diff --git a/.github/actions/setup/action.yml b/.github/actions/setup/action.yml
index 2b8c4c69f322..e78ce2f66d7a 100644
--- a/.github/actions/setup/action.yml
+++ b/.github/actions/setup/action.yml
@@ -16,7 +16,7 @@ runs:
       environment-file: tests/conda/build-environment.yaml
       auto-activate-base: false
       miniforge-version: latest
-      python-version: 3.9
+      python-version: "3.10"
       condarc-file: tests/conda/condarc
   - uses: conda-incubator/setup-miniconda@fc2d68f6413eb2d87b895e92f8584b5b94a10167 # v3.3.0
     if: steps.conda1.outcome == 'failure'
@@ -27,7 +27,7 @@ runs:
       auto-activate-base: false
       miniforge-version: latest
       use-only-tar-bz2: true
-      python-version: 3.9
+      python-version: "3.10"
       condarc-file: tests/conda/condarc
   - name: Conda info
     shell: pwsh
diff --git a/ci/jenkins/generate.py b/ci/jenkins/generate.py
index 8baafa708e13..af9dfbc92f6b 100644
--- a/ci/jenkins/generate.py
+++ b/ci/jenkins/generate.py
@@ -22,7 +22,6 @@
 import textwrap
 from dataclasses import dataclass
 from pathlib import Path
-from typing import List, Optional
 
 import jinja2
 from data import data
@@ -41,7 +40,7 @@ class Change:
 
 @dataclass
 class ChangeData:
-    diff: Optional[str]
+    diff: str | None
     content: str
     destination: Path
     source: Path
@@ -53,7 +52,7 @@ def lines_without_generated_tag(content):
     ]
 
 
-def change_type(lines: List[str]) -> Change:
+def change_type(lines: list[str]) -> Change:
     """
     Return True if 'line' only edits an image tag or if 'line' is not a changed
     line in a diff
diff --git a/ci/scripts/github/github_cc_reviewers.py b/ci/scripts/github/github_cc_reviewers.py
index 1bee762bf153..017878740f28 100755
--- a/ci/scripts/github/github_cc_reviewers.py
+++ b/ci/scripts/github/github_cc_reviewers.py
@@ -23,7 +23,6 @@
 import re
 import sys
 from pathlib import Path
-from typing import List
 from urllib import error
 
 # Hackery to enable importing of utils from ci/scripts/jenkins
@@ -33,7 +32,7 @@
 from git_utils import GitHubRepo, git, parse_remote
 
 
-def find_reviewers(body: str) -> List[str]:
+def find_reviewers(body: str) -> list[str]:
     print(f"Parsing body:\n{body}")
     matches = re.findall(r"(cc( @[-A-Za-z0-9]+)+)", body, flags=re.MULTILINE)
     matches = [full for full, last in matches]
diff --git a/ci/scripts/github/github_commenter.py b/ci/scripts/github/github_commenter.py
index 5b5893b82977..b193dc3b9525 100644
--- a/ci/scripts/github/github_commenter.py
+++ b/ci/scripts/github/github_commenter.py
@@ -21,7 +21,7 @@
 import re
 import sys
 from pathlib import Path
-from typing import Any, Dict, List, Optional, Tuple
+from typing import Any
 
 # Hackery to enable importing of utils from ci/scripts/jenkins
 REPO_ROOT = Path(__file__).resolve().parent.parent.parent.parent
@@ -36,13 +36,13 @@
 class BotCommentBuilder:
     ALLOWLIST_USERS = {"driazati", "gigiblender", "areusch"}
 
-    def __init__(self, github: GitHubRepo, data: Dict[str, Any]):
+    def __init__(self, github: GitHubRepo, data: dict[str, Any]):
         self.github = github
         self.pr_number = data["number"]
         self.comment_data = data["comments"]["nodes"]
         self.author = data["author"]["login"]
 
-    def find_bot_comment(self) -> Optional[Dict[str, Any]]:
+    def find_bot_comment(self) -> dict[str, Any] | None:
         """
         Return the existing bot comment or None if it does not exist
         """
@@ -57,7 +57,7 @@ def find_bot_comment(self) -> Optional[Dict[str, Any]]:
         logging.info("No existing comment found")
         return None
 
-    def find_existing_body(self) -> Dict[str, str]:
+    def find_existing_body(self) -> dict[str, str]:
         """
         Find existing dynamic bullet point items
         """
@@ -84,7 +84,7 @@ def find_existing_body(self) -> Dict[str, str]:
         logging.info(f"Found body items: {items}")
         return items
 
-    def _post_comment(self, body_items: Dict[str, str]):
+    def _post_comment(self, body_items: dict[str, str]):
         comment = BOT_COMMENT_START + "\n\n" + WELCOME_TEXT + "\n\n"
         for key, content in body_items.items():
             line = self.start_key(key) + "\n * " + content.strip() + self.end_key(key)
@@ -118,7 +118,7 @@ def start_key(self, key: str) -> str:
     def end_key(self, key: str) -> str:
         return f"<!--bot-comment-{key}-end-->"
 
-    def post_items(self, items: List[Tuple[str, str]]):
+    def post_items(self, items: list[tuple[str, str]]):
         """
         Update or post bullet points in the PR based on 'items' which is a
         list of (key, text) pairs
diff --git a/ci/scripts/github/github_docs_comment.py b/ci/scripts/github/github_docs_comment.py
index 9422e3c3acfa..287e73653644 100755
--- a/ci/scripts/github/github_docs_comment.py
+++ b/ci/scripts/github/github_docs_comment.py
@@ -16,14 +16,14 @@
 # specific language governing permissions and limitations
 # under the License.
 
-from typing import Any, Dict
+from typing import Any
 
 
 def build_docs_url(base_url_docs, pr_number, build_number):
     return f"{base_url_docs}/PR-{pr_number!s}/{build_number!s}/docs/index.html"
 
 
-def find_target_url(pr_head: Dict[str, Any]):
+def find_target_url(pr_head: dict[str, Any]):
     for status in pr_head["statusCheckRollup"]["contexts"]["nodes"]:
         if status.get("context", "") == "tvm-ci/pr-head":
             return status["targetUrl"]
@@ -39,7 +39,7 @@ def get_pr_and_build_numbers(target_url):
     return {"pr_number": pr_number, "build_number": build_number}
 
 
-def get_doc_url(pr: Dict[str, Any], base_docs_url: str = "https://pr-docs.tlcpack.ai") -> str:
+def get_doc_url(pr: dict[str, Any], base_docs_url: str = "https://pr-docs.tlcpack.ai") -> str:
     pr_head = pr["commits"]["nodes"][0]["commit"]
     target_url = find_target_url(pr_head)
     pr_and_build = get_pr_and_build_numbers(target_url)
diff --git a/ci/scripts/github/github_skipped_tests_comment.py b/ci/scripts/github/github_skipped_tests_comment.py
index 74224195c550..10789f75615b 100755
--- a/ci/scripts/github/github_skipped_tests_comment.py
+++ b/ci/scripts/github/github_skipped_tests_comment.py
@@ -21,7 +21,7 @@
 import os
 import subprocess
 from pathlib import Path
-from typing import Any, Dict, Optional
+from typing import Any
 from xml.etree import ElementTree
 
 
@@ -154,7 +154,7 @@ def build_comment(
     return text
 
 
-def find_target_url(pr_head: Dict[str, Any]):
+def find_target_url(pr_head: dict[str, Any]):
     for status in pr_head["statusCheckRollup"]["contexts"]["nodes"]:
         if status.get("context", "") == "tvm-ci/pr-head":
             return status["targetUrl"]
@@ -163,14 +163,14 @@ def find_target_url(pr_head: Dict[str, Any]):
 
 
 def get_skipped_tests_comment(
-    pr: Dict[str, Any],
+    pr: dict[str, Any],
     github,
     s3_prefix: str = "tvm-jenkins-artifacts-prod",
     jenkins_prefix: str = "ci.tlcpack.ai",
     pr_test_report_dir: str = "pr-reports",
     main_test_report_dir: str = "main-reports",
-    common_commit_sha: Optional[str] = None,
-    common_main_build: Optional[Dict[str, Any]] = None,
+    common_commit_sha: str | None = None,
+    common_main_build: dict[str, Any] | None = None,
     additional_tests_to_check_file: str = "required_tests_to_run.json",
 ) -> str:
     pr_head = pr["commits"]["nodes"][0]["commit"]
diff --git a/ci/scripts/github/github_tag_teams.py b/ci/scripts/github/github_tag_teams.py
index 81956cedd7fe..f2c2102a93b5 100755
--- a/ci/scripts/github/github_tag_teams.py
+++ b/ci/scripts/github/github_tag_teams.py
@@ -24,7 +24,7 @@
 import re
 import sys
 from pathlib import Path
-from typing import Any, Dict, List, Optional, Tuple
+from typing import Any
 
 # Hackery to enable importing of utils from ci/scripts/jenkins
 REPO_ROOT = Path(__file__).resolve().parent.parent.parent.parent
@@ -36,7 +36,7 @@
 GITHUB_NAME_REGEX = r"@[a-zA-Z0-9-]+"
 
 
-def parse_line(line: str) -> Tuple[str, List[str]]:
+def parse_line(line: str) -> tuple[str, list[str]]:
     line = line.lstrip(" -")
     line = line.split()
 
@@ -86,7 +86,7 @@ def fetch_issue(github: GitHubRepo, issue_number: int):
     return r
 
 
-def parse_teams(r: Dict[str, Any], issue_number: int) -> Dict[str, str]:
+def parse_teams(r: dict[str, Any], issue_number: int) -> dict[str, str]:
     """
     Fetch an issue and parse out series of tagged people from the issue body
     and comments
@@ -131,11 +131,11 @@ def add_tag(tag, users):
     return {k.lower(): v for k, v in result.items() if k.strip()}
 
 
-def tags_from_labels(labels: List[Dict[str, Any]]) -> List[str]:
+def tags_from_labels(labels: list[dict[str, Any]]) -> list[str]:
     return [label["name"] for label in labels]
 
 
-def add_ccs_to_body(body: str, to_cc: List[str]) -> str:
+def add_ccs_to_body(body: str, to_cc: list[str]) -> str:
     lines = body.split("\n")
 
     cc_line_idx = None
@@ -174,8 +174,8 @@ def gen_cc_line(users):
 
 
 def determine_users_to_cc(
-    issue: Dict[str, Any], github: GitHubRepo, team_issue: str, issue_data: Optional[Dict[str, Any]]
-) -> List[str]:
+    issue: dict[str, Any], github: GitHubRepo, team_issue: str, issue_data: dict[str, Any] | None
+) -> list[str]:
     if issue_data is None:
         issue_data = fetch_issue(github, issue_number=int(team_issue))
 
@@ -205,7 +205,7 @@ def determine_users_to_cc(
     return to_cc
 
 
-def get_tags(pr_data: Dict[str, Any], github: GitHubRepo, team_issue: int) -> str:
+def get_tags(pr_data: dict[str, Any], github: GitHubRepo, team_issue: int) -> str:
     to_cc = determine_users_to_cc(
         issue=pr_data, github=github, team_issue=team_issue, issue_data=None
     )
diff --git a/ci/scripts/github/github_tvmbot.py b/ci/scripts/github/github_tvmbot.py
index 0d8981b1ba3b..24308cd77cb0 100755
--- a/ci/scripts/github/github_tvmbot.py
+++ b/ci/scripts/github/github_tvmbot.py
@@ -25,8 +25,9 @@
 import sys
 import traceback
 import warnings
+from collections.abc import Callable
 from pathlib import Path
-from typing import Any, Callable, Dict, List, Optional, Union
+from typing import Any
 
 # Hackery to enable importing of utils from ci/scripts/jenkins
 REPO_ROOT = Path(__file__).resolve().parent.parent.parent.parent
@@ -35,9 +36,9 @@
 from cmd_utils import init_log
 from git_utils import GitHubRepo, git, parse_remote, post
 
-Review = Dict[str, Any]
-CIJob = Dict[str, Any]
-Comment = Dict[str, Any]
+Review = dict[str, Any]
+CIJob = dict[str, Any]
+Comment = dict[str, Any]
 CommentChecker = Callable[[Comment], bool]
 
 EXPECTED_JOBS = ["tvm-ci/pr-head"]
@@ -194,7 +195,7 @@ def __init__(
         owner: str,
         repo: str,
         dry_run: bool = False,
-        raw_data: Optional[Dict[str, Any]] = None,
+        raw_data: dict[str, Any] | None = None,
     ):
         self.owner = owner
         self.number = number
@@ -242,7 +243,7 @@ def checker(obj, parent_key):
     def __repr__(self):
         return json.dumps(self.raw, indent=2)
 
-    def react(self, comment: Dict[str, Any], content: str):
+    def react(self, comment: dict[str, Any], content: str):
         """
         React with a thumbs up to a comment
         """
@@ -256,7 +257,7 @@ def react(self, comment: Dict[str, Any], content: str):
     def head_commit(self):
         return self.raw["commits"]["nodes"][0]["commit"]
 
-    def co_authors(self) -> List[str]:
+    def co_authors(self) -> list[str]:
         authors = []
         for commit in self.raw["authorCommits"]["nodes"]:
             # Co-authors always come after the main author according to the
@@ -271,7 +272,7 @@ def co_authors(self) -> List[str]:
     def head_oid(self):
         return self.head_commit()["oid"]
 
-    def ci_jobs(self) -> List[CIJob]:
+    def ci_jobs(self) -> list[CIJob]:
         """
         Get a list of all CI jobs (GitHub Actions and other) in a unified format
         """
@@ -310,14 +311,14 @@ def ci_jobs(self) -> List[CIJob]:
         logging.info(f"Found CI jobs for {self.head_commit()['oid']} {to_json_str(jobs)}")
         return jobs
 
-    def reviews(self) -> List[Review]:
+    def reviews(self) -> list[Review]:
         return self.raw["reviews"]["nodes"]
 
-    def head_commit_reviews(self) -> List[Review]:
+    def head_commit_reviews(self) -> list[Review]:
         """
         Find reviews associated with the head commit
         """
-        commits_to_review_status: Dict[str, List[Review]] = {}
+        commits_to_review_status: dict[str, list[Review]] = {}
 
         for review in self.reviews():
             if not review["authorCanPushToRepository"]:
@@ -347,13 +348,13 @@ def fetch_data(self):
             },
         )["data"]["repository"]["pullRequest"]
 
-    def search_collaborator(self, user: str) -> List[Dict[str, Any]]:
+    def search_collaborator(self, user: str) -> list[dict[str, Any]]:
         """
         Query GitHub for collaborators matching 'user'
         """
         return self.search_users(user, COLLABORATORS_QUERY)["collaborators"]["nodes"]
 
-    def search_users(self, user: str, query: str) -> List[Dict[str, Any]]:
+    def search_users(self, user: str, query: str) -> list[dict[str, Any]]:
         return self.github.graphql(
             query=query,
             variables={
@@ -363,7 +364,7 @@ def search_users(self, user: str, query: str) -> List[Dict[str, Any]]:
             },
         )["data"]["repository"]
 
-    def search_mentionable_users(self, user: str) -> List[Dict[str, Any]]:
+    def search_mentionable_users(self, user: str) -> list[dict[str, Any]]:
         return self.search_users(user, MENTIONABLE_QUERY)["mentionableUsers"]["nodes"]
 
     def comment(self, text: str) -> None:
@@ -449,7 +450,7 @@ def merge(self) -> None:
     def author(self) -> str:
         return self.raw["author"]["login"]
 
-    def find_failed_ci_jobs(self) -> List[CIJob]:
+    def find_failed_ci_jobs(self) -> list[CIJob]:
         # NEUTRAL is GitHub Action's way of saying cancelled
         return [
             job
@@ -457,7 +458,7 @@ def find_failed_ci_jobs(self) -> List[CIJob]:
             if job["status"] not in {"SUCCESS", "SUCCESSFUL", "SKIPPED"}
         ]
 
-    def find_missing_expected_jobs(self) -> List[str]:
+    def find_missing_expected_jobs(self) -> list[str]:
         # Map of job name: has seen in completed jobs
         seen_expected_jobs = {name: False for name in EXPECTED_JOBS}
         logging.info(f"Expected to see jobs: {seen_expected_jobs}")
@@ -485,7 +486,7 @@ def trigger_gha_ci(self, sha: str) -> None:
         )
         logging.info(f"Successful workflow_dispatch: {r}")
 
-    def merge_if_passed_checks(self) -> Optional[Dict[str, Any]]:
+    def merge_if_passed_checks(self) -> dict[str, Any] | None:
         failed_ci_jobs = self.find_failed_ci_jobs()
         all_ci_passed = len(failed_ci_jobs) == 0
         has_one_approval = False
@@ -576,7 +577,7 @@ def rerun_github_actions(self) -> None:
                     else:
                         raise e
 
-    def comment_failure(self, msg: str, exceptions: Union[Exception, List[Exception]]):
+    def comment_failure(self, msg: str, exceptions: Exception | list[Exception]):
         if not isinstance(exceptions, list):
             exceptions = [exceptions]
 
diff --git a/ci/scripts/github/ping_reviewers.py b/ci/scripts/github/ping_reviewers.py
index abe2ac06779f..bd1f434c10ba 100755
--- a/ci/scripts/github/ping_reviewers.py
+++ b/ci/scripts/github/ping_reviewers.py
@@ -24,7 +24,6 @@
 import sys
 import textwrap
 from pathlib import Path
-from typing import List, Optional
 
 # Hackery to enable importing of utils from ci/scripts/jenkins
 REPO_ROOT = Path(__file__).resolve().parent.parent.parent.parent
@@ -35,7 +34,7 @@
 GIT_DATE_FORMAT = "%Y-%m-%dT%H:%M:%SZ"
 
 
-def prs_query(user: str, repo: str, cursor: Optional[str] = None):
+def prs_query(user: str, repo: str, cursor: str | None = None):
     after = ""
     if cursor is not None:
         after = f', before:"{cursor}"'
@@ -86,7 +85,7 @@ def prs_query(user: str, repo: str, cursor: Optional[str] = None):
     """
 
 
-def find_reviewers(body: str) -> List[str]:
+def find_reviewers(body: str) -> list[str]:
     matches = re.findall(r"(cc( @[-A-Za-z0-9]+)+)", body, flags=re.MULTILINE)
     matches = [full for full, last in matches]
 
diff --git a/ci/scripts/github/update_branch.py b/ci/scripts/github/update_branch.py
index 62552989ae74..e5195f7c67f4 100755
--- a/ci/scripts/github/update_branch.py
+++ b/ci/scripts/github/update_branch.py
@@ -22,7 +22,7 @@
 import os
 import sys
 from pathlib import Path
-from typing import Any, Dict, Optional
+from typing import Any
 
 # Hackery to enable importing of utils from ci/scripts/jenkins
 REPO_ROOT = Path(__file__).resolve().parent.parent.parent.parent
@@ -58,7 +58,7 @@
 """
 
 
-def commits_query(user: str, repo: str, cursor: Optional[str] = None):
+def commits_query(user: str, repo: str, cursor: str | None = None):
     """
     Create a GraphQL query to find the last N commits along with their statuses
     and some metadata (paginated after 'cursor')
@@ -101,7 +101,7 @@ def commits_query(user: str, repo: str, cursor: Optional[str] = None):
 ]
 
 
-def commit_passed_ci(commit: Dict[str, Any]) -> bool:
+def commit_passed_ci(commit: dict[str, Any]) -> bool:
     """
     Returns true if all of a commit's statuses are SUCCESS
     """
diff --git a/ci/scripts/jenkins/check_pr.py b/ci/scripts/jenkins/check_pr.py
index 445c636fea9d..683c3fdd6ddb 100755
--- a/ci/scripts/jenkins/check_pr.py
+++ b/ci/scripts/jenkins/check_pr.py
@@ -21,8 +21,9 @@
 import os
 import re
 import textwrap
+from collections.abc import Callable
 from dataclasses import dataclass
-from typing import Any, Callable, List
+from typing import Any
 
 from cmd_utils import init_log, tags_from_title
 from git_utils import GitHubRepo, git, parse_remote
@@ -82,7 +83,7 @@ def trailing_period(s: str):
 ]
 
 
-def run_checks(checks: List[Check], s: str, name: str) -> bool:
+def run_checks(checks: list[Check], s: str, name: str) -> bool:
     print(f"Running checks for {name}")
     print(textwrap.indent(s, prefix="    "))
     passed = True
diff --git a/ci/scripts/jenkins/cmd_utils.py b/ci/scripts/jenkins/cmd_utils.py
index 4b33ce44d261..e1cd8e9b20e9 100644
--- a/ci/scripts/jenkins/cmd_utils.py
+++ b/ci/scripts/jenkins/cmd_utils.py
@@ -21,7 +21,6 @@
 import subprocess
 import sys
 from pathlib import Path
-from typing import List
 
 REPO_ROOT = Path(__file__).resolve().parent.parent.parent.parent
 assert (REPO_ROOT / "Jenkinsfile").exists
@@ -102,7 +101,7 @@ def _default_popen_flags(self):
         }
 
 
-def tags_from_title(title: str) -> List[str]:
+def tags_from_title(title: str) -> list[str]:
     tags = re.findall(r"\[(.*?)\]", title)
     tags = [t.strip() for t in tags]
     return tags
diff --git a/ci/scripts/jenkins/determine_docker_images.py b/ci/scripts/jenkins/determine_docker_images.py
index a0d5b633e536..fed3ab79cc0e 100755
--- a/ci/scripts/jenkins/determine_docker_images.py
+++ b/ci/scripts/jenkins/determine_docker_images.py
@@ -23,7 +23,7 @@
 import logging
 import urllib.error
 from pathlib import Path
-from typing import Any, Dict
+from typing import Any
 
 from cmd_utils import REPO_ROOT, init_log
 from http_utils import get
@@ -35,7 +35,7 @@
 TVM_CI_ECR = "477529581014.dkr.ecr.us-west-2.amazonaws.com"
 
 
-def docker_api(url: str, use_pagination: bool = False) -> Dict[str, Any]:
+def docker_api(url: str, use_pagination: bool = False) -> dict[str, Any]:
     """
     Run a paginated fetch from the public Docker Hub API
     """
diff --git a/ci/scripts/jenkins/git_skip_ci_globs.py b/ci/scripts/jenkins/git_skip_ci_globs.py
index dca941e7c843..651bbeac7c55 100755
--- a/ci/scripts/jenkins/git_skip_ci_globs.py
+++ b/ci/scripts/jenkins/git_skip_ci_globs.py
@@ -19,7 +19,6 @@
 
 import argparse
 import fnmatch
-from typing import Optional
 
 from git_utils import git
 
@@ -36,7 +35,7 @@
 ]
 
 
-def match_any(f: str) -> Optional[str]:
+def match_any(f: str) -> str | None:
     for glob in globs:
         if fnmatch.fnmatch(f, glob):
             return glob
diff --git a/ci/scripts/jenkins/git_utils.py b/ci/scripts/jenkins/git_utils.py
index 05301a046ac0..b16e3f7aa320 100644
--- a/ci/scripts/jenkins/git_utils.py
+++ b/ci/scripts/jenkins/git_utils.py
@@ -23,7 +23,7 @@
 import os
 import re
 import subprocess
-from typing import Any, Dict, List, Optional, Tuple
+from typing import Any
 from urllib import error, request
 
 DRY_RUN = object()
@@ -35,7 +35,7 @@ def compress_query(query: str) -> str:
     return query
 
 
-def post(url: str, body: Optional[Any] = None, auth: Optional[Tuple[str, str]] = None):
+def post(url: str, body: Any | None = None, auth: tuple[str, str] | None = None):
     logging.info(f"Requesting POST to {url} with {body}")
     headers = {}
     req = request.Request(url, headers=headers, method="POST")
@@ -80,7 +80,7 @@ def headers(self):
     def dry_run(self) -> bool:
         return self.token == DRY_RUN
 
-    def graphql(self, query: str, variables: Optional[Dict[str, str]] = None) -> Dict[str, Any]:
+    def graphql(self, query: str, variables: dict[str, str] | None = None) -> dict[str, Any]:
         query = compress_query(query)
         if variables is None:
             variables = {}
@@ -106,7 +106,7 @@ def testing_response(self, method: str, url: str) -> Any:
         logging.info(f"Unknown URL in dry run: {key}")
         return {}
 
-    def _request(self, full_url: str, body: Dict[str, Any], method: str) -> Dict[str, Any]:
+    def _request(self, full_url: str, body: dict[str, Any], method: str) -> dict[str, Any]:
         if self.dry_run():
             logging.info(f"Dry run, would have requested a {method} to {full_url} with {body}")
             return self.testing_response(method, full_url)
@@ -134,16 +134,16 @@ def _request(self, full_url: str, body: Dict[str, Any], method: str) -> Dict[str
 
         return response
 
-    def put(self, url: str, data: Dict[str, Any]) -> Dict[str, Any]:
+    def put(self, url: str, data: dict[str, Any]) -> dict[str, Any]:
         return self._request(self.base + url, data, method="PUT")
 
-    def patch(self, url: str, data: Dict[str, Any]) -> Dict[str, Any]:
+    def patch(self, url: str, data: dict[str, Any]) -> dict[str, Any]:
         return self._request(self.base + url, data, method="PATCH")
 
-    def post(self, url: str, data: Dict[str, Any]) -> Dict[str, Any]:
+    def post(self, url: str, data: dict[str, Any]) -> dict[str, Any]:
         return self._request(self.base + url, data, method="POST")
 
-    def get(self, url: str) -> Dict[str, Any]:
+    def get(self, url: str) -> dict[str, Any]:
         if self.dry_run():
             logging.info(f"Dry run, would have requested a GET to {url}")
             return self.testing_response("GET", url)
@@ -154,7 +154,7 @@ def get(self, url: str) -> Dict[str, Any]:
             response = json.loads(response.read())
         return response
 
-    def delete(self, url: str) -> Dict[str, Any]:
+    def delete(self, url: str) -> dict[str, Any]:
         if self.dry_run():
             logging.info(f"Dry run, would have requested a DELETE to {url}")
             return self.testing_response("DELETE", url)
@@ -166,7 +166,7 @@ def delete(self, url: str) -> Dict[str, Any]:
         return response
 
 
-def parse_remote(remote: str) -> Tuple[str, str]:
+def parse_remote(remote: str) -> tuple[str, str]:
     """
     Get a GitHub (user, repo) pair out of a git remote
     """
@@ -197,7 +197,7 @@ def git(command, **kwargs):
     return proc.stdout.strip()
 
 
-def find_ccs(body: str) -> List[str]:
+def find_ccs(body: str) -> list[str]:
     matches = re.findall(r"(cc( @[-A-Za-z0-9]+)+)", body, flags=re.MULTILINE)
     matches = [full for full, last in matches]
 
diff --git a/ci/scripts/jenkins/http_utils.py b/ci/scripts/jenkins/http_utils.py
index e006c48d477c..05e1b71f24d6 100644
--- a/ci/scripts/jenkins/http_utils.py
+++ b/ci/scripts/jenkins/http_utils.py
@@ -18,11 +18,11 @@
 
 import json
 import logging
-from typing import Any, Dict, Optional
+from typing import Any
 from urllib import request
 
 
-def get(url: str, headers: Optional[Dict[str, str]] = None) -> Dict[str, Any]:
+def get(url: str, headers: dict[str, str] | None = None) -> dict[str, Any]:
     logging.info(f"Requesting GET to {url}")
     if headers is None:
         headers = {}
diff --git a/ci/scripts/jenkins/open_docker_update_pr.py b/ci/scripts/jenkins/open_docker_update_pr.py
index 6809d2c4d916..a948750082e8 100755
--- a/ci/scripts/jenkins/open_docker_update_pr.py
+++ b/ci/scripts/jenkins/open_docker_update_pr.py
@@ -23,7 +23,8 @@
 import os
 import re
 import shlex
-from typing import Any, Callable, Dict, List, Optional
+from collections.abc import Callable
+from typing import Any
 from urllib import error
 
 from cmd_utils import REPO_ROOT, Sh, init_log
@@ -37,10 +38,10 @@
 BRANCH = "nightly-docker-update"
 
 
-def _testing_docker_api(data: Dict[str, Any]) -> Callable[[str], Dict[str, Any]]:
+def _testing_docker_api(data: dict[str, Any]) -> Callable[[str], dict[str, Any]]:
     """Returns a function that can be used in place of docker_api"""
 
-    def mock(url: str) -> Dict[str, Any]:
+    def mock(url: str) -> dict[str, Any]:
         if url in data:
             return data[url]
         else:
@@ -54,11 +55,11 @@ def parse_docker_date(d: str) -> datetime.datetime:
     return datetime.datetime.strptime(d, "%Y-%m-%dT%H:%M:%S.%fZ")
 
 
-def check_tag(tag: Dict[str, Any]) -> bool:
+def check_tag(tag: dict[str, Any]) -> bool:
     return re.match(r"^[0-9]+-[0-9]+-[a-z0-9]+$", tag["name"]) is not None
 
 
-def latest_tag(user: str, repo: str) -> List[Dict[str, Any]]:
+def latest_tag(user: str, repo: str) -> list[dict[str, Any]]:
     """
     Queries Docker Hub and finds the most recent tag for the specified image/repo pair
     """
@@ -73,7 +74,7 @@ def latest_tag(user: str, repo: str) -> List[Dict[str, Any]]:
     return results[-1]
 
 
-def latest_tlcpackstaging_image(source: str) -> Optional[str]:
+def latest_tlcpackstaging_image(source: str) -> str | None:
     """
     Finds the latest full tag to use in the Jenkinsfile or returns None if no
     update is needed
diff --git a/ci/scripts/jenkins/pytest_wrapper.py b/ci/scripts/jenkins/pytest_wrapper.py
index fc00fc8159b8..a66781029350 100755
--- a/ci/scripts/jenkins/pytest_wrapper.py
+++ b/ci/scripts/jenkins/pytest_wrapper.py
@@ -22,7 +22,6 @@
 import textwrap
 import urllib.parse
 from pathlib import Path
-from typing import List, Optional
 
 import junitparser
 from cmd_utils import init_log
@@ -42,7 +41,7 @@ def classname_to_file(classname: str) -> str:
     return classname.replace(".", "/") + ".py"
 
 
-def failed_test_ids() -> List[str]:
+def failed_test_ids() -> list[str]:
     FAILURE_TYPES = (junitparser.Failure, junitparser.Error)
     junit_dir = REPO_ROOT / "build" / "pytest-results"
     failed_node_ids = []
@@ -62,7 +61,7 @@ def failed_test_ids() -> List[str]:
     return list(set(failed_node_ids))
 
 
-def repro_command(build_type: str, failed_node_ids: List[str]) -> Optional[str]:
+def repro_command(build_type: str, failed_node_ids: list[str]) -> str | None:
     """
     Parse available JUnit XML files and output a command that users can run to
     reproduce CI failures locally
@@ -72,7 +71,7 @@ def repro_command(build_type: str, failed_node_ids: List[str]) -> Optional[str]:
     return f"python3 tests/scripts/ci.py {build_type} {test_args_str}"
 
 
-def make_issue_url(failed_node_ids: List[str]) -> str:
+def make_issue_url(failed_node_ids: list[str]) -> str:
     names = [f"`{node_id}`" for node_id in failed_node_ids]
     run_url = os.getenv("RUN_DISPLAY_URL", "<insert run URL>")
     test_bullets = [f"  - `{node_id}`" for node_id in failed_node_ids]
@@ -92,7 +91,7 @@ def make_issue_url(failed_node_ids: List[str]) -> str:
     return "https://github.com/apache/tvm/issues/new?" + urllib.parse.urlencode(params)
 
 
-def show_failure_help(failed_suites: List[str]) -> None:
+def show_failure_help(failed_suites: list[str]) -> None:
     failed_node_ids = failed_test_ids()
 
     if len(failed_node_ids) == 0:
diff --git a/ci/scripts/jenkins/s3.py b/ci/scripts/jenkins/s3.py
index 414348f9e562..dd44490b1e12 100755
--- a/ci/scripts/jenkins/s3.py
+++ b/ci/scripts/jenkins/s3.py
@@ -21,7 +21,6 @@
 import re
 from enum import Enum
 from pathlib import Path
-from typing import List
 
 from cmd_utils import REPO_ROOT, Sh, init_log
 
@@ -40,7 +39,7 @@ def show_md5(item: str) -> None:
         sh.run(f"md5sum {item}")
 
 
-def parse_output_files(stdout: str) -> List[str]:
+def parse_output_files(stdout: str) -> list[str]:
     """
     Grab the list of downloaded files from the output of 'aws s3 cp'. Lines look
     like:
@@ -59,7 +58,7 @@ def parse_output_files(stdout: str) -> List[str]:
     return files
 
 
-def chmod(files: List[str]) -> None:
+def chmod(files: list[str]) -> None:
     """
     S3 has no concept of file permissions so add them back in here to every file
     """
@@ -70,7 +69,7 @@ def chmod(files: List[str]) -> None:
         SH.run(f"chmod +x {' '.join(to_chmod)}")
 
 
-def s3(source: str, destination: str, recursive: bool) -> List[str]:
+def s3(source: str, destination: str, recursive: bool) -> list[str]:
     """
     Send or download the source to the destination in S3
     """
diff --git a/ci/scripts/jenkins/should_rebuild_docker.py b/ci/scripts/jenkins/should_rebuild_docker.py
index 238bea04ef20..0c608dc43a80 100755
--- a/ci/scripts/jenkins/should_rebuild_docker.py
+++ b/ci/scripts/jenkins/should_rebuild_docker.py
@@ -21,7 +21,7 @@
 import json
 import logging
 import subprocess
-from typing import Any, Dict, List
+from typing import Any
 
 from cmd_utils import Sh, init_log
 from http_utils import get
@@ -31,7 +31,7 @@
 TEST_DATA = None
 
 
-def docker_api(url: str) -> Dict[str, Any]:
+def docker_api(url: str) -> dict[str, Any]:
     """
     Run a paginated fetch from the public Docker Hub API
     """
@@ -81,7 +81,7 @@ def does_commit_exist(hash: str) -> bool:
     raise RuntimeError(f"Unexpected failure when running: {cmd}")
 
 
-def find_hash_for_tag(tag: Dict[str, Any]) -> str:
+def find_hash_for_tag(tag: dict[str, Any]) -> str:
     """
     Split the hash off of a name like <date>-<time>-<hash>
     """
@@ -93,7 +93,7 @@ def find_hash_for_tag(tag: Dict[str, Any]) -> str:
     return shorthash
 
 
-def find_commit_in_repo(tags: List[Dict[str, Any]]):
+def find_commit_in_repo(tags: list[dict[str, Any]]):
     """
     Look through all the docker tags, find the most recent one which references
     a commit that is present in the repo
diff --git a/ci/scripts/jenkins/should_run_slow_tests.py b/ci/scripts/jenkins/should_run_slow_tests.py
index 16b3a080820e..31126d7daf4e 100755
--- a/ci/scripts/jenkins/should_run_slow_tests.py
+++ b/ci/scripts/jenkins/should_run_slow_tests.py
@@ -19,7 +19,6 @@
 import argparse
 import os
 import textwrap
-from typing import List, Optional, Tuple
 
 from git_utils import GitHubRepo, git, parse_remote
 
@@ -33,7 +32,7 @@
 ]
 
 
-def check_match(s: str, searches: List[str]) -> Tuple[bool, Optional[str]]:
+def check_match(s: str, searches: list[str]) -> tuple[bool, str | None]:
     for search in searches:
         if search in s:
             return True, search
diff --git a/docs/conf.py b/docs/conf.py
index 57d08025c4fc..80b82f986f01 100644
--- a/docs/conf.py
+++ b/docs/conf.py
@@ -38,7 +38,6 @@
 from importlib import import_module
 from pathlib import Path
 from textwrap import dedent, indent
-from typing import List
 from unittest.mock import patch
 
 # If extensions (or modules to document with autodoc) are in another directory,
@@ -613,7 +612,7 @@ def update_alias_docstring(name, obj, lines):
 }
 
 
-def distinguish_class_name(name: str, lines: List[str]):
+def distinguish_class_name(name: str, lines: list[str]):
     """Distinguish the docstring of type annotations.
 
     In the whole TVM, there are many classes with the same name but in different modules,
diff --git a/docs/download_3rdparty_embeds.py b/docs/download_3rdparty_embeds.py
index 35dcd84486ca..b5660cb4b7c3 100644
--- a/docs/download_3rdparty_embeds.py
+++ b/docs/download_3rdparty_embeds.py
@@ -20,8 +20,8 @@
 import hashlib
 import os
 import re
+from collections.abc import Callable
 from html.parser import HTMLParser
-from typing import Callable, Dict, List, Optional, Set, Tuple, Union
 from urllib.parse import urlparse
 
 import requests
@@ -43,7 +43,7 @@ class ExternalURLParser(HTMLParser):
 
     def __init__(self):
         super().__init__()
-        self.external_urls: List[str] = []
+        self.external_urls: list[str] = []
         self.base_domain = urlparse(BASE_URL).netloc
         # Tags and their attributes that might contain external resources
         self.tags_to_check = {
@@ -57,7 +57,7 @@ def __init__(self):
             "embed": "src",
         }
 
-    def handle_starttag(self, tag: str, attrs: List[tuple[str, Union[str, None]]]) -> None:
+    def handle_starttag(self, tag: str, attrs: list[tuple[str, str | None]]) -> None:
         """Handle HTML start tags to find external URLs."""
         if tag not in self.tags_to_check:
             return
@@ -73,7 +73,7 @@ def handle_starttag(self, tag: str, attrs: List[tuple[str, Union[str, None]]]) -
                     self.external_urls.append(value)
 
 
-def detect_html_external_urls(html_content: str) -> List[str]:
+def detect_html_external_urls(html_content: str) -> list[str]:
     """
     Detect third-party embedded resources in HTML content.
 
@@ -92,7 +92,7 @@ def detect_html_external_urls(html_content: str) -> List[str]:
     return parser.external_urls
 
 
-def detect_css_external_urls(css_content: str) -> List[str]:
+def detect_css_external_urls(css_content: str) -> list[str]:
     """
     Detect external URLs in CSS content.
 
@@ -106,7 +106,7 @@ def detect_css_external_urls(css_content: str) -> List[str]:
     List[str]
         List of external URLs found in the CSS content
     """
-    external_urls: List[str] = []
+    external_urls: list[str] = []
     # Regex to find URLs in CSS
     url_pattern = re.compile(r'url\(["\']?(.*?)["\']?\)')
     matches = url_pattern.findall(css_content)
@@ -116,7 +116,7 @@ def detect_css_external_urls(css_content: str) -> List[str]:
     return external_urls
 
 
-def all_files_in_dir(path: str) -> List[str]:
+def all_files_in_dir(path: str) -> list[str]:
     """
     Get a list of all files in a directory and its subdirectories.
 
@@ -133,7 +133,7 @@ def all_files_in_dir(path: str) -> List[str]:
     return [os.path.join(root, file) for root, _, files in os.walk(path) for file in files]
 
 
-def detect_urls(files: List[str], verbose: bool = False) -> List[str]:
+def detect_urls(files: list[str], verbose: bool = False) -> list[str]:
     """
     Detect external URLs in the given HTML and CSS files.
 
@@ -150,9 +150,9 @@ def detect_urls(files: List[str], verbose: bool = False) -> List[str]:
         List of external URLs found in the files
     """
 
-    external_urls: Set[str] = set()
+    external_urls: set[str] = set()
     for file in files:
-        f_detect: Union[Callable[[str, str], List[str]], None] = None
+        f_detect: Callable[[str, str], list[str]] | None = None
         if file.endswith(".html"):
             f_detect = detect_html_external_urls
         elif file.endswith(".css"):
@@ -183,8 +183,8 @@ def detect_urls(files: List[str], verbose: bool = False) -> List[str]:
 
 
 def download_external_urls(
-    external_urls: List[str], verbose: bool = False
-) -> Tuple[Dict[str, str], List[str]]:
+    external_urls: list[str], verbose: bool = False
+) -> tuple[dict[str, str], list[str]]:
     """
     Download external URLs and save them to docs/_static/downloads.
 
@@ -204,9 +204,9 @@ def download_external_urls(
     """
     download_dir = os.path.join(HTML_DIR, "_static/downloads")
     os.makedirs(download_dir, exist_ok=True)
-    used_file_names: Set[str] = set()
-    downloaded_files: List[str] = []
-    remap_urls: Dict[str, str] = {}
+    used_file_names: set[str] = set()
+    downloaded_files: list[str] = []
+    remap_urls: dict[str, str] = {}
     for url in external_urls:
         query = urlparse(url).query
         if url.startswith("https://fonts.googleapis.com/css2"):
@@ -242,7 +242,7 @@ def download_external_urls(
     return remap_urls, downloaded_files
 
 
-def replace_urls_in_files(remap_urls: Dict[str, str], verbose: bool = False):
+def replace_urls_in_files(remap_urls: dict[str, str], verbose: bool = False):
     """
     Replace external URLs with their downloaded versions in HTML/CSS files.
 
@@ -282,7 +282,7 @@ def replace_urls_in_files(remap_urls: Dict[str, str], verbose: bool = False):
                     print(f"Updated {file_path}")
 
 
-def download_and_replace_urls(files: Optional[List[str]] = None, verbose: bool = False):
+def download_and_replace_urls(files: list[str] | None = None, verbose: bool = False):
     """
     Download external URLs found in files and replace them with local copies.
     Recursively processes any new external URLs found in downloaded content.
diff --git a/docs/how_to/tutorials/optimize_llm.py b/docs/how_to/tutorials/optimize_llm.py
index d8d3842a7941..f698a37f3c09 100644
--- a/docs/how_to/tutorials/optimize_llm.py
+++ b/docs/how_to/tutorials/optimize_llm.py
@@ -60,7 +60,6 @@
 import os
 from pathlib import Path
 from pprint import pprint
-from typing import List, Optional
 
 import tvm
 from tvm import relax, te, tir
@@ -251,7 +250,7 @@ def __init__(self, config: LlamaConfig):
         self.rope_theta = config.rope_theta
         self.dtype = "float32"
 
-    def to(self, dtype: Optional[str] = None):
+    def to(self, dtype: str | None = None):
         super().to(dtype=dtype)
         if dtype is not None:
             self.dtype = dtype
@@ -378,7 +377,7 @@ def get_default_spec(self):
 
 @register_pipeline("opt_llm")
 def _pipeline(  # pylint: disable=too-many-arguments
-    ext_mods: Optional[List[nn.ExternModule]] = None,
+    ext_mods: list[nn.ExternModule] | None = None,
 ):
     ext_mods = ext_mods or []
 
diff --git a/pyproject.toml b/pyproject.toml
index 633e69f5a7ab..e4241426d145 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -26,7 +26,7 @@ version = "0.24.dev0"
 description = "Apache TVM: An End-to-End Deep Learning Compiler Stack"
 readme = "README.md"
 license = { text = "Apache-2.0" }
-requires-python = ">=3.9"
+requires-python = ">=3.10"
 authors = [{ name = "Apache TVM Community", email = "dev@tvm.apache.org" }]
 keywords = ["machine learning", "compiler", "deep learning", "inference"]
 classifiers = [
@@ -197,7 +197,7 @@ include = [
 ]
 line-length = 100
 indent-width = 4
-target-version = "py39"
+target-version = "py310"
 exclude = [
   "3rdparty",
   "build",
@@ -217,11 +217,10 @@ select = [
   "RUF", # ruff, https://docs.astral.sh/ruff/rules/#ruff-specific-rules-ruf
 ]
 ignore = [
-  # pyupgrade (needs `from __future__ import annotations`)
-  "UP006", # non-pep585-annotation
-  "UP007", # non-pep604-annotation-union
-  "UP035", # deprecated-import
-  "UP045", # non-pep604-annotation-optional
+  # pyupgrade (style preference or low-value)
+  "UP028", # yield-in-for-loop — sometimes clearer to keep
+  "UP030", # format-literals — minor
+  "UP031", # printf-string-formatting — style preference
   # ruff (preview rules, not yet stable)
   "RUF043", # pytest-raises-ambiguous-pattern
   "RUF059", # unused-unpacked-variable
@@ -231,6 +230,7 @@ unfixable = []
 
 [tool.ruff.lint.per-file-ignores]
 "__init__.py" = ["E402", "F401", "F403", "F405"]
+"python/tvm/contrib/msc/**" = ["UP"]
 
 [tool.ruff.lint.isort]
 known-first-party = ["tvm"]
diff --git a/python/tvm/arith/analyzer.py b/python/tvm/arith/analyzer.py
index a3abf5670a9f..f7d0ab82aad5 100644
--- a/python/tvm/arith/analyzer.py
+++ b/python/tvm/arith/analyzer.py
@@ -18,7 +18,6 @@
 """Arithmetic data structure and utility"""
 
 import enum
-from typing import Union
 
 import tvm_ffi
 
@@ -267,7 +266,7 @@ def can_prove(
         """
         return self._can_prove(expr, strength)
 
-    def bind(self, var: tir.Var, expr: Union[tir.PrimExpr, ir.Range]) -> None:
+    def bind(self, var: tir.Var, expr: tir.PrimExpr | ir.Range) -> None:
         """Bind a variable to the expression.
 
         Parameters
@@ -355,7 +354,7 @@ def enabled_extensions(self) -> Extension:
         return Extension(value)
 
     @enabled_extensions.setter
-    def enabled_extensions(self, flags: Union[int, Extension]):
+    def enabled_extensions(self, flags: int | Extension):
         """Enable extensions for the analyzer
 
         Parameters
diff --git a/python/tvm/arith/pattern.py b/python/tvm/arith/pattern.py
index 0158f712c137..229283d54b93 100644
--- a/python/tvm/arith/pattern.py
+++ b/python/tvm/arith/pattern.py
@@ -16,8 +16,6 @@
 # under the License.
 """Detect common patterns."""
 
-from typing import Dict
-
 from tvm.tir import PrimExpr
 
 from . import _ffi_api
@@ -65,7 +63,7 @@ def detect_clip_bound(expr, var_list):
     return _ffi_api.DetectClipBound(expr, var_list)
 
 
-def detect_common_subexpr(expr: PrimExpr, threshold: int) -> Dict[PrimExpr, int]:
+def detect_common_subexpr(expr: PrimExpr, threshold: int) -> dict[PrimExpr, int]:
     """Detect common sub expression which shows up more than a threshold times
 
     Parameters
diff --git a/python/tvm/contrib/cc.py b/python/tvm/contrib/cc.py
index 1765fa93f5a6..d63f67a1ceae 100644
--- a/python/tvm/contrib/cc.py
+++ b/python/tvm/contrib/cc.py
@@ -22,7 +22,6 @@
 
 # pylint: disable=invalid-name
 import sys
-from typing import Dict
 
 from ..base import py_str
 from . import tar as _tar
@@ -180,7 +179,7 @@ def create_executable(output, objects, options=None, cc=None, cwd=None, ccache_e
         raise ValueError("Unsupported platform")
 
 
-def get_global_symbol_section_map(path, *, nm=None) -> Dict[str, str]:
+def get_global_symbol_section_map(path, *, nm=None) -> dict[str, str]:
     """Get global symbols from a library via nm -g
 
     Parameters
diff --git a/python/tvm/contrib/cutlass/build.py b/python/tvm/contrib/cutlass/build.py
index c1ad062b9097..f7e02a09852f 100644
--- a/python/tvm/contrib/cutlass/build.py
+++ b/python/tvm/contrib/cutlass/build.py
@@ -23,8 +23,8 @@
 import multiprocessing
 import operator
 import os
+from collections.abc import Sequence
 from functools import reduce
-from typing import Optional, Sequence
 
 from tvm_ffi import register_global_func
 
@@ -363,7 +363,7 @@ def tune_cutlass_kernels(
     return mod, num_cutlass_partition
 
 
-def _get_call_node(expr: relax.Expr, op_name: str) -> Optional[relax.Call]:
+def _get_call_node(expr: relax.Expr, op_name: str) -> relax.Call | None:
     node = None
 
     def fvisit(e):
@@ -417,7 +417,7 @@ def is_shape_valid_for_cutlass_matmul(
     as well as ND x 2D and 2D x ND. For example, it cannot handle matmul with shape
     (2, 1, 4, 8) x (2, 3, 8, 16), because the batch stride of lhs is not constant.
     """
-    if not isinstance(lhs_shape[-1], (tvm.tir.expr.IntImm, int)):
+    if not isinstance(lhs_shape[-1], tvm.tir.expr.IntImm | int):
         # Reduction axis must be constant
         return False
 
diff --git a/python/tvm/contrib/cutlass/library.py b/python/tvm/contrib/cutlass/library.py
index cacff3232f17..62fb952a18eb 100644
--- a/python/tvm/contrib/cutlass/library.py
+++ b/python/tvm/contrib/cutlass/library.py
@@ -143,9 +143,9 @@ def substitute_template(template, values):
     while changed:
         changed = False
         for key, value in values.items():
-            if isinstance(value, (int, IntImm)):
+            if isinstance(value, int | IntImm):
                 value = str(int(value))
-            if isinstance(value, (float, FloatImm)):
+            if isinstance(value, float | FloatImm):
                 value = str(float(value))
             elif isinstance(value, bool):
                 value = str(value).lower()
diff --git a/python/tvm/contrib/download.py b/python/tvm/contrib/download.py
index b93aafae4d6e..65d943b81519 100644
--- a/python/tvm/contrib/download.py
+++ b/python/tvm/contrib/download.py
@@ -166,7 +166,7 @@ def download_testdata(url, relpath, module=None, overwrite=False):
         module_path = ""
     elif isinstance(module, str):
         module_path = module
-    elif isinstance(module, (list, tuple)):
+    elif isinstance(module, list | tuple):
         module_path = Path(*module)
     else:
         raise ValueError("Unsupported module: " + module)
diff --git a/python/tvm/contrib/hexagon/build.py b/python/tvm/contrib/hexagon/build.py
index 27d0183f3609..63203777983f 100644
--- a/python/tvm/contrib/hexagon/build.py
+++ b/python/tvm/contrib/hexagon/build.py
@@ -33,7 +33,6 @@
 import subprocess
 import sys
 import tempfile
-from typing import Optional, Union
 
 from tvm_ffi import libinfo
 
@@ -193,8 +192,8 @@ class HexagonLauncherRPC(metaclass=abc.ABCMeta):
     def __init__(
         self,
         rpc_info: dict,
-        workspace: Optional[Union[str, pathlib.Path]] = None,
-        serial_number: Optional[str] = None,
+        workspace: str | pathlib.Path | None = None,
+        serial_number: str | None = None,
     ):
         self._rpc_info = {
             "rpc_tracker_host": "0.0.0.0",
@@ -222,9 +221,7 @@ def cleanup_directory(self):
         ...
 
     @abc.abstractmethod
-    def _copy_to_remote(
-        self, local_path: Union[str, pathlib.Path], remote_path: Union[str, pathlib.Path]
-    ):
+    def _copy_to_remote(self, local_path: str | pathlib.Path, remote_path: str | pathlib.Path):
         """Copy a local file to a remote location.
 
         Parameters
@@ -237,7 +234,7 @@ def _copy_to_remote(
         ...
 
     @abc.abstractmethod
-    def _create_remote_directory(self, remote_path: Union[str, pathlib.Path]) -> pathlib.Path:
+    def _create_remote_directory(self, remote_path: str | pathlib.Path) -> pathlib.Path:
         """Create a directory in the remote location.
 
         Parameters
@@ -252,7 +249,7 @@ def _create_remote_directory(self, remote_path: Union[str, pathlib.Path]) -> pat
         """
         ...
 
-    def _create_workspace(self, workspace: Union[str, pathlib.Path]) -> pathlib.Path:
+    def _create_workspace(self, workspace: str | pathlib.Path) -> pathlib.Path:
         """Create a working directory for the server.
 
         Parameters
@@ -334,7 +331,7 @@ def __init__(
         self,
         serial_number: str,
         rpc_info: dict,
-        workspace: Optional[Union[str, pathlib.Path]] = None,
+        workspace: str | pathlib.Path | None = None,
         hexagon_debug: bool = False,
         clear_logcat: bool = False,
         sysmon_profile: bool = False,
@@ -381,14 +378,12 @@ def __init__(
 
         super().__init__(rpc_info, workspace, self._serial_number)
 
-    def _copy_to_remote(
-        self, local_path: Union[str, pathlib.Path], remote_path: Union[str, pathlib.Path]
-    ):
+    def _copy_to_remote(self, local_path: str | pathlib.Path, remote_path: str | pathlib.Path):
         """Abstract method implementation. See description in HexagonLauncherRPC."""
 
         _check_call_verbose(self._adb_device_sub_cmd + ["push", str(local_path), str(remote_path)])
 
-    def _create_remote_directory(self, remote_path: Union[str, pathlib.Path]) -> pathlib.Path:
+    def _create_remote_directory(self, remote_path: str | pathlib.Path) -> pathlib.Path:
         """Abstract method implementation. See description in HexagonLauncherRPC."""
         _check_call_verbose(self._adb_device_sub_cmd + ["shell", "mkdir", "-p", str(remote_path)])
         return pathlib.Path(remote_path)
@@ -659,7 +654,7 @@ class HexagonLauncherSimulator(HexagonLauncherRPC):
 
     SIMULATOR_HEXAGON_RPC_FILES = ["tvm_rpc_x86", "libhexagon_rpc_sim.so"]
 
-    def __init__(self, rpc_info: dict, workspace: Optional[Union[str, pathlib.Path]] = None):
+    def __init__(self, rpc_info: dict, workspace: str | pathlib.Path | None = None):
         """Configure a new HexagonLauncherSimulator
 
         Parameters are same as for HexagonLauncherRPC.
@@ -672,18 +667,16 @@ def __init__(self, rpc_info: dict, workspace: Optional[Union[str, pathlib.Path]]
 
         super().__init__(rpc_info, workspace, self._serial_number)
 
-    def _copy_to_remote(
-        self, local_path: Union[str, pathlib.Path], remote_path: Union[str, pathlib.Path]
-    ):
+    def _copy_to_remote(self, local_path: str | pathlib.Path, remote_path: str | pathlib.Path):
         """Abstract method implementation. See description in HexagonLauncherRPC."""
         _check_call_verbose(["cp", str(local_path), str(remote_path)])
 
-    def _create_remote_directory(self, remote_path: Union[str, pathlib.Path]) -> pathlib.Path:
+    def _create_remote_directory(self, remote_path: str | pathlib.Path) -> pathlib.Path:
         """Abstract method implementation. See description in HexagonLauncherRPC."""
         _check_call_verbose(["mkdir", "-p", str(remote_path)])
         return pathlib.Path(os.path.abspath(remote_path))
 
-    def _copy_libcxx(self, dest_dir: Union[str, pathlib.Path]):
+    def _copy_libcxx(self, dest_dir: str | pathlib.Path):
         """Copy libc++ libraries to the remote workspace."""
         # Copy the v68 versions, since we don't have target information.
         # The v68 ones should work everywhere on v68+.
@@ -830,7 +823,7 @@ def farf_config_from_python_log_level(level) -> str:
 def HexagonLauncher(
     serial_number: str,
     rpc_info: dict,
-    workspace: Optional[Union[str, pathlib.Path]] = None,
+    workspace: str | pathlib.Path | None = None,
     hexagon_debug: bool = False,
     clear_logcat: bool = False,
     sysmon_profile: bool = False,
diff --git a/python/tvm/contrib/hexagon/meta_schedule.py b/python/tvm/contrib/hexagon/meta_schedule.py
index a597848e5212..2a52fc54603e 100644
--- a/python/tvm/contrib/hexagon/meta_schedule.py
+++ b/python/tvm/contrib/hexagon/meta_schedule.py
@@ -18,7 +18,7 @@
 
 import os
 import tempfile
-from typing import Callable, Dict, List, Optional
+from collections.abc import Callable
 
 import tvm
 from tvm.contrib.popen_pool import PopenPoolExecutor
@@ -52,11 +52,11 @@ class HexagonRPCRunner(PyRunner):
     def __init__(
         self,
         hexagon_launcher: HexagonLauncherRPC,
-        evaluator_config: Optional[EvaluatorConfig] = None,
+        evaluator_config: EvaluatorConfig | None = None,
         cooldown_sec: float = 0.0,
         alloc_repeat: int = 1,
-        max_workers: Optional[int] = None,
-        initializer: Optional[Callable[[], None]] = None,
+        max_workers: int | None = None,
+        initializer: Callable[[], None] | None = None,
     ):
         """
         Parameters
@@ -89,7 +89,7 @@ def __init__(
             initializer=initializer,
         )
 
-    def run(self, runner_inputs: List[RunnerInput]) -> List[RunnerFuture]:
+    def run(self, runner_inputs: list[RunnerInput]) -> list[RunnerFuture]:
         results = []
         for runner_input in runner_inputs:
             future = RPCRunnerFuture(
@@ -131,7 +131,7 @@ def _worker_func(hexagon_launcher, evaluator_config, alloc_repeat, artifact_path
 
 def get_hexagon_local_builder(
     pass_context: tvm.transform.PassContext = None,
-    max_workers: Optional[int] = None,
+    max_workers: int | None = None,
     timeout_sec: float = 30.0,
 ):
     """Return Hexagon-compatible Builder for meta schedule."""
@@ -141,7 +141,7 @@ def export_func(mod):
         return str(binary_path)
 
     def default_build_with_context(
-        mod: IRModule, target: Target, _params: Optional[Dict[str, Tensor]]
+        mod: IRModule, target: Target, _params: dict[str, Tensor] | None
     ) -> Module:
         with pass_context:
             mod = RemoveWeightLayoutRewriteBlock(skip_tensor_rewrite=True)(mod)
@@ -163,7 +163,7 @@ def get_hexagon_rpc_runner(
     number=3,
     repeat=1,
     min_repeat_ms=100,
-    max_workers: Optional[int] = None,
+    max_workers: int | None = None,
 ):
     """Return Hexagon-compatible RPC Runner for meta schedule.
 
diff --git a/python/tvm/contrib/hexagon/pytest_plugin.py b/python/tvm/contrib/hexagon/pytest_plugin.py
index 4cc7b94d6cc2..41a93d5ed29f 100644
--- a/python/tvm/contrib/hexagon/pytest_plugin.py
+++ b/python/tvm/contrib/hexagon/pytest_plugin.py
@@ -21,7 +21,6 @@
 
 import os
 import random
-from typing import Optional, Union
 
 import pytest
 
@@ -66,7 +65,7 @@ def _compose(args, decs):
 requires_hexagon_toolchain = tvm.testing.requires_hexagon(support_required="compile-only")
 
 
-def android_serial_number() -> Optional[str]:
+def android_serial_number() -> str | None:
     """Return the android serial number"""
     serial = os.getenv(ANDROID_SERIAL_NUMBER, default="")
     # Setting ANDROID_SERIAL_NUMBER to an empty string should be
@@ -114,7 +113,7 @@ def get_free_port() -> int:
 
 
 @pytest.fixture(scope="session")
-def _tracker_info() -> Union[str, int]:
+def _tracker_info() -> str | int:
     env_tracker_host = os.getenv(TVM_TRACKER_HOST, default="")
     env_tracker_port = os.getenv(TVM_TRACKER_PORT, default="")
 
diff --git a/python/tvm/contrib/hexagon/session.py b/python/tvm/contrib/hexagon/session.py
index b92c78309482..4769d3aba127 100644
--- a/python/tvm/contrib/hexagon/session.py
+++ b/python/tvm/contrib/hexagon/session.py
@@ -21,7 +21,6 @@
 import os
 import pathlib
 import tempfile
-from typing import Union
 
 import tvm
 import tvm.contrib.hexagon as hexagon
@@ -62,7 +61,7 @@ class Session:
 
     def __init__(
         self,
-        remote_workspace: Union[str, pathlib.Path],
+        remote_workspace: str | pathlib.Path,
         rpc_tracker: tuple,
         rpc_server_key: str,
         serial_number: str,
@@ -142,7 +141,7 @@ def is_simulator(self):
     def get_function(self, name):
         return self._rpc.get_function(name)
 
-    def upload(self, local_path: Union[str, pathlib.Path], remote_filename: str) -> pathlib.Path:
+    def upload(self, local_path: str | pathlib.Path, remote_filename: str) -> pathlib.Path:
         """Upload a local file to the remote workspace.
 
         Parameters
@@ -164,7 +163,7 @@ def upload(self, local_path: Union[str, pathlib.Path], remote_filename: str) ->
         upload_func(remote_path, data)
         return remote_path
 
-    def load_module(self, module: Union[str, pathlib.Path, tvm.runtime.Module]):
+    def load_module(self, module: str | pathlib.Path | tvm.runtime.Module):
         """Load TVM module.
 
         The session must be established (via __enter__) prior to
@@ -197,13 +196,13 @@ def load_module(self, module: Union[str, pathlib.Path, tvm.runtime.Module]):
         else:
             remote_file_path = module
 
-        assert isinstance(remote_file_path, (str, pathlib.Path)), "Invalid path type:" + str(
+        assert isinstance(remote_file_path, str | pathlib.Path), "Invalid path type:" + str(
             type(remote_file_path)
         )
         return self._rpc.get_function("tvm.hexagon.load_module")(str(remote_file_path))
 
     def get_executor_from_factory(
-        self, module: Union[runtime.executable, str], hexagon_arch: str = "v68"
+        self, module: runtime.executable | str, hexagon_arch: str = "v68"
     ):
         """Create a local GraphModule which consumes a remote libmod.
 
@@ -217,12 +216,12 @@ def get_executor_from_factory(
         hexagon_arch : str
             The hexagon arch to be used
         """
-        if isinstance(module, (runtime.Executable, str)):
+        if isinstance(module, runtime.Executable | str):
             return self._vm_executable_executor(module, hexagon_arch=hexagon_arch)
 
         raise TypeError(f"Unsupported executor type: {type(module)}")
 
-    def _set_device_type(self, module: Union[str, pathlib.Path]):
+    def _set_device_type(self, module: str | pathlib.Path):
         """Set session device type(hexagon, cpu) based on target in module.
 
         Parameters
@@ -244,9 +243,7 @@ def _set_device_type(self, module: Union[str, pathlib.Path]):
             else:
                 self._requires_cpu_device = False
 
-    def _vm_executable_executor(
-        self, executable: Union[runtime.Executable, str], hexagon_arch: str
-    ):
+    def _vm_executable_executor(self, executable: runtime.Executable | str, hexagon_arch: str):
         """Create a local TVM module which consumes a remote vm executable.
 
         Parameters
diff --git a/python/tvm/contrib/hexagon/tools.py b/python/tvm/contrib/hexagon/tools.py
index f9d8b8d1ec48..c1e1191b5e61 100644
--- a/python/tvm/contrib/hexagon/tools.py
+++ b/python/tvm/contrib/hexagon/tools.py
@@ -19,13 +19,13 @@
 """Tools/compilers/linkers for Hexagon"""
 
 import io
+import itertools
 import os
 import pathlib
 import re
 import subprocess
 import sys
 import tarfile
-from typing import List, Union
 
 import numpy
 from tvm_ffi import register_global_func
@@ -76,7 +76,7 @@ def hexagon_clang_plus() -> str:
     return str(HEXAGON_CLANG_PLUS)
 
 
-def toolchain_version(toolchain=None) -> List[int]:
+def toolchain_version(toolchain=None) -> list[int]:
     """Return the version of the Hexagon toolchain.
 
     Parameters
@@ -249,7 +249,7 @@ def __create_shared_mac(so_name, objs, **kwargs):
     register_global_func("tvm.contrib.hexagon.link_shared", f=link_shared, override=True)
 
 
-def create_aot_shared(so_name: Union[str, pathlib.Path], files, hexagon_arch: str, options=None):
+def create_aot_shared(so_name: str | pathlib.Path, files, hexagon_arch: str, options=None):
     """Export Hexagon AOT module."""
     options = options or []
     if not os.access(str(HEXAGON_CLANG_PLUS), os.X_OK):
@@ -427,8 +427,7 @@ def allocate_hexagon_array(
 
     boundaries = [0, *axis_separators, len(tensor_shape)]
     physical_shape = [
-        numpy.prod(tensor_shape[dim_i:dim_f])
-        for dim_i, dim_f in zip(boundaries[:-1], boundaries[1:])
+        numpy.prod(tensor_shape[dim_i:dim_f]) for dim_i, dim_f in itertools.pairwise(boundaries)
     ]
 
     arr = tvm.runtime.empty(physical_shape, dtype=dtype, device=dev, mem_scope=mem_scope)
diff --git a/python/tvm/contrib/ndk.py b/python/tvm/contrib/ndk.py
index c0bbedb3700e..5c2de723cc66 100644
--- a/python/tvm/contrib/ndk.py
+++ b/python/tvm/contrib/ndk.py
@@ -17,14 +17,11 @@
 """Util to invoke NDK compiler toolchain."""
 
 # pylint: disable=invalid-name
-from __future__ import absolute_import as _abs
-
 import os
 import shutil
 import subprocess
 import tempfile
 from pathlib import Path
-from typing import Dict
 
 from tvm_ffi import register_global_func
 
@@ -134,7 +131,7 @@ def create_staticlib(output, inputs):
 create_staticlib.output_format = "a"
 
 
-def get_global_symbol_section_map(path, *, nm=None) -> Dict[str, str]:
+def get_global_symbol_section_map(path, *, nm=None) -> dict[str, str]:
     """Get global symbols from a library via nm -gU in NDK
 
     Parameters
diff --git a/python/tvm/contrib/nvcc.py b/python/tvm/contrib/nvcc.py
index b9770ab95e77..1f933a63fc7b 100644
--- a/python/tvm/contrib/nvcc.py
+++ b/python/tvm/contrib/nvcc.py
@@ -17,14 +17,11 @@
 # pylint: disable=invalid-name
 """Utility to invoke nvcc compiler in the system"""
 
-from __future__ import absolute_import as _abs
-
 import glob
 import os
 import platform
 import subprocess
 import warnings
-from typing import Tuple
 
 import tvm_ffi
 
@@ -298,7 +295,7 @@ def _compile_cuda_nvrtc(
         raise ValueError(f"target_format must be 'cubin' or 'ptx', got: {target_format}")
 
     # Validate options
-    if options is not None and not isinstance(options, (str, list)):
+    if options is not None and not isinstance(options, str | list):
         raise ValueError("options must be str or list of str")
 
     # Auto-detect architecture
@@ -686,7 +683,7 @@ def get_cuda_version(cuda_path=None):
     raise RuntimeError("Cannot read CUDA version file")
 
 
-def find_nvshmem_paths() -> Tuple[str, str]:
+def find_nvshmem_paths() -> tuple[str, str]:
     """
     Searches for the NVSHMEM include and library directories.
 
diff --git a/python/tvm/contrib/rpc.py b/python/tvm/contrib/rpc.py
index 5fbc06a9f1db..882a76b5689b 100644
--- a/python/tvm/contrib/rpc.py
+++ b/python/tvm/contrib/rpc.py
@@ -18,8 +18,6 @@
 """Deprecation RPC module"""
 
 # pylint: disable=unused-import
-from __future__ import absolute_import as _abs
-
 import warnings
 
 from ..rpc import LocalSession, RPCSession, Server, TrackerSession, connect, connect_tracker
diff --git a/python/tvm/contrib/tar.py b/python/tvm/contrib/tar.py
index 0053739da922..a43b4f339261 100644
--- a/python/tvm/contrib/tar.py
+++ b/python/tvm/contrib/tar.py
@@ -18,8 +18,6 @@
 """Util to invoke tarball in the system."""
 
 # pylint: disable=invalid-name
-from __future__ import absolute_import as _abs
-
 import os
 import shutil
 import subprocess
diff --git a/python/tvm/contrib/tvmjs.py b/python/tvm/contrib/tvmjs.py
index a419f7c4d782..084af45d2d1b 100644
--- a/python/tvm/contrib/tvmjs.py
+++ b/python/tvm/contrib/tvmjs.py
@@ -25,8 +25,9 @@
 
 # pylint: disable=unused-import
 import sys
+from collections.abc import Iterator, Mapping
 from types import GeneratorType
-from typing import Any, Iterator, Mapping, Optional, Set, Tuple, Union
+from typing import Any, Optional, Union
 
 import numpy as np
 
@@ -81,7 +82,7 @@ def __init__(
         cache_dir: str,
         prefix: str,
         shard_cap_nbytes: int,
-        initial_shard_records: Optional[Mapping[str, Any]] = None,
+        initial_shard_records: Mapping[str, Any] | None = None,
     ):
         self.cache_dir = cache_dir
         self.prefix = prefix
@@ -90,8 +91,8 @@ def __init__(
         self.shard_records = []
         self.shard_cap_nbytes = shard_cap_nbytes
         self.counter = 0
-        self.name_to_record: Mapping[str, Tuple[int, Mapping[str, Any]]] = {}
-        self.updated_shards: Set[int] = set()
+        self.name_to_record: Mapping[str, tuple[int, Mapping[str, Any]]] = {}
+        self.updated_shards: set[int] = set()
 
         if initial_shard_records is not None:
             self.shard_records = initial_shard_records
@@ -201,10 +202,8 @@ def pending_nbytes(self):
 
 
 def dump_tensor_cache(
-    params: Union[
-        Mapping[str, Union[np.ndarray, tvm.runtime.Tensor]],
-        Iterator[Tuple[str, Union[np.ndarray, tvm.runtime.Tensor]]],
-    ],
+    params: Mapping[str, np.ndarray | tvm.runtime.Tensor]
+    | Iterator[tuple[str, np.ndarray | tvm.runtime.Tensor]],
     cache_dir: str,
     encode_format="f32-to-bf16",
     meta_data=None,
diff --git a/python/tvm/contrib/utils.py b/python/tvm/contrib/utils.py
index 63611997c25c..ca0e70d817c0 100644
--- a/python/tvm/contrib/utils.py
+++ b/python/tvm/contrib/utils.py
@@ -131,7 +131,7 @@ def path(self):
         return pathlib.Path(self.temp_dir)
 
     def __truediv__(self, other):
-        if not isinstance(other, (str, pathlib.Path)):
+        if not isinstance(other, str | pathlib.Path):
             raise TypeError(
                 f"TempDirectory / operator: must supply str or pathlib.Path; got {other!r}"
             )
diff --git a/python/tvm/driver/build_module.py b/python/tvm/driver/build_module.py
index 4599090c40d6..8847b9031d60 100644
--- a/python/tvm/driver/build_module.py
+++ b/python/tvm/driver/build_module.py
@@ -19,7 +19,7 @@
 """The build utils in python."""
 
 import warnings
-from typing import Callable, Optional, Union
+from collections.abc import Callable
 
 import tvm
 from tvm.ir.module import IRModule
@@ -29,9 +29,9 @@
 
 
 def build(
-    mod: Union[PrimFunc, IRModule],
-    target: Optional[Union[str, Target]] = None,
-    pipeline: Optional[Union[str, tvm.transform.Pass]] = "default",
+    mod: PrimFunc | IRModule,
+    target: str | Target | None = None,
+    pipeline: str | tvm.transform.Pass | None = "default",
 ):
     """
     Build a function with a signature, generating code for devices
@@ -60,7 +60,7 @@ def build(
     return tvm.tir.build(mod, target, pipeline)
 
 
-def _contains_relax(mod: Union[PrimFunc, IRModule]) -> bool:
+def _contains_relax(mod: PrimFunc | IRModule) -> bool:
     if isinstance(mod, PrimFunc):
         return False
     if isinstance(mod, IRModule):
@@ -70,11 +70,11 @@ def _contains_relax(mod: Union[PrimFunc, IRModule]) -> bool:
 
 
 def compile(  # pylint: disable=redefined-builtin
-    mod: Union[PrimFunc, IRModule],
-    target: Optional[Target] = None,
+    mod: PrimFunc | IRModule,
+    target: Target | None = None,
     *,
-    relax_pipeline: Optional[Union[tvm.transform.Pass, Callable, str]] = "default",
-    tir_pipeline: Optional[Union[tvm.transform.Pass, Callable, str]] = "default",
+    relax_pipeline: tvm.transform.Pass | Callable | str | None = "default",
+    tir_pipeline: tvm.transform.Pass | Callable | str | None = "default",
 ) -> Executable:
     """
     Compile an IRModule to a runtime executable.
diff --git a/python/tvm/exec/disco_worker.py b/python/tvm/exec/disco_worker.py
index 505dbff5c477..d247a3c2e9c9 100644
--- a/python/tvm/exec/disco_worker.py
+++ b/python/tvm/exec/disco_worker.py
@@ -20,7 +20,7 @@
 
 import os
 import sys
-from typing import Callable
+from collections.abc import Callable
 
 from tvm_ffi import get_global_func, register_global_func
 
diff --git a/python/tvm/ir/analysis.py b/python/tvm/ir/analysis.py
index c79e382c00e6..2baf41f8e064 100644
--- a/python/tvm/ir/analysis.py
+++ b/python/tvm/ir/analysis.py
@@ -19,8 +19,6 @@
 
 """Common analysis across all IR variants."""
 
-from typing import Dict, List
-
 import tvm
 
 from . import _ffi_analysis_api as _ffi
@@ -28,7 +26,7 @@
 
 def collect_call_map(
     module: "tvm.ir.IRModule",
-) -> Dict["tvm.ir.GlobalVar", List["tvm.ir.GlobalVar"]]:
+) -> dict["tvm.ir.GlobalVar", list["tvm.ir.GlobalVar"]]:
     """Collect the call map of a module
 
     Parameters
diff --git a/python/tvm/ir/expr.py b/python/tvm/ir/expr.py
index ba3dc305ea04..b534213f8164 100644
--- a/python/tvm/ir/expr.py
+++ b/python/tvm/ir/expr.py
@@ -32,7 +32,7 @@
 class BaseExpr(Node):
     """Base class of all the expressions."""
 
-    span: Optional[Span]
+    span: Span | None
 
 
 @tvm_ffi.register_object("ir.PrimExpr")
@@ -101,7 +101,7 @@ def __call__(self, *args: RelaxExpr) -> BaseExpr:
 
             return relax.Call(self, args)
 
-        elif all(isinstance(x, (Number, PrimExpr)) for x in args):
+        elif all(isinstance(x, Number | PrimExpr) for x in args):
             return tvm.tir.call_tir(self, *args)
 
         arg_types = [type(x) for x in args]
@@ -135,17 +135,15 @@ class Range(Node, Scriptable):
 
     min: PrimExpr
     extent: PrimExpr
-    span: Optional[Span]
+    span: Span | None
 
     def __init__(
-        self, begin: PrimExpr, end: Optional[PrimExpr] = None, span: Optional[Span] = None
+        self, begin: PrimExpr, end: PrimExpr | None = None, span: Span | None = None
     ) -> None:
         self.__init_handle_by_constructor__(_ffi_api.Range, begin, end, span)
 
     @staticmethod
-    def from_min_extent(
-        min_value: PrimExpr, extent: PrimExpr, span: Optional[Span] = None
-    ) -> "Range":
+    def from_min_extent(min_value: PrimExpr, extent: PrimExpr, span: Span | None = None) -> "Range":
         """Construct a Range by min and extent.
 
         This constructs a range in [min_value, min_value + extent)
diff --git a/python/tvm/ir/function.py b/python/tvm/ir/function.py
index f7928350be2d..4a1317022624 100644
--- a/python/tvm/ir/function.py
+++ b/python/tvm/ir/function.py
@@ -18,7 +18,6 @@
 """Function definitions."""
 
 from enum import IntEnum
-from typing import Dict, Union
 
 import tvm_ffi
 
@@ -76,7 +75,7 @@ def with_attr(self, attr_key_or_dict, attr_value=None) -> "BaseFunc":
             res._move(), attr_key_or_dict, tvm.runtime.convert(attr_value)
         )
 
-    def with_attrs(self, attr_map: Union[DictAttrs, Dict[str, Object]]) -> "BaseFunc":
+    def with_attrs(self, attr_map: DictAttrs | dict[str, Object]) -> "BaseFunc":
         """Copy the IRModule and add the given attribute map to it.
         Parameters
         ----------
diff --git a/python/tvm/ir/global_info.py b/python/tvm/ir/global_info.py
index a56c6d028a16..1039320f4bc5 100644
--- a/python/tvm/ir/global_info.py
+++ b/python/tvm/ir/global_info.py
@@ -61,7 +61,7 @@ def __init__(
         vdevice_id: int = 0,
         memory_scope: str = "global",
     ) -> None:
-        if isinstance(target, (dict, str)):
+        if isinstance(target, dict | str):
             target = tvm.target.Target(tvm.runtime.convert(target))
         if isinstance(target, Device):
             target = tvm.target.Target.from_device(target)
diff --git a/python/tvm/ir/instrument.py b/python/tvm/ir/instrument.py
index 48774ce0f68c..e3bc201e06d1 100644
--- a/python/tvm/ir/instrument.py
+++ b/python/tvm/ir/instrument.py
@@ -22,7 +22,6 @@
 import re
 import shutil
 from pathlib import Path
-from typing import Union
 
 import tvm_ffi
 
@@ -300,7 +299,7 @@ def run_before_pass(self, mod, info):
 class DumpIR:
     """Dump the IR after the pass runs."""
 
-    def __init__(self, dump_dir: Union[Path, str], refresh: bool = False):
+    def __init__(self, dump_dir: Path | str, refresh: bool = False):
         if isinstance(dump_dir, Path):
             self.dump_dir = dump_dir
         else:
diff --git a/python/tvm/ir/module.py b/python/tvm/ir/module.py
index f73db9463e63..b64baa7edc93 100644
--- a/python/tvm/ir/module.py
+++ b/python/tvm/ir/module.py
@@ -16,10 +16,6 @@
 # under the License.
 """IRModule that holds the functions and type definitions."""
 
-from __future__ import annotations
-
-from typing import Dict, Union
-
 import tvm_ffi
 
 import tvm
@@ -70,7 +66,7 @@ def __init__(self, functions=None, attrs=None, global_infos=None):
         )
         self.pyfuncs = {}
 
-    def clone(self) -> IRModule:
+    def clone(self) -> "IRModule":
         return _ffi_api.Module_Clone(self)
 
     def functions_items(self):
@@ -125,10 +121,10 @@ def __getitem__(self, var):
         assert isinstance(var, _expr.GlobalVar)
         return _ffi_api.Module_Lookup(self, var)
 
-    def __delitem__(self, var: Union[str, _expr.GlobalVar]):
+    def __delitem__(self, var: str | _expr.GlobalVar):
         _ffi_api.Module_Remove(self, var)
 
-    def __contains__(self, var: Union[str, _expr.GlobalVar]) -> bool:
+    def __contains__(self, var: str | _expr.GlobalVar) -> bool:
         return _ffi_api.Module_Contains(self, var)
 
     def update(self, other):
@@ -202,8 +198,8 @@ def get_global_vars(self):
 
     def replace_global_vars(
         self,
-        replacements: Dict[Union[str, _expr.GlobalVar], Union[str, _expr.GlobalVar]],
-    ) -> IRModule:
+        replacements: dict[str | _expr.GlobalVar, str | _expr.GlobalVar],
+    ) -> "IRModule":
         """Replace GlobalVar instances within the module
 
         Replace GlobalVars within the IRModule.  Since the IRModule
@@ -284,7 +280,7 @@ def with_attr(self, attr_key, attr_value):
 
         return _ffi_api.Module_WithAttr(self, attr_key, attr_value)
 
-    def without_attr(self, attr_key: str) -> IRModule:
+    def without_attr(self, attr_key: str) -> "IRModule":
         """Copy the IRModule and remove an attribute key and its associated value.
         Parameters
         ----------
@@ -298,7 +294,7 @@ def without_attr(self, attr_key: str) -> IRModule:
 
         return _ffi_api.Module_WithoutAttr(self, attr_key)
 
-    def with_attrs(self, attr_map: Union[DictAttrs, Dict[str, Object]]) -> IRModule:
+    def with_attrs(self, attr_map: DictAttrs | dict[str, Object]) -> "IRModule":
         """Copy the IRModule and add the given attribute map to it.
         Parameters
         ----------
diff --git a/python/tvm/ir/supply.py b/python/tvm/ir/supply.py
index 8e1a7986b666..8e73ec4d1205 100644
--- a/python/tvm/ir/supply.py
+++ b/python/tvm/ir/supply.py
@@ -100,7 +100,7 @@ def __init__(self, value=None):
             self.__init_handle_by_constructor__(_ffi_api.GlobalVarSupply_NameSupply, name_supply)
         elif isinstance(value, NameSupply):
             self.__init_handle_by_constructor__(_ffi_api.GlobalVarSupply_NameSupply, value)
-        elif isinstance(value, (list, tvm.container.Array)):
+        elif isinstance(value, list | tvm.container.Array):
             self.__init_handle_by_constructor__(_ffi_api.GlobalVarSupply_IRModules, value)
         elif isinstance(value, IRModule):
             self.__init_handle_by_constructor__(_ffi_api.GlobalVarSupply_IRModule, value)
diff --git a/python/tvm/ir/transform.py b/python/tvm/ir/transform.py
index 9b7d8872dc05..28c187d7deb8 100644
--- a/python/tvm/ir/transform.py
+++ b/python/tvm/ir/transform.py
@@ -84,15 +84,15 @@ def __init__(
         config=None,
     ):
         required = list(required_pass) if required_pass else []
-        if not isinstance(required, (list, tuple)):
+        if not isinstance(required, list | tuple):
             raise TypeError("required_pass is expected to be the type of " + "list/tuple/set.")
 
         disabled = list(disabled_pass) if disabled_pass else []
-        if not isinstance(disabled, (list, tuple)):
+        if not isinstance(disabled, list | tuple):
             raise TypeError("disabled_pass is expected to be the type of " + "list/tuple/set.")
 
         instruments = list(instruments) if instruments else []
-        if not isinstance(instruments, (list, tuple)):
+        if not isinstance(instruments, list | tuple):
             raise TypeError("instruments is expected to be the type of " + "list/tuple/set.")
 
         config = config if config else None
@@ -209,11 +209,11 @@ class Sequential(Pass):
 
     def __init__(self, passes=None, opt_level=0, name="sequential", required=None, traceable=False):
         passes = passes if passes else []
-        if not isinstance(passes, (list, tuple)):
+        if not isinstance(passes, list | tuple):
             raise TypeError("passes must be a list of Pass objects.")
 
         required = required if required else []
-        if not isinstance(required, (list, tuple)):
+        if not isinstance(required, list | tuple):
             raise TypeError("Required is expected to be the type of list/tuple.")
 
         self.__init_handle_by_constructor__(
@@ -332,7 +332,7 @@ def transform(mod, ctx):
         raise ValueError("Please provide opt_level for the module pass.")
 
     required = required if required else []
-    if not isinstance(required, (list, tuple)):
+    if not isinstance(required, list | tuple):
         raise TypeError("Required is expected to be the type of " + "list/tuple.")
 
     def create_module_pass(pass_arg):
diff --git a/python/tvm/relax/analysis/analysis.py b/python/tvm/relax/analysis/analysis.py
index 3545203b5367..d3bf850740dc 100644
--- a/python/tvm/relax/analysis/analysis.py
+++ b/python/tvm/relax/analysis/analysis.py
@@ -21,8 +21,8 @@
 configuring the passes and scripting them in Python.
 """
 
+from collections.abc import Callable
 from enum import IntEnum
-from typing import Callable, Dict, List, Optional, Union
 
 import tvm
 from tvm import IRModule, tir
@@ -52,8 +52,8 @@ def get_static_type(sinfo: StructInfo) -> Type:
 
 def erase_to_well_defined(
     sinfo: StructInfo,
-    shape_var_map: Optional[Dict[tir.Var, tir.PrimExpr]] = None,
-    var_map: Optional[Dict[Var, Expr]] = None,
+    shape_var_map: dict[tir.Var, tir.PrimExpr] | None = None,
+    var_map: dict[Var, Expr] | None = None,
 ) -> StructInfo:
     """Erase sinfo into a well defined form.
 
@@ -167,7 +167,7 @@ def struct_info_lca(lhs: StructInfo, rhs: StructInfo) -> StructInfo:
     return _ffi_api.StructInfoLCA(lhs, rhs)  # type: ignore
 
 
-def tir_vars_in_struct_info(sinfo: StructInfo) -> List[tir.Var]:
+def tir_vars_in_struct_info(sinfo: StructInfo) -> list[tir.Var]:
     """Get the TIR variables that appear in the input struct info.
     The returned list is deduplicated - each TIR variable will appear at most once.
 
@@ -184,7 +184,7 @@ def tir_vars_in_struct_info(sinfo: StructInfo) -> List[tir.Var]:
     return _ffi_api.TIRVarsInStructInfo(sinfo)  # type: ignore
 
 
-def definable_tir_vars_in_struct_info(sinfo: StructInfo) -> List[tir.Var]:
+def definable_tir_vars_in_struct_info(sinfo: StructInfo) -> list[tir.Var]:
     """Get the TIR variables that may be defined from input struct info.
     The returned list is deduplicated - each TIR variable will appear at most once.
 
@@ -202,7 +202,7 @@ def definable_tir_vars_in_struct_info(sinfo: StructInfo) -> List[tir.Var]:
     return _ffi_api.DefinableTIRVarsInStructInfo(sinfo)  # type: ignore
 
 
-def collect_non_negative_expressions(sinfo: StructInfo) -> List[tir.PrimExpr]:
+def collect_non_negative_expressions(sinfo: StructInfo) -> list[tir.PrimExpr]:
     """Collect TIR expressions used in non-negative contexts
 
     Get TIR variables that are non-negative within the context where
@@ -229,7 +229,7 @@ def collect_non_negative_expressions(sinfo: StructInfo) -> List[tir.PrimExpr]:
     return _ffi_api.CollectNonNegativeExpressions(sinfo)  # type: ignore
 
 
-def defined_symbolic_vars(func: Function) -> List[Var]:
+def defined_symbolic_vars(func: Function) -> list[Var]:
     """Get the TIR variables that defined in the input function.
     The returned list is deduplicated - each TIR variable will appear at most once.
 
@@ -246,7 +246,7 @@ def defined_symbolic_vars(func: Function) -> List[Var]:
     return _ffi_api.DefinedSymbolicVars(func)  # type: ignore
 
 
-def free_symbolic_vars(func: Function) -> List[Var]:
+def free_symbolic_vars(func: Function) -> list[Var]:
     """Get the TIR variables that are used but not defined in the input function.
     The returned list is deduplicated - each TIR variable will appear at most once.
 
@@ -263,7 +263,7 @@ def free_symbolic_vars(func: Function) -> List[Var]:
     return _ffi_api.FreeSymbolicVars(func)  # type: ignore
 
 
-def bound_vars(expr: Expr) -> List[Var]:
+def bound_vars(expr: Expr) -> list[Var]:
     """
     Return all bound variables from expression expr.
     Bound variables are all variables that are declared in the expr.
@@ -280,7 +280,7 @@ def bound_vars(expr: Expr) -> List[Var]:
     return _ffi_api.bound_vars(expr)
 
 
-def free_vars(expr: Expr) -> List[Var]:
+def free_vars(expr: Expr) -> list[Var]:
     """
     Return all free variables from expression expr.
     Free variables are variables that are not bound by a
@@ -297,7 +297,7 @@ def free_vars(expr: Expr) -> List[Var]:
     return _ffi_api.free_vars(expr)
 
 
-def all_vars(expr: Expr) -> List[Var]:
+def all_vars(expr: Expr) -> list[Var]:
     """
     Return all (local) variables from expression expr.
     Parameters
@@ -312,7 +312,7 @@ def all_vars(expr: Expr) -> List[Var]:
     return _ffi_api.all_vars(expr)
 
 
-def used_vars(expr: Expr) -> List[Var]:
+def used_vars(expr: Expr) -> list[Var]:
     """
     Return all variables used in an expression.
 
@@ -332,7 +332,7 @@ def used_vars(expr: Expr) -> List[Var]:
     return _ffi_api.used_vars(expr)  # type: ignore
 
 
-def all_global_vars(expr: Expr) -> List[GlobalVar]:
+def all_global_vars(expr: Expr) -> list[GlobalVar]:
     """
     Return all global variables from expression expr.
     Parameters
@@ -392,7 +392,7 @@ def has_reshape_pattern(func: tir.PrimFunc) -> bool:
     return _ffi_api.has_reshape_pattern(func)  # type: ignore
 
 
-def contains_impure_call(expr: Expr, own_name: Optional[Union[Var, GlobalVar]] = None) -> bool:
+def contains_impure_call(expr: Expr, own_name: Var | GlobalVar | None = None) -> bool:
     """
     Check if the given expression (likely a function body) contains any impure calls.
 
@@ -420,7 +420,7 @@ def contains_impure_call(expr: Expr, own_name: Optional[Union[Var, GlobalVar]] =
     return _ffi_api.contains_impure_call(expr, own_name)
 
 
-def get_var2val(func: Function) -> Dict[Var, Expr]:
+def get_var2val(func: Function) -> dict[Var, Expr]:
     """
     Get a mapping from Var to Expr for each variable in the function.
 
@@ -437,7 +437,7 @@ def get_var2val(func: Function) -> Dict[Var, Expr]:
     return _ffi_api.get_var2val(func)  # type: ignore
 
 
-def udchain(dfb: DataflowBlock) -> Dict[Var, List[Var]]:
+def udchain(dfb: DataflowBlock) -> dict[Var, list[Var]]:
     """
     Analyze the variable use-def chain in a dataflow block.
 
@@ -454,7 +454,7 @@ def udchain(dfb: DataflowBlock) -> Dict[Var, List[Var]]:
     return _ffi_api.udchain(dfb)  # type: ignore
 
 
-def name_to_binding(func: Function) -> Dict[str, List[Binding]]:
+def name_to_binding(func: Function) -> dict[str, list[Binding]]:
     """Return a map from variable name to its bindings."""
     return _ffi_api.name_to_binding(func)  # type: ignore
 
@@ -481,7 +481,7 @@ def remove_all_unused(func: Function) -> Function:
     return _ffi_api.remove_all_unused(func)  # type: ignore
 
 
-def well_formed(obj: Union[IRModule, Function], check_struct_info: bool = True) -> bool:
+def well_formed(obj: IRModule | Function, check_struct_info: bool = True) -> bool:
     """Check if the IRModule is well formed.
 
     Parameters
@@ -516,8 +516,8 @@ def _get_prim_func_default_dtype(func: PrimFunc):
 
 
 def suggest_layout_transforms(
-    func: PrimFunc, write_buffer_transforms: List[Union[IndexMap, Callable]]
-) -> Dict[SBlock, Dict[Union[SBlock, Buffer], IndexMap]]:
+    func: PrimFunc, write_buffer_transforms: list[IndexMap | Callable]
+) -> dict[SBlock, dict[SBlock | Buffer, IndexMap]]:
     """Suggest Layout transformations of blocks and buffers in a PrimFunc.
 
     Parameters
@@ -545,7 +545,7 @@ def suggest_layout_transforms(
     return _ffi_api.suggest_layout_transforms(func, write_buffer_index_maps)  # type: ignore
 
 
-def detect_recursion(mod: tvm.IRModule) -> List[List[GlobalVar]]:
+def detect_recursion(mod: tvm.IRModule) -> list[list[GlobalVar]]:
     """
     Find all sets of recursive or mutually recursive functions in the module.
 
@@ -577,7 +577,7 @@ def detect_recursion(mod: tvm.IRModule) -> List[List[GlobalVar]]:
     return _ffi_api.detect_recursion(mod)  # type: ignore
 
 
-def computable_at_compile_time(func: Function) -> List[Var]:
+def computable_at_compile_time(func: Function) -> list[Var]:
     """Collect variables whose value can be computed at compile-time
 
     If a function has the `kNumInput` attribute, then the first
diff --git a/python/tvm/relax/analysis/estimate_memory_usage.py b/python/tvm/relax/analysis/estimate_memory_usage.py
index ec4ed4176eac..0de0c7ea344f 100644
--- a/python/tvm/relax/analysis/estimate_memory_usage.py
+++ b/python/tvm/relax/analysis/estimate_memory_usage.py
@@ -16,7 +16,6 @@
 # under the License.
 # pylint: disable=abstract-method,unused-argument
 # pylint: disable=missing-function-docstring,missing-module-docstring
-from typing import Union
 
 import tvm
 from tvm.ir import Op
@@ -26,7 +25,7 @@
 from ..expr_functor import PyExprVisitor, visitor
 
 
-def estimate_memory_usage(mod: Union[IRModule, Function]) -> str:
+def estimate_memory_usage(mod: IRModule | Function) -> str:
     """Analysis function that estimates the memory usage of Relax functions
     in an IRModule. The estimation includes the total memory size needed to
     be allocated before and after memory planning.
diff --git a/python/tvm/relax/backend/adreno/transform/transform.py b/python/tvm/relax/backend/adreno/transform/transform.py
index f38d94b5f05d..0a68f560433b 100644
--- a/python/tvm/relax/backend/adreno/transform/transform.py
+++ b/python/tvm/relax/backend/adreno/transform/transform.py
@@ -17,15 +17,13 @@
 # pylint: disable=invalid-name
 """Adreno Relax transformation passes."""
 
-from typing import Optional
-
 import tvm.ir
 from tvm.target import Target
 
 from . import _ffi_api
 
 
-def AnnotateCustomMemoryScope(target: Optional[Target] = None) -> tvm.ir.transform.Pass:
+def AnnotateCustomMemoryScope(target: Target | None = None) -> tvm.ir.transform.Pass:
     """Allocate the memory scope information. This is Adreno specific pass to annotate
     The memory scope information and realize the same with RealizeVDevice pass followed by
     updating the Prim Function var_buffer mapping using SpecializePrimFuncBasedOnCallSite.
diff --git a/python/tvm/relax/backend/contrib/nnapi.py b/python/tvm/relax/backend/contrib/nnapi.py
index 0f4089627743..b8f9125cd5d3 100644
--- a/python/tvm/relax/backend/contrib/nnapi.py
+++ b/python/tvm/relax/backend/contrib/nnapi.py
@@ -17,12 +17,7 @@
 
 """Pattern table for NNAPI backend"""
 
-from typing import (
-    List,
-    Mapping,
-    Optional,
-    Tuple,
-)
+from collections.abc import Mapping
 
 from tvm.ir import IRModule
 from tvm.relax.dpl.pattern import (
@@ -35,7 +30,7 @@
 from ..pattern_registry import get_patterns_with_prefix, register_patterns
 
 
-def elementwise_binary_patterns() -> List[Tuple[str, DFPattern, Mapping[str, DFPattern]]]:
+def elementwise_binary_patterns() -> list[tuple[str, DFPattern, Mapping[str, DFPattern]]]:
     """
     Returns a list of tuples representing elementwise binary operation patterns mapped
     between NNAPI and Relax frameworks.
@@ -44,7 +39,7 @@ def elementwise_binary_patterns() -> List[Tuple[str, DFPattern, Mapping[str, DFP
     def _elementwise_binary_pattern(
         pattern_name: str,
         op_name: str,
-    ) -> Tuple[str, DFPattern, Mapping[str, DFPattern]]:
+    ) -> tuple[str, DFPattern, Mapping[str, DFPattern]]:
         input0 = wildcard()
         input1 = wildcard()
 
@@ -69,7 +64,7 @@ def _elementwise_binary_pattern(
     ]
 
 
-def unary_patterns() -> List[Tuple[str, DFPattern, Mapping[str, DFPattern]]]:
+def unary_patterns() -> list[tuple[str, DFPattern, Mapping[str, DFPattern]]]:
     """
     Returns a list of tuples representing unary operation patterns mapped
     between NNAPI and Relax frameworks.
@@ -77,7 +72,7 @@ def unary_patterns() -> List[Tuple[str, DFPattern, Mapping[str, DFPattern]]]:
 
     def _unary_pattern(
         pattern_name: str, op_name: str
-    ) -> Tuple[str, DFPattern, Mapping[str, DFPattern]]:
+    ) -> tuple[str, DFPattern, Mapping[str, DFPattern]]:
         input0 = wildcard()
         pattern = is_op(op_name)(input0)
         return (pattern_name, pattern, {})
@@ -98,7 +93,7 @@ def _unary_pattern(
     ]
 
 
-def matmul_pattern() -> Tuple[str, DFPattern, Mapping[str, DFPattern]]:
+def matmul_pattern() -> tuple[str, DFPattern, Mapping[str, DFPattern]]:
     """
     Returns a tuple representing matmul operation patterns mapped
     between NNAPI and Relax frameworks.
@@ -109,7 +104,7 @@ def matmul_pattern() -> Tuple[str, DFPattern, Mapping[str, DFPattern]]:
     return ("nnapi.batch_matmul", pattern, {})
 
 
-def permute_dims_pattern() -> Tuple[str, DFPattern, Mapping[str, DFPattern]]:
+def permute_dims_pattern() -> tuple[str, DFPattern, Mapping[str, DFPattern]]:
     """
     Returns a tuple representing permute operation patterns mapped
     between NNAPI and Relax frameworks.
@@ -119,7 +114,7 @@ def permute_dims_pattern() -> Tuple[str, DFPattern, Mapping[str, DFPattern]]:
     return ("nnapi.transpose", pattern, {})
 
 
-def astype_pattern() -> Tuple[str, DFPattern, Mapping[str, DFPattern]]:
+def astype_pattern() -> tuple[str, DFPattern, Mapping[str, DFPattern]]:
     """
     Returns a tuple representing astype operation patterns mapped
     between NNAPI and Relax frameworks.
@@ -132,7 +127,7 @@ def astype_pattern() -> Tuple[str, DFPattern, Mapping[str, DFPattern]]:
     return ("nnapi.cast", pattern, {})
 
 
-def mean_pattern() -> Tuple[str, DFPattern, Mapping[str, DFPattern]]:
+def mean_pattern() -> tuple[str, DFPattern, Mapping[str, DFPattern]]:
     """
     Returns a tuple representing mean operation patterns mapped
     between NNAPI and Relax frameworks.
@@ -143,7 +138,7 @@ def mean_pattern() -> Tuple[str, DFPattern, Mapping[str, DFPattern]]:
     return ("nnapi.mean", pattern, {})
 
 
-def conv2d_pattern() -> Tuple[str, DFPattern, Mapping[str, DFPattern]]:
+def conv2d_pattern() -> tuple[str, DFPattern, Mapping[str, DFPattern]]:
     """
     Returns a tuple representing conv2d operation patterns mapped
     between NNAPI and Relax frameworks.
@@ -156,7 +151,7 @@ def conv2d_pattern() -> Tuple[str, DFPattern, Mapping[str, DFPattern]]:
     return ("nnapi.conv2d", pattern, {})
 
 
-def max_pool2d_pattern() -> Tuple[str, DFPattern, Mapping[str, DFPattern]]:
+def max_pool2d_pattern() -> tuple[str, DFPattern, Mapping[str, DFPattern]]:
     """
     Returns a tuple representing max_pool2d operation patterns mapped
     between NNAPI and Relax frameworks.
@@ -303,7 +298,7 @@ def min_feature_level(pattern_name: str) -> int:
     return levels[pattern_name]
 
 
-def partition_for_nnapi(mod: IRModule, feature_level: Optional[int] = None) -> IRModule:
+def partition_for_nnapi(mod: IRModule, feature_level: int | None = None) -> IRModule:
     """Partition the graph greedily offloading supported operators to NNAPI.
 
     Parameters
diff --git a/python/tvm/relax/backend/cuda/cublas.py b/python/tvm/relax/backend/cuda/cublas.py
index 53948dc48b12..006f2153af9d 100644
--- a/python/tvm/relax/backend/cuda/cublas.py
+++ b/python/tvm/relax/backend/cuda/cublas.py
@@ -74,7 +74,7 @@ def _check_matmul(context: PatternCheckContext) -> bool:
     lhs_shape = lhs.struct_info.shape.values
     rhs_shape = rhs.struct_info.shape.values
 
-    if not isinstance(lhs_shape[-1], (tvm.tir.expr.IntImm, int)):
+    if not isinstance(lhs_shape[-1], tvm.tir.expr.IntImm | int):
         # Reduction axis must be constant
         return False
 
@@ -82,7 +82,7 @@ def _check_matmul(context: PatternCheckContext) -> bool:
         if lhs_shape[-1] % 4 != 0:
             # Reduction axis must be multiples of 4 for IGEMM
             return False
-        if not isinstance(rhs_shape[-1], (tvm.tir.expr.IntImm, int)) or rhs_shape[-1] % 4 != 0:
+        if not isinstance(rhs_shape[-1], tvm.tir.expr.IntImm | int) or rhs_shape[-1] % 4 != 0:
             # Rows number must be multiples of 4 for IGEMM
             return False
     elif lhs_dtype == "float8_e4m3fn" and rhs_dtype == "float8_e4m3fn":
@@ -102,12 +102,12 @@ def _check_matmul(context: PatternCheckContext) -> bool:
 
         # cuBLAS FP8 operations require all tensors being aligned to 16 bytes.
         if (
-            not isinstance(rhs_shape[-1], (tvm.tir.expr.IntImm, int))
+            not isinstance(rhs_shape[-1], tvm.tir.expr.IntImm | int)
             or rhs_shape[-1] % (16 // DataType(lhs_dtype).itemsize) != 0
         ):
             return False
         if (
-            not isinstance(rhs_shape[-2], (tvm.tir.expr.IntImm, int))
+            not isinstance(rhs_shape[-2], tvm.tir.expr.IntImm | int)
             or rhs_shape[-2] % (16 // DataType(out_dtype).itemsize) != 0
         ):
             return False
@@ -122,7 +122,7 @@ def _check_matmul(context: PatternCheckContext) -> bool:
         bias = context.annotated_expr["bias"]
         bias_shape = bias.struct_info.shape.values
         bias_batches = reduce(operator.mul, bias_shape[:-1], 1)
-        if not isinstance(bias_batches, (tvm.tir.expr.IntImm, int)) or int(bias_batches) > 1:
+        if not isinstance(bias_batches, tvm.tir.expr.IntImm | int) or int(bias_batches) > 1:
             # cuBLAS only supports bias vector
             return False
 
diff --git a/python/tvm/relax/backend/cuda/cudnn.py b/python/tvm/relax/backend/cuda/cudnn.py
index dfbdfe4ebee1..3cc8720f4269 100644
--- a/python/tvm/relax/backend/cuda/cudnn.py
+++ b/python/tvm/relax/backend/cuda/cudnn.py
@@ -182,7 +182,7 @@ def visit_function_(self, f):
             out_size_1d = _shape_1d(f.ret_struct_info.shape)
             # This needs to be in sync with the actual value that the kernel expects.
             workspace_size_bytes = out_size_1d * {"float16": 2, "float32": 4}[out_dtype]
-            if not isinstance(workspace_size_bytes, (int, tvm.tir.expr.IntImm)):
+            if not isinstance(workspace_size_bytes, int | tvm.tir.expr.IntImm):
                 # Tempororay workaround for dynamic shape workload. Will be removed when
                 # workspace for dynamic shape workload is implemented.
                 workspace_size_bytes = 8
diff --git a/python/tvm/relax/backend/cuda/cutlass.py b/python/tvm/relax/backend/cuda/cutlass.py
index 1e57df319f7d..c656a6ae77f9 100644
--- a/python/tvm/relax/backend/cuda/cutlass.py
+++ b/python/tvm/relax/backend/cuda/cutlass.py
@@ -19,8 +19,8 @@
 """Pattern table for CUTLASS backend"""
 
 import operator
+from collections.abc import Mapping, Sequence
 from functools import reduce
-from typing import Mapping, Sequence
 
 import tvm
 from tvm.contrib.cutlass.build import is_shape_valid_for_cutlass_matmul
@@ -551,7 +551,7 @@ def visit_function_(self, f):
             out_size_1d = _shape_1d(f.ret_struct_info.shape)
             # This needs to be in sync with the actual value that the kernel expects.
             workspace_size_bytes = out_size_1d * {"float16": 2, "float32": 4}[out_dtype]
-            if not isinstance(workspace_size_bytes, (int, tvm.tir.expr.IntImm)):
+            if not isinstance(workspace_size_bytes, int | tvm.tir.expr.IntImm):
                 # Tempororay workaround for dynamic shape workload. Will be removed when
                 # workspace for dynamic shape workload is implemented.
                 workspace_size_bytes = 8
diff --git a/python/tvm/relax/backend/cuda/flashinfer.py b/python/tvm/relax/backend/cuda/flashinfer.py
index cab8fac24c4b..a6e1fd995456 100644
--- a/python/tvm/relax/backend/cuda/flashinfer.py
+++ b/python/tvm/relax/backend/cuda/flashinfer.py
@@ -19,13 +19,12 @@
 
 import re
 from pathlib import Path
-from typing import List
 
 import tvm
 from tvm.target import Target
 
 
-def _rename_exported_func_names(source_paths: List[Path], prefix: str):
+def _rename_exported_func_names(source_paths: list[Path], prefix: str):
     """Rename the ffi-exported function names in the source files to the given prefix."""
     pattern = re.compile(r"^(\s*TVM_FFI_DLL_EXPORT_TYPED_FUNC\()([A-Za-z0-9_]+)(,.*)$")
     for source_path in source_paths:
@@ -49,7 +48,7 @@ def _rename_exported_func_names(source_paths: List[Path], prefix: str):
             source_path.write_text("".join(lines), encoding="utf-8")
 
 
-def _load_flashinfer_modules(object_files: List[Path]) -> List[tvm.runtime.Module]:
+def _load_flashinfer_modules(object_files: list[Path]) -> list[tvm.runtime.Module]:
     return [
         tvm.runtime.load_static_library(str(obj_path.absolute()), func_names=[])
         for obj_path in object_files
@@ -64,7 +63,7 @@ def gen_flashinfer_prefill_module(
     v_head_dim: int,
     enable_inline_rope: bool,
     return_static_libs: bool = False,
-) -> List[tvm.runtime.Module]:
+) -> list[tvm.runtime.Module]:
     """Generate a FlashInfer module for prefill.
 
     Parameters
@@ -159,7 +158,7 @@ def gen_flashinfer_decode_module(
     v_head_dim: int,
     enable_inline_rope: bool,
     return_static_libs: bool = False,
-) -> List[tvm.runtime.Module]:
+) -> list[tvm.runtime.Module]:
     """Generate a FlashInfer module for decode.
 
     Parameters
@@ -237,7 +236,7 @@ def gen_flashinfer_mla_module(
     head_dim_ckv: int,
     head_dim_kpe: int,
     return_static_libs: bool = False,
-) -> List[tvm.runtime.Module]:
+) -> list[tvm.runtime.Module]:
     """Generate a FlashInfer module for MLA.
 
     Parameters
@@ -301,7 +300,7 @@ def gen_flashinfer_mla_module(
 
 def gen_grouped_gemm_module(
     target: Target, return_static_libs: bool = False
-) -> List[tvm.runtime.Module]:
+) -> list[tvm.runtime.Module]:
     """Generate a FlashInfer module for FP8 grouped GEMM.
 
     Parameters
diff --git a/python/tvm/relax/backend/dispatch_sort_scan.py b/python/tvm/relax/backend/dispatch_sort_scan.py
index 62290cf33388..b0d74769d9a8 100644
--- a/python/tvm/relax/backend/dispatch_sort_scan.py
+++ b/python/tvm/relax/backend/dispatch_sort_scan.py
@@ -19,7 +19,6 @@
 
 from functools import reduce
 from operator import mul
-from typing import Dict
 
 from tvm import DataType, relax, topi
 from tvm.contrib.thrust import can_use_thrust
@@ -36,7 +35,7 @@
 class SortScanDispatcher(BackendDispatcher):
     """Dispatcher to dispatch sort and scan."""
 
-    calls_to_update: Dict[GlobalVar, Target]
+    calls_to_update: dict[GlobalVar, Target]
 
     def __init__(self, mod):
         super().__init__(mod)
diff --git a/python/tvm/relax/backend/gpu_generic/cumsum.py b/python/tvm/relax/backend/gpu_generic/cumsum.py
index 324a256bc5e8..e0e58ee7cdbd 100644
--- a/python/tvm/relax/backend/gpu_generic/cumsum.py
+++ b/python/tvm/relax/backend/gpu_generic/cumsum.py
@@ -18,7 +18,6 @@
 """Backend kernels for cumsum operator."""
 
 import math
-from typing import Optional
 
 from tvm.script import tir as T
 from tvm.tir import PrimFunc
@@ -34,7 +33,7 @@ def gpu_2d_continuous_cumsum(
     tx_len: int = 32,
     thread_elem: int = 4,
     in_dtype: str = "int32",
-    out_dtype: Optional[str] = None,
+    out_dtype: str | None = None,
 ) -> PrimFunc:
     """Generate GPU kernel for 2D continuous cumsum, i.e. The cumsum axis is -1
 
diff --git a/python/tvm/relax/backend/gpu_generic/sampling.py b/python/tvm/relax/backend/gpu_generic/sampling.py
index 4bd1ab82a8c4..6bcfc0c5e372 100644
--- a/python/tvm/relax/backend/gpu_generic/sampling.py
+++ b/python/tvm/relax/backend/gpu_generic/sampling.py
@@ -18,7 +18,7 @@
 """Backend kernels for sampling operator."""
 
 import math
-from typing import Callable, Optional
+from collections.abc import Callable
 
 import tvm
 from tvm.script import tir as T
@@ -165,7 +165,7 @@ def block_reduce_with_mask(
         output_local: T.Buffer,
         dtype: str,
         reduce_op: Callable,  # T.macro
-        mask_local: Optional[T.Buffer] = None,
+        mask_local: T.Buffer | None = None,
     ):
         with T.sblock():
             local_sum = T.alloc_buffer((), dtype, scope="local")
diff --git a/python/tvm/relax/backend/pattern_registry.py b/python/tvm/relax/backend/pattern_registry.py
index 5ec57164eb6b..73f695baf971 100644
--- a/python/tvm/relax/backend/pattern_registry.py
+++ b/python/tvm/relax/backend/pattern_registry.py
@@ -18,7 +18,7 @@
 """Pattern registry for BYOC backends"""
 
 import atexit
-from typing import Callable, List, Mapping, Optional, Set, Tuple, Union
+from collections.abc import Callable, Mapping
 
 from tvm.relax.dpl import DFPattern
 from tvm.relax.transform import FusionPattern
@@ -26,7 +26,7 @@
 from ..expr import Expr
 from . import _ffi_api
 
-_REGISTERED_PATTERN_NAMES: Set[str] = set()
+_REGISTERED_PATTERN_NAMES: set[str] = set()
 
 
 def _cleanup_registered_patterns():
@@ -52,15 +52,15 @@ def _ensure_cleanup_function_registered():
 
 
 CheckFunc = Callable[[Mapping[DFPattern, Expr], Expr], bool]
-Pattern = Union[
-    FusionPattern,
-    Tuple[str, DFPattern],
-    Tuple[str, DFPattern, Mapping[str, DFPattern]],
-    Tuple[str, DFPattern, Mapping[str, DFPattern], CheckFunc],
-]
+Pattern = (
+    FusionPattern
+    | tuple[str, DFPattern]
+    | tuple[str, DFPattern, Mapping[str, DFPattern]]
+    | tuple[str, DFPattern, Mapping[str, DFPattern], CheckFunc]
+)
 
 
-def register_patterns(patterns: List[Pattern]):
+def register_patterns(patterns: list[Pattern]):
     """
     Register patterns which will be used to partition the DataflowBlock into
     subgraphs that are supported by external backends.
@@ -85,7 +85,7 @@ def register_patterns(patterns: List[Pattern]):
     _ffi_api.RegisterPatterns(entries)
 
 
-def get_patterns_with_prefix(prefix: str) -> List[FusionPattern]:
+def get_patterns_with_prefix(prefix: str) -> list[FusionPattern]:
     """
     Get a list of patterns whose names startwith `prefix`.
 
@@ -102,7 +102,7 @@ def get_patterns_with_prefix(prefix: str) -> List[FusionPattern]:
     return _ffi_api.GetPatternsWithPrefix(prefix)
 
 
-def get_pattern(name: str) -> Optional[FusionPattern]:
+def get_pattern(name: str) -> FusionPattern | None:
     """
     Find the pattern with a particular name.
 
diff --git a/python/tvm/relax/backend/patterns.py b/python/tvm/relax/backend/patterns.py
index 4baea6d5c152..9d47bb8354d7 100644
--- a/python/tvm/relax/backend/patterns.py
+++ b/python/tvm/relax/backend/patterns.py
@@ -17,7 +17,7 @@
 # pylint: disable=invalid-name
 """Common patterns used in BYOC"""
 
-from typing import Dict, Mapping, Optional, Tuple, Union
+from collections.abc import Mapping
 
 from tvm.relax.dpl.pattern import (
     DFPattern,
@@ -34,10 +34,10 @@
 
 def _with_bias_activation_pattern(
     out: DFPattern,
-    annotations: Dict[str, DFPattern],
+    annotations: dict[str, DFPattern],
     with_bias: bool = False,
-    activation: Optional[str] = None,
-) -> Tuple[DFPattern, Mapping[str, DFPattern]]:
+    activation: str | None = None,
+) -> tuple[DFPattern, Mapping[str, DFPattern]]:
     if with_bias:
         annotations["bias"] = bias = wildcard()
         out = is_op("relax.add")(out, bias)
@@ -51,8 +51,8 @@ def _with_bias_activation_pattern(
 def make_fused_bias_activation_pattern(
     op_name: str,
     with_bias: bool = False,
-    activation: Optional[str] = None,
-) -> Tuple[DFPattern, Mapping[str, DFPattern]]:
+    activation: str | None = None,
+) -> tuple[DFPattern, Mapping[str, DFPattern]]:
     """
     A simple utility to create patterns for an operation fused with bias addition and activation.
 
@@ -86,10 +86,10 @@ def make_fused_bias_activation_pattern(
 
 
 def make_residual_block_pattern(
-    node_output: Union[DFPattern, Tuple[DFPattern, Mapping[str, DFPattern]]],
+    node_output: DFPattern | tuple[DFPattern, Mapping[str, DFPattern]],
     binary_op="relax.add",
     activation=None,
-) -> Tuple[DFPattern, Mapping[str, DFPattern]]:
+) -> tuple[DFPattern, Mapping[str, DFPattern]]:
     """
     Create pattern for residual block.
 
@@ -133,8 +133,8 @@ def make_residual_block_pattern(
 
 def make_conv2d_pattern(
     with_bias: bool = False,
-    activation: Optional[str] = None,
-) -> Tuple[DFPattern, Mapping[str, DFPattern]]:
+    activation: str | None = None,
+) -> tuple[DFPattern, Mapping[str, DFPattern]]:
     """
     Create pattern for 2D convolution.
 
@@ -169,9 +169,9 @@ def make_conv2d_pattern(
 
 def make_matmul_pattern(
     with_bias: bool = False,
-    activation: Optional[str] = None,
+    activation: str | None = None,
     transposed_rhs: bool = False,
-) -> Tuple[DFPattern, Mapping[str, DFPattern]]:
+) -> tuple[DFPattern, Mapping[str, DFPattern]]:
     """
     Create pattern for matrix multiplication.
 
@@ -358,7 +358,7 @@ def make_rms_norm_pattern():
 
 def make_matmul_dequantize_pattern(
     transposed_rhs: bool = False,
-) -> Tuple[DFPattern, Mapping[str, DFPattern]]:
+) -> tuple[DFPattern, Mapping[str, DFPattern]]:
     """
     Create pattern for matrix multiplication and dequantize operation.
 
@@ -398,7 +398,7 @@ def make_matmul_dequantize_pattern(
 
 def make_matmul_multiply_pattern(
     transposed_rhs: bool = False,
-) -> Tuple[DFPattern, Mapping[str, DFPattern]]:
+) -> tuple[DFPattern, Mapping[str, DFPattern]]:
     """
     Create pattern for matrix multiplication and multiply operation.
 
diff --git a/python/tvm/relax/backend/rocm/hipblas.py b/python/tvm/relax/backend/rocm/hipblas.py
index 0bc7038e1508..1c43777e23fd 100644
--- a/python/tvm/relax/backend/rocm/hipblas.py
+++ b/python/tvm/relax/backend/rocm/hipblas.py
@@ -56,7 +56,7 @@ def _check_matmul(context: PatternCheckContext) -> bool:
     lhs_shape = lhs.struct_info.shape.values
     rhs_shape = rhs.struct_info.shape.values
 
-    if not isinstance(lhs_shape[-1], (tvm.tir.expr.IntImm, int)):
+    if not isinstance(lhs_shape[-1], tvm.tir.expr.IntImm | int):
         # Reduction axis must be constant
         return False
 
@@ -75,7 +75,7 @@ def _check_matmul(context: PatternCheckContext) -> bool:
         bias = context.annotated_expr["bias"]
         bias_shape = bias.struct_info.shape.values
         bias_batches = reduce(operator.mul, bias_shape[:-1], 1)
-        if not isinstance(bias_batches, (tvm.tir.expr.IntImm, int)) or int(bias_batches) > 1:
+        if not isinstance(bias_batches, tvm.tir.expr.IntImm | int) or int(bias_batches) > 1:
             # hipblas only supports bias vector
             return False
 
diff --git a/python/tvm/relax/backend/utils.py b/python/tvm/relax/backend/utils.py
index 87e866cb031a..9d4598e669b8 100644
--- a/python/tvm/relax/backend/utils.py
+++ b/python/tvm/relax/backend/utils.py
@@ -17,8 +17,6 @@
 # pylint: disable=invalid-name
 """Utils for BYOC pattern matching"""
 
-from typing import Tuple
-
 from tvm import relax
 from tvm.relax import DataflowVar, PyExprMutator
 from tvm.relax.transform import PatternCheckContext
@@ -37,7 +35,7 @@ def is_gpu_target(target: Target) -> bool:
         return "gpu" in target.keys
 
     @staticmethod
-    def get_shape_dtype(expr: relax.Expr) -> Tuple[relax.ShapeExpr, str]:
+    def get_shape_dtype(expr: relax.Expr) -> tuple[relax.ShapeExpr, str]:
         """Get shape and dtype from an expression.
         If the shape and dtype is unknown, raise an error."""
         sinfo = expr.struct_info
diff --git a/python/tvm/relax/base_py_module.py b/python/tvm/relax/base_py_module.py
index e38ee4d4b966..016bd27034d1 100644
--- a/python/tvm/relax/base_py_module.py
+++ b/python/tvm/relax/base_py_module.py
@@ -19,7 +19,7 @@
 
 import inspect
 import os
-from typing import Any, Dict, List, Optional, Tuple, Union
+from typing import Any, Optional, Union
 
 import numpy as np
 
@@ -66,7 +66,7 @@ def __init__(
         self,
         ir_mod: IRModule,
         device: Device,
-        target: Optional[Target] = None,
+        target: Target | None = None,
     ):
         """Initialize BasePyModule with JIT compilation and DLPack conversion."""
         self.device = device
@@ -100,12 +100,12 @@ def _getattr_python_function(name: str) -> Any:
 
         self.__getattr__ = _getattr_python_function
 
-        self.compiled_tir_funcs: Dict[str, PackedFunc] = {}
-        self.extern_funcs: Dict[str, PackedFunc] = {}
-        self.tir_func_names: List[str] = []
-        self.relax_func_names: List[str] = []
-        self.relax_vm: Optional[relax.VirtualMachine] = None
-        self.pyfuncs: Dict[str, Any] = {}
+        self.compiled_tir_funcs: dict[str, PackedFunc] = {}
+        self.extern_funcs: dict[str, PackedFunc] = {}
+        self.tir_func_names: list[str] = []
+        self.relax_func_names: list[str] = []
+        self.relax_vm: relax.VirtualMachine | None = None
+        self.pyfuncs: dict[str, Any] = {}
 
         if target is None:
             target = Target.from_device(device)
@@ -294,8 +294,8 @@ def _create_output_tensors(self, out_sinfo, in_args=None):
         sinfo_list = out_sinfo if isinstance(out_sinfo, list) else [out_sinfo]
         out_tensors = []
         for sinfo in sinfo_list:
-            if isinstance(sinfo, (tuple, list)) and all(
-                isinstance(x, (int, np.integer)) for x in sinfo
+            if isinstance(sinfo, tuple | list) and all(
+                isinstance(x, int | np.integer) for x in sinfo
             ):
                 out_tensors.append(torch.zeros(list(map(int, sinfo)), dtype=torch.float32))
                 continue
@@ -313,7 +313,7 @@ def _infer_concrete_shape_from_args(self, shape, in_args):
         concrete = []
         symbolic_positions = []
         for idx, dim in enumerate(shape):
-            if isinstance(dim, (int, np.integer)):
+            if isinstance(dim, int | np.integer):
                 concrete.append(int(dim))
             elif isinstance(dim, tir.IntImm):
                 concrete.append(int(dim.value))
@@ -326,10 +326,10 @@ def _infer_concrete_shape_from_args(self, shape, in_args):
 
         candidates = []
         if in_args is not None:
-            if not isinstance(in_args, (list, tuple)):
+            if not isinstance(in_args, list | tuple):
                 in_args = [in_args]
             for obj in in_args:
-                if hasattr(obj, "shape") and isinstance(obj.shape, (tuple, list)):
+                if hasattr(obj, "shape") and isinstance(obj.shape, tuple | list):
                     try:
                         candidates.append(tuple(int(x) for x in obj.shape))
                         continue
@@ -366,13 +366,13 @@ def _convert_tvm_dtype_to_torch(self, tvm_dtype: str) -> "torch.dtype":
         return dtype_mapping.get(str(tvm_dtype), torch.float32)
 
     def _convert_pytorch_to_tvm(
-        self, tensors: Union[Any, List[Any], Tuple[Any, ...]]
-    ) -> Union[Tensor, List[Tensor]]:
+        self, tensors: Any | list[Any] | tuple[Any, ...]
+    ) -> Tensor | list[Tensor]:
         """Convert PyTorch tensors to TVM Tensors using DLPack."""
         # pylint: disable=import-outside-toplevel
         import torch
 
-        if isinstance(tensors, (list, tuple)):
+        if isinstance(tensors, list | tuple):
             return [self._convert_single_pytorch_to_tvm(t) for t in tensors]
         return self._convert_single_pytorch_to_tvm(tensors)
 
@@ -424,10 +424,10 @@ def _convert_single_pytorch_to_tvm(self, tensor: Any) -> Tensor:
             ) from error
 
     def _convert_tvm_to_pytorch(
-        self, tvm_tensors: Union[Any, List[Any]]
-    ) -> Union["torch.Tensor", List["torch.Tensor"]]:
+        self, tvm_tensors: Any | list[Any]
+    ) -> Union["torch.Tensor", list["torch.Tensor"]]:
         """Convert TVM Tensors to PyTorch tensors using DLPack."""
-        if isinstance(tvm_tensors, (list, tuple)):
+        if isinstance(tvm_tensors, list | tuple):
             return [self._convert_single_tvm_to_pytorch(tensor) for tensor in tvm_tensors]
         return self._convert_single_tvm_to_pytorch(tvm_tensors)
 
@@ -457,7 +457,7 @@ def _convert_single_tvm_to_pytorch(self, tvm_tensor: Any) -> "torch.Tensor":
             numpy_array = tvm_tensor.numpy()
             return torch.from_numpy(numpy_array)
 
-    def get_function(self, name: str) -> Optional[PackedFunc]:
+    def get_function(self, name: str) -> PackedFunc | None:
         """Get a compiled function by name."""
         if name in self.compiled_tir_funcs:
             return self.compiled_tir_funcs[name]
@@ -472,7 +472,7 @@ def get_function(self, name: str) -> Optional[PackedFunc]:
                 print(f"Warning: Failed to get Relax function '{name}': {error}")
         return None
 
-    def list_functions(self) -> Dict[str, List[str]]:
+    def list_functions(self) -> dict[str, list[str]]:
         """List all available functions."""
         return {
             "tir": self.tir_func_names,
@@ -504,7 +504,7 @@ def wrapper(*args, **kwargs):
     def script(
         self,
         *,
-        name: Optional[str] = None,
+        name: str | None = None,
         show_meta: bool = False,
         ir_prefix: str = "I",
         tir_prefix: str = "T",
@@ -585,7 +585,7 @@ def _insert_python_functions(self, base_script: str, indent_spaces: int) -> str:
 
         return "\n".join(result_lines)
 
-    def _get_function_source(self, func: callable) -> Optional[str]:
+    def _get_function_source(self, func: callable) -> str | None:
         """Get the source code of a Python function."""
         try:
             source = inspect.getsource(func)
@@ -610,9 +610,7 @@ def _format_python_function(self, _func_name: str, func_source: str, indent: int
 
         return "\n".join(formatted_lines)
 
-    def show(
-        self, style: Optional[str] = None, black_format: Optional[bool] = None, **kwargs
-    ) -> None:
+    def show(self, style: str | None = None, black_format: bool | None = None, **kwargs) -> None:
         """A sugar for print highlighted TVM script with Python function support.
 
         This method extends the standard IRModule show() method to handle
diff --git a/python/tvm/relax/binding_rewrite.py b/python/tvm/relax/binding_rewrite.py
index 1fd2896e9814..01287893689e 100644
--- a/python/tvm/relax/binding_rewrite.py
+++ b/python/tvm/relax/binding_rewrite.py
@@ -17,8 +17,6 @@
 # pylint: disable=no-else-return, invalid-name
 """Developer API of add/remove/replace bindings in Relax."""
 
-from typing import Optional
-
 import tvm_ffi
 
 import tvm
@@ -74,7 +72,7 @@ def replace_all_uses(self, old_var: Var, new_var: Var) -> None:
     def add_binding(self, binding: Binding) -> None:
         return _ffi_api.dfb_rewrite_add_binding(self, binding)  # type: ignore
 
-    def add(self, expr: Expr, name: Optional[str] = None, is_dfvar: bool = False) -> None:
+    def add(self, expr: Expr, name: str | None = None, is_dfvar: bool = False) -> None:
         """
         Add a new statement to the DataflowBlock with an automatically generated variable name.
 
diff --git a/python/tvm/relax/block_builder.py b/python/tvm/relax/block_builder.py
index c0673e0fbf2a..b10ddf5ebd02 100644
--- a/python/tvm/relax/block_builder.py
+++ b/python/tvm/relax/block_builder.py
@@ -18,7 +18,8 @@
 # ruff: noqa: RUF012
 """Developer API of constructing Relax AST."""
 
-from typing import Any, Callable, Dict, List, Optional, Sequence, Union
+from collections.abc import Callable, Sequence
+from typing import Any, Optional
 
 import tvm_ffi
 
@@ -164,7 +165,7 @@ def current() -> Optional["BlockBuilder"]:
 
     def __init__(self, mod: IRModule = None):
         # Which functions are currently being defined
-        self._func_stack: List[FunctionScope] = []
+        self._func_stack: list[FunctionScope] = []
         self.__init_handle_by_constructor__(_ffi_api.BlockBuilderCreate, mod)  # type: ignore
 
     def _begin_dataflow_block(self) -> None:
@@ -210,8 +211,8 @@ def _exit_function_scope(self, exc_type, exc_val, exc_tb):
     def function(
         self,
         name: str,
-        params: Optional[Union[Var, Tuple, List[Var]]] = None,
-        attrs: Optional[Dict[str, Object]] = None,
+        params: Var | Tuple | list[Var] | None = None,
+        attrs: dict[str, Object] | None = None,
         pure: bool = True,
         private: bool = False,
     ) -> FunctionScope:
@@ -246,7 +247,7 @@ def function(
         """
         if isinstance(params, rx.Var):
             params = [params]
-        elif isinstance(params, (list, tuple)):
+        elif isinstance(params, list | tuple):
             for param in params:
                 if not isinstance(param, rx.Var):
                     raise TypeError(
@@ -262,7 +263,7 @@ def function(
 
         return FunctionScope(self, name, params, attrs, is_pure=pure)
 
-    def testing_scope(self, def_vars: List[tir.Var]) -> TestingScope:
+    def testing_scope(self, def_vars: list[tir.Var]) -> TestingScope:
         """Start a scope for unit-testing purposes.
 
         Parameters
@@ -287,7 +288,7 @@ def dataflow(self) -> DataflowScope:
         """
         return DataflowScope(self)
 
-    def _normalize_python_tuple(self, expr: Union[Expr, Sequence[Expr]]):
+    def _normalize_python_tuple(self, expr: Expr | Sequence[Expr]):
         """Internal utility function to convert to relax.Tuple
 
         The `emit`, `emit_output`, and `emit_func_output` can be
@@ -297,7 +298,7 @@ def _normalize_python_tuple(self, expr: Union[Expr, Sequence[Expr]]):
         `tvm.runtime.Array`.  In addition, any nested tuple objects
         should be converted.
         """
-        if isinstance(expr, (list, tuple)):
+        if isinstance(expr, list | tuple):
             return Tuple([self._normalize_python_tuple(element) for element in expr])
         elif expr is None:
             from . import op
@@ -371,7 +372,7 @@ def call_te_with_grad(
         func: Callable,
         *args: Any,
         te_grad_name: str,
-        te_grad_kwargs: Optional[Dict[str, Object]] = None,
+        te_grad_kwargs: dict[str, Object] | None = None,
         **kwargs: Any,
     ) -> Expr:
         """Generate a call node according to the te function.
@@ -570,7 +571,7 @@ def match_cast(self, value: Expr, struct_info: StructInfo, name_hint: str = "")
             name_hint,
         )  # type: ignore
 
-    def emit_output(self, output: Union[Expr, Tuple, List[Expr]], name_hint: str = "") -> Var:
+    def emit_output(self, output: Expr | Tuple | list[Expr], name_hint: str = "") -> Var:
         """Emit output for the current dataflow block or function.
 
         Parameters
@@ -591,8 +592,8 @@ def emit_output(self, output: Union[Expr, Tuple, List[Expr]], name_hint: str = "
 
     def emit_func_output(
         self,
-        output: Union[Expr, Tuple, List[Expr]],
-        params: Optional[Union[Var, Tuple, List[Var]]] = None,
+        output: Expr | Tuple | list[Expr],
+        params: Var | Tuple | list[Var] | None = None,
     ) -> GlobalVar:
         """Emit output for the function.
 
@@ -765,7 +766,7 @@ def emit_normalized(self, binding: Binding) -> None:
         """
         _ffi_api.BlockBuilderEmitNormalized(self, binding)  # type: ignore
 
-    def lookup_binding(self, var: Var) -> Optional[Expr]:
+    def lookup_binding(self, var: Var) -> Expr | None:
         """Lookup a var in the binding table.
 
         Parameters
@@ -780,7 +781,7 @@ def lookup_binding(self, var: Var) -> Optional[Expr]:
         """
         return _ffi_api.BlockBuilderLookupBinding(self, var)  # type: ignore
 
-    def begin_scope(self, params: Optional[List[Var]] = None) -> None:
+    def begin_scope(self, params: list[Var] | None = None) -> None:
         """Begin a new scope, with optional parameters that
         are visible within the scope.
 
diff --git a/python/tvm/relax/distributed/global_info.py b/python/tvm/relax/distributed/global_info.py
index e648d02399eb..576ab0c70eb7 100644
--- a/python/tvm/relax/distributed/global_info.py
+++ b/python/tvm/relax/distributed/global_info.py
@@ -17,8 +17,6 @@
 # pylint: disable=redefined-builtin, invalid-name
 """Global Info Data structures for distributed tensor."""
 
-from typing import List, Tuple, Union
-
 import tvm_ffi
 
 from tvm.ir import Range
@@ -41,10 +39,8 @@ class DeviceMesh(GlobalInfo):
         Represents the device id in the mesh
     """
 
-    def __init__(
-        self, shape: Union[ShapeTuple, List[int], Tuple[int]], device_ids: Union[List[int], Range]
-    ):
-        if isinstance(shape, (list, tuple)):
+    def __init__(self, shape: ShapeTuple | list[int] | tuple[int], device_ids: list[int] | Range):
+        if isinstance(shape, list | tuple):
             shape = ShapeTuple(shape)
         device_range = None
         if isinstance(device_ids, Range):
@@ -53,7 +49,7 @@ def __init__(
         self.__init_handle_by_constructor__(ffi.DeviceMesh, shape, device_ids, device_range)  # type: ignore
 
 
-def device_mesh(shape: ShapeTuple, device_ids: Union[List[int], Range]) -> DeviceMesh:
+def device_mesh(shape: ShapeTuple, device_ids: list[int] | Range) -> DeviceMesh:
     """Create a device mesh expression.
     Parameters
     ----------
diff --git a/python/tvm/relax/distributed/struct_info.py b/python/tvm/relax/distributed/struct_info.py
index e8b457af0bc3..2d85b2fa6d14 100644
--- a/python/tvm/relax/distributed/struct_info.py
+++ b/python/tvm/relax/distributed/struct_info.py
@@ -18,7 +18,6 @@
 """Struct Info for distributed tensor."""
 
 import enum
-from typing import List
 
 import tvm_ffi
 
@@ -93,7 +92,7 @@ class Placement(Object):
         The placement spec for each dimension of the device mesh.
     """
 
-    def __init__(self, dim_specs: List[PlacementSpec]):
+    def __init__(self, dim_specs: list[PlacementSpec]):
         self.__init_handle_by_constructor__(_ffi_api.Placement, dim_specs)  # type: ignore
 
     @staticmethod
diff --git a/python/tvm/relax/dpl/context.py b/python/tvm/relax/dpl/context.py
index b019ffba8ee6..adc955ebbcf0 100644
--- a/python/tvm/relax/dpl/context.py
+++ b/python/tvm/relax/dpl/context.py
@@ -17,8 +17,6 @@
 
 """The Graph Matching Context Manager for Dataflow Pattern Language."""
 
-from typing import Dict
-
 import tvm
 
 from ..expr import DataflowBlock, Var
@@ -64,7 +62,7 @@ def current() -> "PatternContext":
     def match_dfb(
         self,
         dfb: DataflowBlock,
-    ) -> Dict[DFPattern, Var]:
+    ) -> dict[DFPattern, Var]:
         """
         Match a DataflowBlock via a graph of DFPattern and corresponding constraints
 
diff --git a/python/tvm/relax/dpl/pattern.py b/python/tvm/relax/dpl/pattern.py
index 179246d8d234..ad7d1b7421b1 100644
--- a/python/tvm/relax/dpl/pattern.py
+++ b/python/tvm/relax/dpl/pattern.py
@@ -20,8 +20,7 @@
 # pylint: disable=no-member
 # pylint: disable=pointless-statement
 
-import typing
-from typing import Dict, List, Optional, Tuple, Union
+from typing import Union
 
 import tvm_ffi
 
@@ -108,7 +107,7 @@ def __invert__(self) -> "NotPattern":
         """
         return reject(self)
 
-    def has_attr(self, attrs: Dict[str, Object]) -> "AttrPattern":
+    def has_attr(self, attrs: dict[str, Object]) -> "AttrPattern":
         """
         Add an attribute constraint to this pattern
 
@@ -143,7 +142,7 @@ def has_dtype(self, dtype: str) -> "DataTypePattern":
         """
         return has_dtype(dtype, self)
 
-    def has_shape(self, shape: List[PrimExpr]) -> "ShapePattern":
+    def has_shape(self, shape: list[PrimExpr]) -> "ShapePattern":
         """
         Add a shape constraint to this pattern
 
@@ -162,11 +161,11 @@ def has_shape(self, shape: List[PrimExpr]) -> "ShapePattern":
         has_shape assumes that the matched relax.Expr only has one
         output tensor. Use is_tuple for those with multiple outputs.
         """
-        if not isinstance(shape, (list, tuple, tvm.ir.PrimExpr)):
+        if not isinstance(shape, list | tuple | tvm.ir.PrimExpr):
             raise ValueError("has_shape takes a list or tuple as input.")
         return ShapePattern(pattern=self, shape=shape)
 
-    def match(self, expr, var2val: Optional[Dict[Var, Expr]] = None) -> bool:
+    def match(self, expr, var2val: dict[Var, Expr] | None = None) -> bool:
         """
         Match a relax.Expr syntactically
 
@@ -193,8 +192,8 @@ def match(self, expr, var2val: Optional[Dict[Var, Expr]] = None) -> bool:
         return ffi.match_expr(self, expr, var2val)  # type: ignore
 
     def extract_matched_expr(
-        self, expr, var2val: Optional[Dict[Var, Expr]] = None
-    ) -> Optional[Dict["DFPattern", Expr]]:
+        self, expr, var2val: dict[Var, Expr] | None = None
+    ) -> dict["DFPattern", Expr] | None:
         """
         Match a relax.Expr and return a map from matching patterns to matched expressions.
 
@@ -277,7 +276,7 @@ def fork_to(self, *args) -> None:
         for v in args:
             self ^ v
 
-    def same_shape_as(self, *args: List["DFPattern"]) -> "SameShapeConstraint":
+    def same_shape_as(self, *args: list["DFPattern"]) -> "SameShapeConstraint":
         """
         The current pattern with the same shape as another pattern (sequence)
 
@@ -409,7 +408,7 @@ class CallPattern(DFPattern):
     def __init__(
         self,
         op: "DFPattern",
-        args: Union[List["DFPattern"], typing.Tuple["DFPattern", ...]],
+        args: list["DFPattern"] | tuple["DFPattern", ...],
         varg_default_wildcard: bool = False,
         add_constraint=True,
     ):
@@ -441,7 +440,7 @@ class FunctionPattern(DFPattern):
 
     def __init__(
         self,
-        params: List["DFPattern"],
+        params: list["DFPattern"],
         body: "DFPattern",
     ):
         self.__init_handle_by_constructor__(ffi.FunctionPattern, params, body)  # type: ignore
@@ -460,7 +459,7 @@ class TuplePattern(DFPattern):
     def __init__(self, fields: list):
         self.__init_handle_by_constructor__(ffi.TuplePattern, fields)  # type: ignore
 
-    def __getitem__(self, index: Optional[int]) -> "TupleGetItemPattern":
+    def __getitem__(self, index: int | None) -> "TupleGetItemPattern":
         if index is not None:
             # support negative index for being pythonic
             if index < 0:
@@ -505,7 +504,7 @@ class TupleGetItemPattern(DFPattern):
         The index to match; Default (None) to match a TupleGetItem with any index.
     """
 
-    def __init__(self, tuple_value: "DFPattern", index: Optional[int] = None):
+    def __init__(self, tuple_value: "DFPattern", index: int | None = None):
         match_index = index if index is not None else -1
         self.__init_handle_by_constructor__(
             ffi.TupleGetItemPattern,
@@ -619,7 +618,7 @@ class ShapePattern(DFPattern):
         The shape to match.
     """
 
-    def __init__(self, pattern: "DFPattern", shape: List[tvm.ir.PrimExpr]):
+    def __init__(self, pattern: "DFPattern", shape: list[tvm.ir.PrimExpr]):
         self.__init_handle_by_constructor__(ffi.ShapePattern, pattern, shape)  # type: ignore
 
 
@@ -633,7 +632,7 @@ class SameShapeConstraint(DFConstraint):
         A set of patterns which must all provide the same shape.
     """
 
-    def __init__(self, *args: List[DFPattern]):
+    def __init__(self, *args: list[DFPattern]):
         self.__init_handle_by_constructor__(ffi.SameShapeConstraint, args)  # type: ignore
 
 
@@ -648,7 +647,7 @@ class PrimArrPattern(DFPattern):
         The shape to match.
     """
 
-    def __init__(self, shape: List[tvm.ir.PrimExpr]):
+    def __init__(self, shape: list[tvm.ir.PrimExpr]):
         self.__init_handle_by_constructor__(ffi.PrimArrPattern, shape)  # type: ignore
 
     def __getitem__(self, index: int):
@@ -757,9 +756,7 @@ def is_op(op_name: str) -> ExprPattern:
     return ExprPattern(op)
 
 
-def is_tuple(
-    fields: Union[Array, List, Tuple], unordered=False
-) -> Union[TuplePattern, UnorderedTuplePattern]:
+def is_tuple(fields: Array | list | tuple, unordered=False) -> TuplePattern | UnorderedTuplePattern:
     """
     Syntatic sugar for creating an ExprPattern.
 
@@ -773,14 +770,14 @@ def is_tuple(
     result: tvm.relax.dpl.DFPattern
         The resulting pattern.
     """
-    if not isinstance(fields, (list, tuple, Array)):
+    if not isinstance(fields, list | tuple | Array):
         raise ValueError("fields must be a list, tuple, or Array")
     if unordered:
         return UnorderedTuplePattern(fields)
     return TuplePattern(fields)
 
 
-def is_tuple_get_item(tuple_value: DFPattern, index: Optional[int] = None) -> TupleGetItemPattern:
+def is_tuple_get_item(tuple_value: DFPattern, index: int | None = None) -> TupleGetItemPattern:
     """
     Syntatic sugar for creating an ExprPattern.
 
@@ -834,7 +831,7 @@ def has_dtype(dtype: str, pattern: DFPattern = None) -> DataTypePattern:
     return DataTypePattern(pattern, dtype)
 
 
-def is_shape(shape: List[tvm.ir.PrimExpr]) -> "PrimArrPattern":
+def is_shape(shape: list[tvm.ir.PrimExpr]) -> "PrimArrPattern":
     """
     Directly matches a shape which is an array of PrimExpr
 
@@ -859,7 +856,7 @@ def is_shape(shape: List[tvm.ir.PrimExpr]) -> "PrimArrPattern":
     puts assumptions on the shape of the tensor matched by pattern p. While
     is_shape directly matches the shape (an array of PrimExpr).
     """
-    if not isinstance(shape, (list, tuple, tvm.ir.Array)):
+    if not isinstance(shape, list | tuple | tvm.ir.Array):
         raise ValueError("is_shape takes a list or tuple as input.")
     return PrimArrPattern(shape)
 
@@ -867,12 +864,12 @@ def is_shape(shape: List[tvm.ir.PrimExpr]) -> "PrimArrPattern":
 # Todo(relax-team): Dataflow pattern for StructInfo, and match out_sinfo
 def _is_call_tir(
     func_pattern: DFPattern,
-    args: Union[List, Tuple, TuplePattern] = None,
-    tir_vars: Optional[DFPattern] = None,
+    args: list | tuple | TuplePattern = None,
+    tir_vars: DFPattern | None = None,
 ) -> CallPattern:
     if args is None:
         args = wildcard()
-    elif isinstance(args, (list, tuple)):
+    elif isinstance(args, list | tuple):
         args = TuplePattern(args)
 
     if tir_vars is None:
@@ -883,8 +880,8 @@ def _is_call_tir(
 # Todo(relax-team): Dataflow pattern for StructInfo, and match out_sinfo
 def is_call_tir(
     func_name: str,
-    args: Union[List, Tuple, TuplePattern] = None,
-    tir_vars: Optional[DFPattern] = None,
+    args: list | tuple | TuplePattern = None,
+    tir_vars: DFPattern | None = None,
 ) -> CallPattern:
     """
     Syntax sugar for creating a CallPattern for call_tir that calls an function through global var.
@@ -908,11 +905,11 @@ def is_call_tir(
 
 def _is_call_dps_packed(
     func_pattern: DFPattern,
-    args: Union[List, Tuple, TuplePattern] = None,
+    args: list | tuple | TuplePattern = None,
 ) -> CallPattern:
     if args is None:
         args = wildcard()
-    elif isinstance(args, (list, tuple)):
+    elif isinstance(args, list | tuple):
         args = TuplePattern(args)
 
     return is_op("relax.call_dps_packed")(func_pattern, args, add_constraint=False)
@@ -920,7 +917,7 @@ def _is_call_dps_packed(
 
 def is_call_dps_packed(
     func_name: str,
-    args: Union[List, Tuple, TuplePattern] = None,
+    args: list | tuple | TuplePattern = None,
 ) -> CallPattern:
     """Syntax sugar for creating a CallPattern for call_dps_packed
 
@@ -941,7 +938,7 @@ def is_call_dps_packed(
 
 
 def is_call_packed(
-    func_name: str, args: Optional[Union[List[DFPattern], Tuple[DFPattern]]] = None
+    func_name: str, args: list[DFPattern] | tuple[DFPattern] | None = None
 ) -> CallPattern:
     """
     Syntax sugar for creating a CallPattern for call_packed
@@ -1006,7 +1003,7 @@ def has_attr(attrs, pattern=None) -> AttrPattern:
 class PatternSeq(Node):
     """A sequence of patterns with consecutive constraints"""
 
-    def __init__(self, patterns: List[DFPattern], only_use=False):
+    def __init__(self, patterns: list[DFPattern], only_use=False):
         """
         Initializer to PatternSeq
 
@@ -1105,8 +1102,8 @@ def dup(self) -> "PatternSeq":
 
 
 def _used_by(
-    lhs: Union[DFPattern, PatternSeq],
-    rhs: Union[DFPattern, PatternSeq],
+    lhs: DFPattern | PatternSeq,
+    rhs: DFPattern | PatternSeq,
     index=-1,
 ) -> PatternSeq:
     if isinstance(lhs, DFPattern):
@@ -1116,9 +1113,7 @@ def _used_by(
     return ffi.used_by(lhs, rhs, index)  # type: ignore
 
 
-def _only_used_by(
-    lhs: Union[DFPattern, PatternSeq], rhs: Union[DFPattern, PatternSeq], index=-1
-) -> PatternSeq:
+def _only_used_by(lhs: DFPattern | PatternSeq, rhs: DFPattern | PatternSeq, index=-1) -> PatternSeq:
     if isinstance(lhs, DFPattern):
         lhs = PatternSeq([lhs])
     if isinstance(rhs, DFPattern):
diff --git a/python/tvm/relax/dpl/rewrite.py b/python/tvm/relax/dpl/rewrite.py
index 647455d7c764..5cc928ff2599 100644
--- a/python/tvm/relax/dpl/rewrite.py
+++ b/python/tvm/relax/dpl/rewrite.py
@@ -16,7 +16,7 @@
 # under the License.
 """APIs for pattern-based rewriting."""
 
-from typing import Callable, Dict, Union
+from collections.abc import Callable
 
 from tvm_ffi import register_object
 
@@ -36,7 +36,7 @@ class PatternMatchingRewriter(Object):
     @staticmethod
     def from_pattern(
         pattern: DFPattern,
-        func: Callable[[Expr, Dict[DFPattern, Expr]], Expr],
+        func: Callable[[Expr, dict[DFPattern, Expr]], Expr],
     ) -> "PatternMatchingRewriter":
         """Construct from a pattern and rewriter-function
 
@@ -131,7 +131,7 @@ def replacement(A: R.Tensor):
         """
         return ffi.PatternMatchingRewriterFromModule(mod)  # type: ignore
 
-    def __call__(self, obj: Union[Expr, IRModule]) -> Union[Expr, IRModule]:
+    def __call__(self, obj: Expr | IRModule) -> Expr | IRModule:
         """Apply the rewriter
 
         Parameters
@@ -204,7 +204,7 @@ def __init__(self, patterns, func):
 
 def rewrite_call(
     pattern: DFPattern,
-    rewriter: Callable[[Expr, Dict[DFPattern, Expr]], Expr],
+    rewriter: Callable[[Expr, dict[DFPattern, Expr]], Expr],
     func: Function,
 ) -> Function:
     """
@@ -242,7 +242,7 @@ def rewriter(orig, matchings):
 
 def rewrite_bindings(
     ctx: PatternContext,
-    rewriter: Callable[[Dict[DFPattern, Var], Dict[Var, Expr]], Dict[Var, Expr]],
+    rewriter: Callable[[dict[DFPattern, Var], dict[Var, Expr]], dict[Var, Expr]],
     func: Function,
 ) -> Function:
     """
diff --git a/python/tvm/relax/exec_builder.py b/python/tvm/relax/exec_builder.py
index e530b7e7c63d..65ca991071a3 100644
--- a/python/tvm/relax/exec_builder.py
+++ b/python/tvm/relax/exec_builder.py
@@ -19,7 +19,6 @@
 """A builder to build Relax VM executable."""
 
 from enum import IntEnum
-from typing import List, Optional, Union
 
 import tvm_ffi
 
@@ -47,7 +46,7 @@ class VMFuncKind(IntEnum):
 class VMFuncScope:
     """An object corresponds to each VM function, working as a context manager."""
 
-    stack: List["VMFuncScope"] = []
+    stack: list["VMFuncScope"] = []
 
     def __init__(self, exit_callback):
         self.exit_callback = exit_callback
@@ -95,7 +94,7 @@ def declare_function(self, func_name: str, kind: VMFuncKind = VMFuncKind.PACKED_
         _ffi_api.ExecBuilderDeclareFunction(self, func_name, kind)  # type: ignore
 
     def function(
-        self, func_name: str, num_inputs: Optional[int] = 0, param_names: Optional[List[str]] = None
+        self, func_name: str, num_inputs: int | None = 0, param_names: list[str] | None = None
     ) -> VMFuncScope:
         """annotate a VM function."""
         _ffi_api.ExecBuilderEmitFunction(self, func_name, num_inputs, param_names)  # type: ignore
@@ -111,8 +110,8 @@ def convert_constant(self, const: object) -> int:
     def emit_call(
         self,
         name: str,
-        args: Optional[List[Union[tvm.runtime.Tensor, tvm.DataType]]] = None,
-        dst: Optional[int] = None,
+        args: list[tvm.runtime.Tensor | tvm.DataType] | None = None,
+        dst: int | None = None,
     ) -> None:
         """emit a call instruction which calls a packed function."""
         self._check_scope()
@@ -125,7 +124,7 @@ def emit_call(
                     shape_tuple = ShapeTuple(arg)
                     new_arg = self.convert_constant(shape_tuple)
                     args_.append(new_arg)
-                elif isinstance(arg, (tvm.runtime.Tensor, tvm.DataType, ShapeTuple)):
+                elif isinstance(arg, tvm.runtime.Tensor | tvm.DataType | ShapeTuple):
                     new_arg = self.convert_constant(arg)
                     args_.append(new_arg)
                 else:
diff --git a/python/tvm/relax/expr.py b/python/tvm/relax/expr.py
index 5b9581818a9a..c96aaf8648c5 100644
--- a/python/tvm/relax/expr.py
+++ b/python/tvm/relax/expr.py
@@ -18,8 +18,9 @@
 """The expression nodes of Relax."""
 
 import typing
+from collections.abc import Callable, Mapping
 from numbers import Number
-from typing import Any, Callable, Dict, List, Mapping, Optional, Union
+from typing import Any, Optional, Union
 
 import numpy as _np  # type: ignore
 import tvm_ffi
@@ -38,9 +39,9 @@
 # It is a workaround for mypy: https://github.com/python/mypy/issues/7866#issuecomment-549454370
 # This feature is not supported until python 3.10:
 # https://docs.python.org/3.10/whatsnew/3.10.html#pep-613-typealias
-Expr = Union[tvm.ir.RelaxExpr]
-Type = Union[tvm.ir.Type]  # pylint: disable=invalid-name
-GlobalVar = Union[tvm.ir.GlobalVar]
+Expr = tvm.ir.RelaxExpr
+Type = tvm.ir.Type  # pylint: disable=invalid-name
+GlobalVar = tvm.ir.GlobalVar
 
 
 @tvm_ffi.register_object("relax.Id")
@@ -116,7 +117,7 @@ def _binary_rhs_helper(rhs: "ExprWithOp") -> "ExprWithOp":
 class ExprWithOp(Expr, Scriptable):
     """Basetype of all relax expressions that defines op overloading."""
 
-    def astype(self, dtype: Union[str, DataType]) -> "ExprWithOp":
+    def astype(self, dtype: str | DataType) -> "ExprWithOp":
         """Cast the content type of the current data to dtype.
 
         Parameters
@@ -197,7 +198,7 @@ def __pow__(self, other: Expr) -> "ExprWithOp":
     def __rpow__(self, other: Expr) -> "ExprWithOp":
         return _binary_rhs_helper(other)
 
-    def __call__(self, *args: List[Expr], attrs: Optional[Dict[str, Any]] = None) -> "ExprWithOp":
+    def __call__(self, *args: list[Expr], attrs: dict[str, Any] | None = None) -> "ExprWithOp":
         """Call the variable (if it represents a function).
 
         Parameters
@@ -424,12 +425,12 @@ def asobject(self):
             f"and the DLTensor::shape array can be accessed as {self.tensor}.shape[i]"
         )
 
-    def __getitem__(self, axis: Union[int, PrimExpr, Expr]) -> Expr:
+    def __getitem__(self, axis: int | PrimExpr | Expr) -> Expr:
         """Returns the extent of a tensor axis
 
         Parameters
         ----------
-        axis: Union[int, PrimExpr, Expr]
+        axis: int | PrimExpr | Expr
 
             The tensor axis whose extent should be returned.  For ease
             of use, any python integers or TIR expressions are
@@ -494,12 +495,12 @@ def asobject(self):
             f"and the DLTensor::strides array can be accessed as {self.tensor}.strides[i]"
         )
 
-    def __getitem__(self, axis: Union[int, PrimExpr, Expr]) -> Expr:
+    def __getitem__(self, axis: int | PrimExpr | Expr) -> Expr:
         """Returns the extent of a tensor axis
 
         Parameters
         ----------
-        axis: Union[int, PrimExpr, Expr]
+        axis: int | PrimExpr | Expr
 
             The tensor axis whose extent should be returned.  For ease
             of use, any python integers or TIR expressions are
@@ -557,18 +558,18 @@ class Call(ExprWithOp):
     """
 
     op: Expr
-    args: List[Expr]
+    args: list[Expr]
     attrs: tvm.ir.Attrs
-    sinfo_args: List[StructInfo]
-    span: Optional[Span]
+    sinfo_args: list[StructInfo]
+    span: Span | None
 
     def __init__(
         self,
-        op: Union[Expr, tvm.ir.Op],
-        args: Union[List[Expr], typing.Tuple[Expr, ...]],
-        attrs: Optional[tvm.ir.Attrs] = None,
-        sinfo_args: Optional[Union[List[StructInfo], typing.Tuple[StructInfo, ...]]] = None,
-        span: Optional[Span] = None,
+        op: Expr | tvm.ir.Op,
+        args: list[Expr] | tuple[Expr, ...],
+        attrs: tvm.ir.Attrs | None = None,
+        sinfo_args: list[StructInfo] | tuple[StructInfo, ...] | None = None,
+        span: Span | None = None,
     ):
         if not sinfo_args:
             sinfo_args = []
@@ -604,11 +605,9 @@ class If(ExprWithOp):
     cond: Expr
     true_branch: Expr
     false_branch: Expr
-    span: Optional[Span]
+    span: Span | None
 
-    def __init__(
-        self, cond: Expr, true_branch: Expr, false_branch: Expr, span: Optional[Span] = None
-    ):
+    def __init__(self, cond: Expr, true_branch: Expr, false_branch: Expr, span: Span | None = None):
         self.__init_handle_by_constructor__(
             _ffi_api.If,
             cond,
@@ -631,12 +630,10 @@ class Tuple(ExprWithOp):
         Span that points to original source code
     """
 
-    fields: List[Expr]
-    span: Optional[Span]
+    fields: list[Expr]
+    span: Span | None
 
-    def __init__(
-        self, fields: Union[List[Expr], typing.Tuple[Expr, ...]], span: Optional[Span] = None
-    ):
+    def __init__(self, fields: list[Expr] | tuple[Expr, ...], span: Span | None = None):
         if isinstance(fields, tvm.relax.Tuple):
             fields = fields.fields
         elif isinstance(getattr(fields, "struct_info_", None), tvm.relax.TupleStructInfo):
@@ -671,9 +668,9 @@ class TupleGetItem(ExprWithOp):
 
     tuple_value: Expr
     index: int
-    span: Optional[Span]
+    span: Span | None
 
-    def __init__(self, tuple_value: Expr, index: int, span: Optional[Span] = None):
+    def __init__(self, tuple_value: Expr, index: int, span: Span | None = None):
         self.__init_handle_by_constructor__(
             _ffi_api.TupleGetItem,
             tuple_value,
@@ -695,13 +692,13 @@ class ShapeExpr(ExprWithOp):
         Span that points to original source code
     """
 
-    values: List[PrimExpr]
-    span: Optional[Span]
+    values: list[PrimExpr]
+    span: Span | None
 
     def __init__(
         self,
-        values: Union[List[PrimExpr], typing.Tuple[PrimExpr, ...], tvm.ir.Array],
-        span: Optional[Span] = None,
+        values: list[PrimExpr] | tuple[PrimExpr, ...] | tvm.ir.Array,
+        span: Span | None = None,
     ) -> None:
         self.__init_handle_by_constructor__(_ffi_api.ShapeExpr, values, span)  # type: ignore
 
@@ -714,8 +711,8 @@ def __len__(self):
         return len(self.values)
 
 
-def make_shape(shape: Union[List[Any], typing.Tuple[Any, ...]]) -> ShapeExpr:
-    if isinstance(shape, (list, tuple)):
+def make_shape(shape: list[Any] | tuple[Any, ...]) -> ShapeExpr:
+    if isinstance(shape, list | tuple):
         return ShapeExpr(shape)
     raise ValueError("Wrong type")
 
@@ -741,13 +738,13 @@ class Constant(ExprWithOp):
     """
 
     data: tvm.runtime.Tensor
-    span: Optional[Span]
+    span: Span | None
 
     def __init__(
         self,
         data: tvm.runtime.Tensor,
-        struct_info: Optional[StructInfo] = None,
-        span: Optional[Span] = None,
+        struct_info: StructInfo | None = None,
+        span: Span | None = None,
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.Constant,
@@ -763,7 +760,7 @@ class Var(ExprWithOp):
 
     Parameters
     ----------
-    name_hint: Union[str, Id]
+    name_hint: str | Id
         The name hint of the variable.
 
     struct_info: Optional[StructInfo]
@@ -774,13 +771,13 @@ class Var(ExprWithOp):
     """
 
     vid: Id
-    span: Optional[Span]
+    span: Span | None
 
     def __init__(
         self,
-        name_hint: Union[str, Id],
-        struct_info: Optional[StructInfo] = None,
-        span: Optional[Span] = None,
+        name_hint: str | Id,
+        struct_info: StructInfo | None = None,
+        span: Span | None = None,
     ) -> None:
         if struct_info is not None:
             struct_info = tvm.runtime.convert(struct_info)
@@ -812,7 +809,7 @@ class DataflowVar(Var):
 
     Parameters
     ----------
-    name_hint: Union[str, Id]
+    name_hint: str | Id
         The name hint of the variable.
 
     struct_info: Optional[StructInfo]
@@ -823,13 +820,13 @@ class DataflowVar(Var):
     """
 
     vid: Id
-    span: Optional[Span]
+    span: Span | None
 
     def __init__(
         self,
-        name_hint: Union[str, Id],
-        struct_info: Optional[StructInfo] = None,
-        span: Optional[Span] = None,
+        name_hint: str | Id,
+        struct_info: StructInfo | None = None,
+        span: Span | None = None,
     ) -> None:
         # pylint: disable=super-init-not-called
         if struct_info is not None:
@@ -859,7 +856,7 @@ class PrimValue(Expr, Scriptable):
 
     value: PrimExpr
 
-    def __init__(self, value: Union[PrimExpr, int], span: Optional[Span] = None) -> None:
+    def __init__(self, value: PrimExpr | int, span: Span | None = None) -> None:
         if isinstance(value, int):
             value = tvm.tir.IntImm("int64", value)
         self.__init_handle_by_constructor__(_ffi_api.PrimValue, value, span)  # type: ignore
@@ -870,9 +867,9 @@ class StringImm(Expr, Scriptable):
     """Represent a string literal constant."""
 
     value: str
-    span: Optional[Span]
+    span: Span | None
 
-    def __init__(self, value: str, span: Optional[Span] = None) -> None:
+    def __init__(self, value: str, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.StringImm, value, span)  # type: ignore
 
 
@@ -881,9 +878,9 @@ class DataTypeImm(Expr, Scriptable):
     """Represent a data type constant."""
 
     value: DataType
-    span: Optional[Span]
+    span: Span | None
 
-    def __init__(self, value: Union[DataType, str], span: Optional[Span] = None) -> None:
+    def __init__(self, value: DataType | str, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.DataTypeImm, value, span)  # type: ignore
 
 
@@ -892,7 +889,7 @@ class Binding(Node, Scriptable):
     """The base class of a binding in Relax."""
 
     var: Var
-    span: Optional[Span]
+    span: Span | None
 
 
 @tvm_ffi.register_object("relax.expr.MatchCast")
@@ -917,10 +914,10 @@ class MatchCast(Binding):
 
     struct_info: StructInfo
     value: Expr
-    span: Optional[Span]
+    span: Span | None
 
     def __init__(
-        self, var: Var, value: Expr, struct_info: StructInfo, span: Optional[Span] = None
+        self, var: Var, value: Expr, struct_info: StructInfo, span: Span | None = None
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.MatchCast,
@@ -947,9 +944,9 @@ class VarBinding(Binding):
 
     var: Var
     value: Expr
-    span: Optional[Span]
+    span: Span | None
 
-    def __init__(self, var: Var, value: Expr, span: Optional[Span] = None) -> None:
+    def __init__(self, var: Var, value: Expr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.VarBinding, var, value, span)  # type: ignore
 
 
@@ -958,10 +955,10 @@ class BindingBlock(Node, Scriptable):
     """base class of binding block, bindings inside can be impure
     (with side effect or control flow)"""
 
-    bindings: List[Binding]
-    span: Optional[Span]
+    bindings: list[Binding]
+    span: Span | None
 
-    def __init__(self, bindings: List[Binding], span: Optional[Span] = None) -> None:
+    def __init__(self, bindings: list[Binding], span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.BindingBlock, bindings, span)  # type: ignore
 
 
@@ -969,10 +966,10 @@ def __init__(self, bindings: List[Binding], span: Optional[Span] = None) -> None
 class DataflowBlock(BindingBlock):
     """dataflow block, bindings inside are pure (no side effect and no control flow)"""
 
-    bindings: List[Binding]
-    span: Optional[Span]
+    bindings: list[Binding]
+    span: Span | None
 
-    def __init__(self, bindings: List[Binding], span: Optional[Span] = None) -> None:
+    def __init__(self, bindings: list[Binding], span: Span | None = None) -> None:
         # pylint: disable=super-init-not-called
         self.__init_handle_by_constructor__(_ffi_api.DataflowBlock, bindings, span)  # type: ignore
 
@@ -981,11 +978,11 @@ def __init__(self, bindings: List[Binding], span: Optional[Span] = None) -> None
 class SeqExpr(ExprWithOp):
     """A sequence of binding blocks followed by an expression."""
 
-    blocks: List[BindingBlock]
+    blocks: list[BindingBlock]
     body: Expr
-    span: Optional[Span]
+    span: Span | None
 
-    def __init__(self, blocks: List[BindingBlock], body: Expr, span: Optional[Span] = None) -> None:
+    def __init__(self, blocks: list[BindingBlock], body: Expr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.SeqExpr, blocks, body, span)  # type: ignore
 
 
@@ -993,21 +990,21 @@ def __init__(self, blocks: List[BindingBlock], body: Expr, span: Optional[Span]
 class Function(BaseFunc, Scriptable):
     """A Relax function."""
 
-    params: List[Var]
+    params: list[Var]
     body: Expr
     ret_struct_info: StructInfo
     is_pure: bool
     attrs: tvm.ir.DictAttrs
-    span: Optional[Span]
+    span: Span | None
 
     def __init__(
         self,
-        params: List[Var],
+        params: list[Var],
         body: Expr,
-        ret_struct_info: Optional[StructInfo] = None,
-        is_pure: Optional[bool] = True,
-        attrs: Optional[tvm.ir.DictAttrs] = None,
-        span: Optional[Span] = None,
+        ret_struct_info: StructInfo | None = None,
+        is_pure: bool | None = True,
+        attrs: tvm.ir.DictAttrs | None = None,
+        span: Span | None = None,
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.Function,
@@ -1021,11 +1018,11 @@ def __init__(
 
     @staticmethod
     def create_empty(
-        params: List[Var],
+        params: list[Var],
         ret_struct_info: StructInfo,
-        is_pure: Optional[bool] = True,
-        attrs: Optional[tvm.ir.DictAttrs] = None,
-        span: Optional[Span] = None,
+        is_pure: bool | None = True,
+        attrs: tvm.ir.DictAttrs | None = None,
+        span: Span | None = None,
     ):
         """Construct a relax.Function but without body"""
         return _ffi_api.FunctionCreateEmpty(params, ret_struct_info, is_pure, attrs, span)  # type: ignore
@@ -1040,14 +1037,12 @@ def __call__(self, *args):
         """
         return Call(self, args, None, None)
 
-    def bind_symbolic_vars(
-        self, binding_map: Mapping[Union[str, tvm.tir.Var], PrimExpr]
-    ) -> "Function":
+    def bind_symbolic_vars(self, binding_map: Mapping[str | tvm.tir.Var, PrimExpr]) -> "Function":
         """Return a new function with updated symbolic variable
 
         Parameters
         ----------
-        binding_map: Mapping[Union[str, tvm.tir.Var], PrimExpr]
+        binding_map: Mapping[str | tvm.tir.Var, PrimExpr]
 
             The mapping of values to be replaced.  Keys may be either
             a `tir.Var` or a string name of the variable.  If the
@@ -1073,8 +1068,8 @@ def bind_symbolic_vars(
     def bind_params(
         self,
         binding_map: Mapping[
-            Union[str, Var],
-            Union[int, float, PrimExpr, tvm.runtime.Tensor, _np.ndarray, Expr],
+            str | Var,
+            int | float | PrimExpr | tvm.runtime.Tensor | _np.ndarray | Expr,
         ],
     ) -> "Function":
         """Return a new function with updated symbolic variable
@@ -1082,8 +1077,8 @@ def bind_params(
         Parameters
         ----------
         binding_map: Mapping[
-                Union[str, Var],
-                Union[int, float, PrimExpr, tvm.runtime.Tensor, _np.ndarray, Expr],
+                str | Var,
+                int | float | PrimExpr | tvm.runtime.Tensor | _np.ndarray | Expr,
         ]
 
             The mapping of values to be replaced.
@@ -1111,7 +1106,7 @@ def _normalize_value(value):
                 # Relax uses int64 for symbolic variables, but the FFI
                 # converts python integers into int32.
                 return tvm.tir.const(value, "int64")
-            elif isinstance(value, (_np.ndarray, tvm.runtime.Tensor)):
+            elif isinstance(value, _np.ndarray | tvm.runtime.Tensor):
                 return tvm.relax.const(value)
             else:
                 return value
@@ -1121,7 +1116,7 @@ def _normalize_value(value):
         return _ffi_api.FunctionBindParams(self, binding_map)  # type: ignore
 
     def inline_functions(
-        self, function_map: Mapping[Union[str, tvm.ir.GlobalVar], "Function"]
+        self, function_map: Mapping[str | tvm.ir.GlobalVar, "Function"]
     ) -> "Function":
         return _ffi_api.FunctionInlineFunctions(self, function_map)  # type: ignore
 
@@ -1131,13 +1126,13 @@ class ExternFunc(BaseFunc, ExprWithOp):
     """extern function, which represents a PackedFunc."""
 
     global_symbol: String
-    span: Optional[Span]
+    span: Span | None
 
     def __init__(
         self,
         global_symbol: String,
-        struct_info: Optional[StructInfo] = None,
-        span: Optional[Span] = None,
+        struct_info: StructInfo | None = None,
+        span: Span | None = None,
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.ExternFunc,
@@ -1147,19 +1142,19 @@ def __init__(
         )
 
 
-def extern(name: str, struct_info: Optional[StructInfo] = None, span: Optional[Span] = None):
+def extern(name: str, struct_info: StructInfo | None = None, span: Span | None = None):
     """Create extern function."""
     return ExternFunc(name, struct_info, span)
 
 
 def const(
-    value: Union[bool, int, float, _np.ndarray, tvm.runtime.Tensor], dtype: Optional[str] = None
+    value: bool | int | float | _np.ndarray | tvm.runtime.Tensor, dtype: str | None = None
 ) -> Constant:
     """Create a constant value.
 
     Parameters
     ----------
-    value: Union[bool, int, float, numpy.ndarray, tvm.runtime.Tensor]
+    value: bool | int | float | numpy.ndarray | tvm.runtime.Tensor
         The constant value.
 
     dtype: Optional[str]
@@ -1177,7 +1172,7 @@ def const(
     # Needed for bf16 and fp8 support (does not come with numpy)
     import ml_dtypes  # pylint: disable=unused-import,import-outside-toplevel
 
-    if isinstance(value, (Number, (bool, list))):
+    if isinstance(value, Number | (bool | list)):
         value = _np.array(value, dtype=dtype)
 
     if not dtype:
@@ -1190,7 +1185,7 @@ def const(
             None,  # type: ignore
         )
 
-    if isinstance(value, (_np.ndarray, _np.generic)):
+    if isinstance(value, _np.ndarray | _np.generic):
         if dtype is not None:
             value = value.astype(dtype)
         value = tvm.runtime.tensor(value)
@@ -1207,7 +1202,7 @@ class TEPlaceholderOp(tvm.te.tensor.Operation):
 
 
 def te_tensor(
-    value: Expr, tir_var_map: Dict[tvm.tir.Var, tvm.tir.PrimExpr], name: str = "rxplaceholder"
+    value: Expr, tir_var_map: dict[tvm.tir.Var, tvm.tir.PrimExpr], name: str = "rxplaceholder"
 ):
     """Create a TE tensor from relax expression, with TIR variables in the
     tensor shape substituted by the given mapping
@@ -1251,5 +1246,5 @@ def get_shape_of(expr: Expr) -> Expr:
     return _ffi_api.GetShapeOf(expr)  # type: ignore
 
 
-def _update_struct_info(expr: Expr, struct_info: Optional[StructInfo]) -> None:
+def _update_struct_info(expr: Expr, struct_info: StructInfo | None) -> None:
     _ffi_api.UpdateStructInfo(expr, struct_info)  # type: ignore
diff --git a/python/tvm/relax/expr_functor.py b/python/tvm/relax/expr_functor.py
index 7cbd4d2a380f..5ac77da3c04d 100644
--- a/python/tvm/relax/expr_functor.py
+++ b/python/tvm/relax/expr_functor.py
@@ -18,7 +18,7 @@
 # ruff: noqa: RUF012
 """The expression functor of Relax."""
 
-from typing import Callable, Optional
+from collections.abc import Callable
 
 import tvm_ffi
 
@@ -277,33 +277,33 @@ class _PyExprVisitor(tvm_ffi.core.Object):
 
     def __init__(
         self,
-        f_visit_expr: Optional[Callable] = None,
-        f_visit_constant_: Optional[Callable] = None,
-        f_visit_tuple_: Optional[Callable] = None,
-        f_visit_var_: Optional[Callable] = None,
-        f_visit_dataflow_var_: Optional[Callable] = None,
-        f_visit_shape_expr_: Optional[Callable] = None,
-        f_visit_extern_func_: Optional[Callable] = None,
-        f_visit_global_var_: Optional[Callable] = None,
-        f_visit_function_: Optional[Callable] = None,
-        f_visit_call_: Optional[Callable] = None,
-        f_visit_seq_expr_: Optional[Callable] = None,
-        f_visit_if_: Optional[Callable] = None,
-        f_visit_op_: Optional[Callable] = None,
-        f_visit_tuple_getitem_: Optional[Callable] = None,
-        f_visit_prim_value_: Optional[Callable] = None,
-        f_visit_string_imm_: Optional[Callable] = None,
-        f_visit_data_type_imm_: Optional[Callable] = None,
-        f_visit_binding: Optional[Callable] = None,
-        f_visit_var_binding_: Optional[Callable] = None,
-        f_visit_match_cast_: Optional[Callable] = None,
-        f_visit_binding_block: Optional[Callable] = None,
-        f_visit_binding_block_: Optional[Callable] = None,
-        f_visit_dataflow_block_: Optional[Callable] = None,
-        f_visit_var_def: Optional[Callable] = None,
-        f_visit_var_def_: Optional[Callable] = None,
-        f_visit_dataflow_var_def_: Optional[Callable] = None,
-        f_visit_span: Optional[Callable] = None,
+        f_visit_expr: Callable | None = None,
+        f_visit_constant_: Callable | None = None,
+        f_visit_tuple_: Callable | None = None,
+        f_visit_var_: Callable | None = None,
+        f_visit_dataflow_var_: Callable | None = None,
+        f_visit_shape_expr_: Callable | None = None,
+        f_visit_extern_func_: Callable | None = None,
+        f_visit_global_var_: Callable | None = None,
+        f_visit_function_: Callable | None = None,
+        f_visit_call_: Callable | None = None,
+        f_visit_seq_expr_: Callable | None = None,
+        f_visit_if_: Callable | None = None,
+        f_visit_op_: Callable | None = None,
+        f_visit_tuple_getitem_: Callable | None = None,
+        f_visit_prim_value_: Callable | None = None,
+        f_visit_string_imm_: Callable | None = None,
+        f_visit_data_type_imm_: Callable | None = None,
+        f_visit_binding: Callable | None = None,
+        f_visit_var_binding_: Callable | None = None,
+        f_visit_match_cast_: Callable | None = None,
+        f_visit_binding_block: Callable | None = None,
+        f_visit_binding_block_: Callable | None = None,
+        f_visit_dataflow_block_: Callable | None = None,
+        f_visit_var_def: Callable | None = None,
+        f_visit_var_def_: Callable | None = None,
+        f_visit_dataflow_var_def_: Callable | None = None,
+        f_visit_span: Callable | None = None,
     ) -> None:
         """Constructor."""
 
@@ -798,33 +798,33 @@ class _PyExprMutator(Object):
     def __init__(
         self,
         builder: BlockBuilder = None,
-        f_visit_expr: Optional[Callable] = None,
-        f_visit_constant_: Optional[Callable] = None,
-        f_visit_tuple_: Optional[Callable] = None,
-        f_visit_var_: Optional[Callable] = None,
-        f_visit_dataflow_var_: Optional[Callable] = None,
-        f_visit_shape_expr_: Optional[Callable] = None,
-        f_visit_extern_func_: Optional[Callable] = None,
-        f_visit_global_var_: Optional[Callable] = None,
-        f_visit_function_: Optional[Callable] = None,
-        f_visit_call_: Optional[Callable] = None,
-        f_visit_seq_expr_: Optional[Callable] = None,
-        f_visit_if_: Optional[Callable] = None,
-        f_visit_op_: Optional[Callable] = None,
-        f_visit_tuple_getitem_: Optional[Callable] = None,
-        f_visit_prim_value_: Optional[Callable] = None,
-        f_visit_string_imm_: Optional[Callable] = None,
-        f_visit_data_type_imm_: Optional[Callable] = None,
-        f_visit_binding: Optional[Callable] = None,
-        f_visit_var_binding_: Optional[Callable] = None,
-        f_visit_match_cast_: Optional[Callable] = None,
-        f_visit_binding_block: Optional[Callable] = None,
-        f_visit_binding_block_: Optional[Callable] = None,
-        f_visit_dataflow_block_: Optional[Callable] = None,
-        f_visit_var_def: Optional[Callable] = None,
-        f_visit_var_def_: Optional[Callable] = None,
-        f_visit_dataflow_var_def_: Optional[Callable] = None,
-        f_visit_span: Optional[Callable] = None,
+        f_visit_expr: Callable | None = None,
+        f_visit_constant_: Callable | None = None,
+        f_visit_tuple_: Callable | None = None,
+        f_visit_var_: Callable | None = None,
+        f_visit_dataflow_var_: Callable | None = None,
+        f_visit_shape_expr_: Callable | None = None,
+        f_visit_extern_func_: Callable | None = None,
+        f_visit_global_var_: Callable | None = None,
+        f_visit_function_: Callable | None = None,
+        f_visit_call_: Callable | None = None,
+        f_visit_seq_expr_: Callable | None = None,
+        f_visit_if_: Callable | None = None,
+        f_visit_op_: Callable | None = None,
+        f_visit_tuple_getitem_: Callable | None = None,
+        f_visit_prim_value_: Callable | None = None,
+        f_visit_string_imm_: Callable | None = None,
+        f_visit_data_type_imm_: Callable | None = None,
+        f_visit_binding: Callable | None = None,
+        f_visit_var_binding_: Callable | None = None,
+        f_visit_match_cast_: Callable | None = None,
+        f_visit_binding_block: Callable | None = None,
+        f_visit_binding_block_: Callable | None = None,
+        f_visit_dataflow_block_: Callable | None = None,
+        f_visit_var_def: Callable | None = None,
+        f_visit_var_def_: Callable | None = None,
+        f_visit_dataflow_var_def_: Callable | None = None,
+        f_visit_span: Callable | None = None,
     ) -> None:
         """Constructor."""
 
@@ -972,7 +972,7 @@ def MyExprMutator(PyExprMutator):
         ],
     }
 
-    def __init__(self, mod: Optional[IRModule] = None) -> None:
+    def __init__(self, mod: IRModule | None = None) -> None:
         """Constructor"""
         self.builder_ = BlockBuilder(mod)
 
@@ -1504,7 +1504,7 @@ def visit_with_new_scope(self, expr: Expr) -> Expr:
         # Using self._outer() to ref _PyExprMutator
         return _ffi_api.PyExprMutatorVisitWithNewScope(self._outer(), expr)  # type: ignore
 
-    def lookup_binding(self, var: Var) -> Optional[Expr]:
+    def lookup_binding(self, var: Var) -> Expr | None:
         """Look up the value bound to a variable.
         Note: For function parameters, this function returns std::nullopt.
 
diff --git a/python/tvm/relax/frontend/common.py b/python/tvm/relax/frontend/common.py
index e58ca35d1f3a..63b58cd8f113 100644
--- a/python/tvm/relax/frontend/common.py
+++ b/python/tvm/relax/frontend/common.py
@@ -17,15 +17,13 @@
 # pylint: disable=invalid-name
 """Commons for Relax frontend."""
 
-from typing import Dict, List, Tuple
-
 import numpy as _np
 
 import tvm
 from tvm import topi
 
 
-def detach_params(mod: tvm.IRModule) -> Tuple[tvm.IRModule, Dict[str, List[tvm.runtime.Tensor]]]:
+def detach_params(mod: tvm.IRModule) -> tuple[tvm.IRModule, dict[str, list[tvm.runtime.Tensor]]]:
     """Detach the attribute "params" in the functions of the input IRModule as
     separate dictionary of params.
 
diff --git a/python/tvm/relax/frontend/nn/_tensor_op.py b/python/tvm/relax/frontend/nn/_tensor_op.py
index 0a5142f3a1d0..d7acc48fd22a 100644
--- a/python/tvm/relax/frontend/nn/_tensor_op.py
+++ b/python/tvm/relax/frontend/nn/_tensor_op.py
@@ -17,8 +17,6 @@
 # ruff: noqa: F821
 """Adding member operators to nn.Tensor."""
 
-from typing import Optional
-
 from tvm import tir
 
 
@@ -33,9 +31,9 @@ def _convert_scalar(scalar, ref) -> "Tensor":
 
     if isinstance(scalar, Tensor):
         return scalar
-    if isinstance(scalar, (tir.FloatImm, tir.IntImm)):
+    if isinstance(scalar, tir.FloatImm | tir.IntImm):
         return Tensor.from_scalar(scalar.value, dtype=ref.dtype)
-    if isinstance(scalar, (int, float)):
+    if isinstance(scalar, int | float):
         return Tensor.from_scalar(scalar, dtype=ref.dtype)
     return scalar
 
@@ -102,5 +100,5 @@ def reshape(self, *shape):
     def permute_dims(self, *axes):
         return _op().permute_dims(self, axes)
 
-    def repeat(self, repeats: int, axis: Optional[int] = None):
+    def repeat(self, repeats: int, axis: int | None = None):
         return _op().repeat(self, repeats, axis)
diff --git a/python/tvm/relax/frontend/nn/core.py b/python/tvm/relax/frontend/nn/core.py
index 875da8d9cb91..2ac461cbff89 100644
--- a/python/tvm/relax/frontend/nn/core.py
+++ b/python/tvm/relax/frontend/nn/core.py
@@ -25,16 +25,10 @@
 """
 
 from collections import OrderedDict
+from collections.abc import Callable, Iterator, Sequence
 from typing import (
     TYPE_CHECKING,
     Any,
-    Callable,
-    Dict,
-    Iterator,
-    List,
-    Optional,
-    Sequence,
-    Tuple,
     Union,
 )
 
@@ -123,7 +117,7 @@ def from_const(data) -> "Tensor":
         return Tensor(_expr=rx.const(data))
 
     @staticmethod
-    def from_scalar(data: Union[int, float], dtype: str) -> "Tensor":
+    def from_scalar(data: int | float, dtype: str) -> "Tensor":
         """Construct a tensor from a scalar with dtype specified."""
         return Tensor(_expr=rx.const(data, dtype=dtype))
 
@@ -139,7 +133,7 @@ def from_struct_info(struct_info: rx.TensorStructInfo, name: str = "tensor") ->
 
     @staticmethod
     def placeholder(
-        shape: Sequence[Union[int, str, tir.PrimExpr]],
+        shape: Sequence[int | str | tir.PrimExpr],
         dtype: str,
         name: str = "tensor",
     ) -> "Tensor":
@@ -151,7 +145,7 @@ def placeholder(
         """
         new_shape = []
         for expr in shape:
-            if isinstance(expr, (int, tir.IntImm)):
+            if isinstance(expr, int | tir.IntImm):
                 expr = int(expr)
                 assert expr >= 0
                 new_shape.append(expr)
@@ -175,7 +169,7 @@ def placeholder(
         )
 
     @property
-    def shape(self) -> List[Union[int, tir.PrimExpr]]:
+    def shape(self) -> list[int | tir.PrimExpr]:
         """Returns the shape of the tensor as a list of integers.
 
         An integer can be a python int or tvm.tir.PrimExpr, depending on whether the shape is
@@ -226,13 +220,13 @@ class Parameter(Tensor):
     it is called a bound parameter, otherwise it is called an unbound parameter.
     """
 
-    _data: Optional[Tensor]
-    attrs: Dict[str, Any]
+    _data: Tensor | None
+    attrs: dict[str, Any]
 
     def __init__(
         self,
-        shape: Sequence[Union[int, str, tir.PrimExpr]],
-        dtype: Optional[str] = None,
+        shape: Sequence[int | str | tir.PrimExpr],
+        dtype: str | None = None,
     ) -> None:
         """Create a parameter with given shape and dtype. The parameter is not bound to any
         concrete values.
@@ -252,7 +246,7 @@ def __init__(
         self.attrs = OrderedDict()
 
     @property
-    def data(self) -> Optional[Tensor]:
+    def data(self) -> Tensor | None:
         """Returns the concrete value of the parameter if it is bound to a concrete value,
         otherwise returns None. The returned value is a tvm.runtime.Tensor."""
         return self._data
@@ -283,7 +277,7 @@ def data(self, data: Union[None, tvm.runtime.Tensor, np.ndarray, "torch.Tensor"]
             raise ValueError(f"Dtype mismatch: expected {self.dtype}, got {data.dtype}")
         self._data = data
 
-    def to(self, dtype: Optional[str] = None) -> None:  # pylint: disable=invalid-name
+    def to(self, dtype: str | None = None) -> None:  # pylint: disable=invalid-name
         """Change the dtype of the parameter if it is not bound to any concrete data"""
         if dtype is not None:
             if self._data is not None:
@@ -318,24 +312,24 @@ class Effect:
     effects, for example, print. It is used to represent the output of a computation.
     """
 
-    def emit_init(self, name_hint: str, builder: BlockBuilder) -> List[rx.DataflowVar]:
+    def emit_init(self, name_hint: str, builder: BlockBuilder) -> list[rx.DataflowVar]:
         """Emit the initialization of the effect. This method is called by the compiler to
         initialize the effect."""
         raise NotImplementedError
 
-    def create(self, name_hint: str) -> List[rx.Var]:
+    def create(self, name_hint: str) -> list[rx.Var]:
         """Create the implicit inputs to a relax.Function that represents the side effect"""
         raise NotImplementedError
 
-    def set_state(self, state_vars: List[rx.Var]) -> None:
+    def set_state(self, state_vars: list[rx.Var]) -> None:
         """Set the variables that represents the effect"""
         raise NotImplementedError
 
-    def finalize(self) -> List[rx.Var]:
+    def finalize(self) -> list[rx.Var]:
         """finalize the effect as the implicit return value of a relax.Function"""
         raise NotImplementedError
 
-    def to(self, dtype: Optional[str] = None) -> None:  # pylint: disable=invalid-name
+    def to(self, dtype: str | None = None) -> None:  # pylint: disable=invalid-name
         """Convert the effect to specific dtype. Usually it is no-op for most of the effects"""
 
 
@@ -343,7 +337,7 @@ class Module(SubroutineMixin):
     """Base class for neural network components. Subclass it to build your models.
     Modules can nest within each other in a tree structure using regular attribute assignment."""
 
-    def named_parameters(self, prefix: str = "") -> Iterator[Tuple[str, Parameter]]:
+    def named_parameters(self, prefix: str = "") -> Iterator[tuple[str, Parameter]]:
         """This method provides an iterator over module parameters,
         yielding both the parameter name and its corresponding value.
 
@@ -372,8 +366,8 @@ def parameters(self) -> Iterator[Parameter]:
             yield param
 
     def state_dict(
-        self, *, prefix: str = "", destination: Optional[Dict[str, Parameter]] = None
-    ) -> Dict[str, Parameter]:
+        self, *, prefix: str = "", destination: dict[str, Parameter] | None = None
+    ) -> dict[str, Parameter]:
         """Returns a dictionary containing references to the whole state of the module.
 
         Parameters
@@ -397,8 +391,8 @@ def state_dict(
         return destination
 
     def load_state_dict(
-        self, state_dict: Dict[str, Parameter], strict: bool = True
-    ) -> Tuple[List[str], List[str]]:
+        self, state_dict: dict[str, Parameter], strict: bool = True
+    ) -> tuple[list[str], list[str]]:
         """This function copies parameters and buffers from the state_dict into the current module
         and its descendants. If `strict` is set to True, the keys in the `state_dict` must exactly
         match the keys returned by the `state_dict()` function of this module.
@@ -417,8 +411,8 @@ def load_state_dict(
             A tuple of two lists: the missing keys and the unexpected keys.
         """
         self_state_dict = self.state_dict()
-        missing_keys: List[str] = []
-        unexpected_keys: List[str] = []
+        missing_keys: list[str] = []
+        unexpected_keys: list[str] = []
         for key, value in state_dict.items():
             if key not in self_state_dict:
                 unexpected_keys.append(key)
@@ -437,7 +431,7 @@ def __call__(self, *args: Any, **kwargs: Any) -> Any:
             raise NotImplementedError(f"Module {type(self)} does not have a `forward` method")
         return self.forward(*args, **kwargs)  # pylint: disable=no-member
 
-    def to(self, dtype: Optional[str] = None) -> None:  # pylint: disable=invalid-name
+    def to(self, dtype: str | None = None) -> None:  # pylint: disable=invalid-name
         """Convert the module to specific dtype recursively"""
         for _, item in self.__dict__.items():
             if hasattr(item, "to") and callable(item.to):
@@ -450,17 +444,10 @@ def export_tvm(
         spec: "_spec.ModuleSpecType",
         debug: bool = False,
         allow_extern: bool = False,
-    ) -> Union[
-        Tuple[
-            IRModule,
-            List[Tuple[str, Parameter]],
-        ],
-        Tuple[
-            IRModule,
-            List[Tuple[str, Parameter]],
-            List["ExternModule"],
-        ],
-    ]:
+    ) -> (
+        tuple[IRModule, list[tuple[str, Parameter]]]
+        | tuple[IRModule, list[tuple[str, Parameter]], list["ExternModule"]]
+    ):
         """Export the module to TVM IRModule and parameters
 
         Parameters
@@ -500,8 +487,8 @@ def export_tvm(
     def jit(  # pylint: disable=too-many-arguments
         self,
         spec: "_spec.ModuleSpec",
-        device: Union[str, Device] = "cpu",
-        pipeline: Union[None, str, Pass] = "default_build",
+        device: str | Device = "cpu",
+        pipeline: None | str | Pass = "default_build",
         out_format: str = "torch",
         debug: bool = False,
     ) -> Any:
@@ -544,7 +531,7 @@ def _compile(spec, device, pipeline, debug):
 class ModuleDict(Module):
     """Holds submodules in a dict."""
 
-    def __init__(self, modules: Optional[OrderedDict[str, Module]] = None):
+    def __init__(self, modules: OrderedDict[str, Module] | None = None):
         if modules is None:
             self.modules = OrderedDict()
         else:
@@ -568,13 +555,13 @@ def keys(self) -> Iterator[str]:
     def values(self) -> Iterator[Module]:
         return self.modules.values()
 
-    def items(self) -> Iterator[Tuple[str, Module]]:
+    def items(self) -> Iterator[tuple[str, Module]]:
         return self.modules.items()
 
-    def get(self, key: str, default: Optional[Module] = None) -> Optional[Module]:
+    def get(self, key: str, default: Module | None = None) -> Module | None:
         return self.modules.get(key, default)
 
-    def update(self, modules: Dict[str, Module]) -> None:
+    def update(self, modules: dict[str, Module]) -> None:
         self.modules.update(modules)
 
     def clear(self) -> None:
@@ -586,7 +573,7 @@ def pop(self, key: str) -> Module:
     def __contains__(self, key: str) -> bool:
         return key in self.modules
 
-    def to(self, dtype: Optional[str] = None) -> None:  # pylint: disable=invalid-name
+    def to(self, dtype: str | None = None) -> None:  # pylint: disable=invalid-name
         for module in self.modules.values():
             module.to(dtype=dtype)
 
@@ -594,7 +581,7 @@ def to(self, dtype: Optional[str] = None) -> None:  # pylint: disable=invalid-na
 class ModuleList(Module):
     """Holds submodules in a list."""
 
-    def __init__(self, modules: List[Module]):
+    def __init__(self, modules: list[Module]):
         self.modules = modules
 
     def __iter__(self):
@@ -613,7 +600,7 @@ def append(self, module: Module):
         """Add a module to the end of the ModuleList"""
         self.modules.append(module)
 
-    def to(self, dtype: Optional[str] = None) -> None:  # pylint: disable=invalid-name
+    def to(self, dtype: str | None = None) -> None:  # pylint: disable=invalid-name
         for module in self.modules:
             module.to(dtype=dtype)
 
@@ -624,7 +611,7 @@ def forward(self, x):  # pylint: disable=invalid-name
         return x
 
 
-def wrap_nested(expr: rx.Expr, name: str) -> Union[Tensor, Sequence[Tensor]]:
+def wrap_nested(expr: rx.Expr, name: str) -> Tensor | Sequence[Tensor]:
     """Wrap the given relax.Expr, emit it using the current BlockBuilder,
     and automatically handle nested cases if the expr represents a Tuple.
 
@@ -708,8 +695,8 @@ def _from_dlpack(tensor) -> tvm.runtime.Tensor:
 
 
 def _param_to_tensor(
-    params: List[Tuple[str, Parameter]], device: Device
-) -> List[tvm.runtime.Tensor]:
+    params: list[tuple[str, Parameter]], device: Device
+) -> list[tvm.runtime.Tensor]:
     results = []
     missing = []
     for name, param in params:
diff --git a/python/tvm/relax/frontend/nn/exporter.py b/python/tvm/relax/frontend/nn/exporter.py
index 07db1821f64c..a6cf334d1acb 100644
--- a/python/tvm/relax/frontend/nn/exporter.py
+++ b/python/tvm/relax/frontend/nn/exporter.py
@@ -50,7 +50,7 @@ class Exporter:
 
     builder: BlockBuilder
     io_effect: core.Effect
-    extern_mods: typing.List[extern.ExternModule]
+    extern_mods: list[extern.ExternModule]
 
     def __init__(self, debug: bool) -> None:
         self.builder = BlockBuilder()
@@ -75,7 +75,7 @@ def __exit__(self, exc_type, exc, traceback) -> None:
     def add_external_module(self, mod: extern.ExternModule) -> None:
         """Add an external module to the exporter."""
         # pylint: disable=protected-access
-        all_symbols: typing.List[str] = []
+        all_symbols: list[str] = []
         for extern_mod in self.extern_mods:
             all_symbols.extend(extern_mod._symbols.keys())
         duplicated_symbols = list(set(mod._symbols.keys()) & set(all_symbols))
@@ -87,15 +87,15 @@ def add_external_module(self, mod: extern.ExternModule) -> None:
     def build(  # pylint: disable=too-many-locals
         self,
         spec: _spec.ModuleSpec,
-    ) -> typing.Tuple[
+    ) -> tuple[
         IRModule,
-        typing.List[typing.Tuple[str, core.Parameter]],
-        typing.List[extern.ExternModule],
+        list[tuple[str, core.Parameter]],
+        list[extern.ExternModule],
     ]:
         """Build the ModuleSpec to TVM IRModule. Returns the IRModule and the parameters."""
 
         # pylint: disable=protected-access
-        def _params() -> typing.List[typing.Tuple[str, core.Parameter]]:
+        def _params() -> list[tuple[str, core.Parameter]]:
             params = []
             for name, param in core._attribute_finder(
                 spec.module, prefix="", condition_yield=lambda x: isinstance(x, core.Parameter)
@@ -103,7 +103,7 @@ def _params() -> typing.List[typing.Tuple[str, core.Parameter]]:
                 params.append((name, param))
             return params
 
-        def _effects() -> typing.List[typing.Tuple[str, core.Effect]]:
+        def _effects() -> list[tuple[str, core.Effect]]:
             result = []
             if self.io_effect is not None:
                 result.append(("", self.io_effect))
@@ -146,7 +146,7 @@ def _effects() -> typing.List[typing.Tuple[str, core.Effect]]:
 
 def _emit_effect_init(
     builder: BlockBuilder,
-    effects: typing.List[typing.Tuple[str, core.Effect]],
+    effects: list[tuple[str, core.Effect]],
 ):
     outputs = []
     for prefix, effect in effects:
@@ -160,15 +160,15 @@ def _emit_effect_init(
 def _emit_method(  # pylint: disable=too-many-locals,too-many-branches,too-many-statements
     builder: BlockBuilder,
     spec: _spec.MethodSpec,
-    params: typing.List[typing.Tuple[str, core.Parameter]],
-    effects: typing.Optional[typing.List[typing.Tuple[str, core.Effect]]],
+    params: list[tuple[str, core.Parameter]],
+    effects: list[tuple[str, core.Effect]] | None,
 ):
     # pylint: disable=protected-access
     # symbolic shape's name mapping to its tir.Var for reuse
-    str2var_params: typing.Dict[str, tir.Var] = {}
+    str2var_params: dict[str, tir.Var] = {}
 
     def _unwrap_ret(expr: typing.Any) -> typing.Any:
-        if isinstance(expr, (core.Tensor, core.Object)):
+        if isinstance(expr, core.Tensor | core.Object):
             return expr._expr
         if isinstance(expr, tuple):
             return rx.Tuple([_unwrap_ret(x) for x in expr])
@@ -179,7 +179,7 @@ def _unwrap_ret(expr: typing.Any) -> typing.Any:
     def _convert_input(arg):
         if isinstance(arg, tir.Var):
             return rx.Var(arg.name, struct_info=ShapeStructInfo(values=[arg]))
-        if isinstance(arg, (core.Tensor, core.Object)):
+        if isinstance(arg, core.Tensor | core.Object):
             return arg._expr  # pylint: disable=protected-access
         if isinstance(arg, _spec.Tuple):
             return rx.Var(
@@ -190,8 +190,8 @@ def _convert_input(arg):
             )
         raise TypeError(f"Unsupported input type: {type(arg)}")
 
-    def _params(mode: str) -> typing.List[rx.Var]:
-        inputs: typing.List[rx.Var] = []
+    def _params(mode: str) -> list[rx.Var]:
+        inputs: list[rx.Var] = []
 
         def _get_var(shape_var: tir.Var) -> tir.Var:
             name = shape_var.name
@@ -222,13 +222,13 @@ def _get_var(shape_var: tir.Var) -> tir.Var:
             return [input_var]
         raise ValueError(f"Invalid param_mode: {mode}")
 
-    def _effects(mode: str) -> typing.List[rx.Var]:
-        unflat_inputs: typing.List[typing.List[rx.Var]] = []
+    def _effects(mode: str) -> list[rx.Var]:
+        unflat_inputs: list[list[rx.Var]] = []
         for name, effect in effects:
             effect_input = effect.create(name)
             effect.set_state(effect_input)
             unflat_inputs.append(effect_input)
-        inputs: typing.List[rx.Var] = functools.reduce(operator.iadd, unflat_inputs, [])
+        inputs: list[rx.Var] = functools.reduce(operator.iadd, unflat_inputs, [])
         if mode == "none":
             return []
         if mode == "plain":
@@ -292,9 +292,9 @@ def _detuple(arg, var: rx.Var, builder: BlockBuilder):
 
 def _method_spec_to_inputs(
     spec: _spec.MethodSpec,
-) -> typing.List[typing.Union[tir.Var, core.Tensor]]:
+) -> list[tir.Var | core.Tensor]:
     """Convert the MethodSpec to a list of inputs to Module's method."""
-    str2var: typing.Dict[str, tir.Var] = {}
+    str2var: dict[str, tir.Var] = {}
 
     def _get_var(name: str) -> tir.Var:
         if name in str2var:
diff --git a/python/tvm/relax/frontend/nn/extern.py b/python/tvm/relax/frontend/nn/extern.py
index 4ad55a0c7e85..50d8b390c9a7 100644
--- a/python/tvm/relax/frontend/nn/extern.py
+++ b/python/tvm/relax/frontend/nn/extern.py
@@ -21,8 +21,8 @@
 import shutil
 import sys
 import tempfile
+from collections.abc import Callable
 from pathlib import Path
-from typing import Callable, Dict, List, Optional, Union
 
 from tvm import tir
 from tvm.contrib import cc as _cc
@@ -38,9 +38,9 @@ class ExternModule:
     incorporate user-provided handcrafted kernels into the exported TVM IRModule.
     """
 
-    _symbols: Dict[str, Callable]
+    _symbols: dict[str, Callable]
 
-    def __init__(self, symbols: Dict[str, Callable]) -> None:
+    def __init__(self, symbols: dict[str, Callable]) -> None:
         self._symbols = symbols
 
     def __getitem__(self, func_name: str) -> Callable:
@@ -60,7 +60,7 @@ def _convert(arg, name: str):
                     return rx.StringImm(arg)
                 if isinstance(arg, tir.PrimExpr):
                     return rx.PrimValue(arg)
-                if isinstance(arg, (tuple, list)):
+                if isinstance(arg, tuple | list):
                     return rx.Tuple([_convert(e, f"{name}_{i}") for i, e in enumerate(arg)])
                 raise TypeError(f"Unsupported input type: {type(arg)}")
 
@@ -86,7 +86,7 @@ class ObjectModule(ExternModule):  # pylint: disable=too-few-public-methods
 
     def __init__(
         self,
-        symbols: Dict[str, Callable],
+        symbols: dict[str, Callable],
         filepath: Path,
     ) -> None:
         if not isinstance(filepath, Path):
@@ -168,11 +168,11 @@ def shape_dtype_inference(a, b):
 
     def __init__(  # pylint: disable=too-many-arguments
         self,
-        symbols: Dict[str, Callable],
-        source_code: Union[str, Path],
+        symbols: dict[str, Callable],
+        source_code: str | Path,
         source_format: str,  # "cpp", "cu"
-        compile_options: Optional[List[str]] = None,
-        compiler: Optional[str] = None,
+        compile_options: list[str] | None = None,
+        compiler: str | None = None,
         output_format: str = "obj",  # "obj", "wasm"
     ):
         """Constructs a `nn.SourceModule` from source code.
@@ -290,7 +290,7 @@ def tvm_home() -> Path:
         return tvm_path.resolve()
 
     @staticmethod
-    def get_includes(tvm_pkg: Optional[List[str]] = None) -> List[Path]:
+    def get_includes(tvm_pkg: list[str] | None = None) -> list[Path]:
         """Returns the default include paths according to `tvm_home()`.
         By default, it includes TVM, DLPack. With `tvm_pkg` provided, it also
         includes the specified package under `tvm_home/3rdparty`.
@@ -323,8 +323,8 @@ def get_includes(tvm_pkg: Optional[List[str]] = None) -> List[Path]:
     @staticmethod
     def get_compile_options(
         source_format: str,
-        tvm_pkg: Optional[List[str]] = None,
-    ) -> List[str]:
+        tvm_pkg: list[str] | None = None,
+    ) -> list[str]:
         """Returns the default compile options depending on `source_format`, including the default
         inlcude paths w.r.t. `tvm_home()`, and by default,
         it uses "-O3" and "-std=c++17".
diff --git a/python/tvm/relax/frontend/nn/llm/kv_cache.py b/python/tvm/relax/frontend/nn/llm/kv_cache.py
index 21d238b7c264..4e48eefef4f3 100644
--- a/python/tvm/relax/frontend/nn/llm/kv_cache.py
+++ b/python/tvm/relax/frontend/nn/llm/kv_cache.py
@@ -21,7 +21,7 @@
 # pylint: disable=too-many-statements,too-many-lines,too-many-arguments,invalid-name
 import enum
 import math
-from typing import Any, Dict, List, Literal, Optional, Tuple, Union
+from typing import Any, Literal
 
 import tvm
 from tvm import relax as rx
@@ -105,7 +105,7 @@ class RopeMode(enum.IntEnum):
 class PagedKVCache(Object):  # pylint: disable=too-few-public-methods
     """The Paged KV Cache used in LLM batching for efficient attention computation."""
 
-    extern_mods: List[tvm.runtime.Module] = []
+    extern_mods: list[tvm.runtime.Module] = []
 
     def attention_with_fused_qkv(
         self,
@@ -151,7 +151,7 @@ def self_attention(  # pylint: disable=too-many-locals
         k: Tensor,
         v: Tensor,
         sm_scale: float,
-    ) -> Tuple[Tensor, Tensor]:
+    ) -> tuple[Tensor, Tensor]:
         """Fine-grained API that computes ragged self attention with Q/K/V data."""
         # pylint: disable=protected-access
         b, s, h_qo, d_qk = q._expr.struct_info.shape
@@ -189,7 +189,7 @@ def cross_attention(
         q: Tensor,
         v_head_dim: int,
         sm_scale: float,
-    ) -> Tuple[Tensor, Tensor]:
+    ) -> tuple[Tensor, Tensor]:
         """Fine-grained API that computes paged cross attention with Q and in-cache KV data."""
         # pylint: disable=protected-access
         b, s, h_qo, d_qk = q._expr.struct_info.shape
@@ -238,7 +238,7 @@ def merge_attn_output_inplace(
         lse_self_attn: Tensor,
         o_cross_attn: Tensor,
         lse_cross_attn: Tensor,
-    ) -> Tuple[Tensor, Tensor]:
+    ) -> tuple[Tensor, Tensor]:
         """Fine-grained API that merges the attention output from two sources.
         The first two tensors will be inplace updated.
         """
@@ -299,9 +299,9 @@ def get_query_positions(self, total_length: tir.PrimExpr) -> Tensor:
 
 
 def _prepare_yarn_rope_scaling(
-    rope_scaling: Optional[Dict[str, Any]],
-    rope_theta: Optional[float],
-) -> Optional[Dict[str, Any]]:
+    rope_scaling: dict[str, Any] | None,
+    rope_theta: float | None,
+) -> dict[str, Any] | None:
     """Ensure Yarn-specific scaling configs include the theta metadata."""
     if rope_scaling is None:
         return None
@@ -320,7 +320,7 @@ class FlashInferPagedKVCache(PagedKVCache):  # pylint: disable=too-few-public-me
 
     def __init__(  # pylint: disable=too-many-locals
         self,
-        attn_kind: Union[Literal["mha", "mla"], List[Literal["mha", "mla", "mha_sliding"]]],
+        attn_kind: Literal["mha", "mla"] | list[Literal["mha", "mla", "mha_sliding"]],
         max_batch_size: tir.Var,
         max_total_seq_len: tir.Var,
         prefill_chunk_size: tir.Var,
@@ -337,7 +337,7 @@ def __init__(  # pylint: disable=too-many-locals
         rope_mode: RopeMode,
         rope_scale: int,
         rope_theta: int,
-        rope_scaling: Dict[str, Any],
+        rope_scaling: dict[str, Any],
         rope_ext_factors: rx.Expr,
         rotary_dim: int,
         enable_disaggregation: bool,
@@ -392,7 +392,7 @@ def __init__(  # pylint: disable=too-many-locals
         assert rope_mode != RopeMode.INLINE, "FlashInfer RoPE does not support inline mode."
         rope_scaling = _prepare_yarn_rope_scaling(rope_scaling, rope_theta)
 
-        attn_kind_single = attn_kind[0] if isinstance(attn_kind, List) else attn_kind
+        attn_kind_single = attn_kind[0] if isinstance(attn_kind, list) else attn_kind
         if attn_kind_single == "mha_sliding":
             attn_kind_single = "mha"
         flashinfer_prefill_mods = rx.backend.cuda.flashinfer.gen_flashinfer_prefill_module(
@@ -454,7 +454,7 @@ def __init__(  # pylint: disable=too-many-locals
         if attn_kind_single == "mla":
             attn_merge_functions.append(bb.add_func(_merge_state_inplace(num_attention_heads, mla_original_v_head_dim, dtype, target, "tir_attention_merge_state_mla"), "tir_attention_merge_state_mla"))
 
-        if isinstance(attn_kind, List):
+        if isinstance(attn_kind, list):
             attn_kind = [int(getattr(AttnKind, layer_kind.upper())) for layer_kind in attn_kind]
         else:
             attn_kind = [int(getattr(AttnKind, attn_kind.upper())) for _ in range(num_hidden_layers)]
@@ -508,7 +508,7 @@ class TIRPagedKVCache(PagedKVCache):  # pylint: disable=too-few-public-methods
 
     def __init__(  # pylint: disable=too-many-locals
         self,
-        attn_kind: Union[Literal["mha", "mla"], List[Literal["mha", "mla", "mha_sliding"]]],
+        attn_kind: Literal["mha", "mla"] | list[Literal["mha", "mla", "mha_sliding"]],
         max_batch_size: tir.Var,
         max_total_seq_len: tir.Var,
         prefill_chunk_size: tir.Var,
@@ -525,7 +525,7 @@ def __init__(  # pylint: disable=too-many-locals
         rope_mode: RopeMode,
         rope_scale: int,
         rope_theta: int,
-        rope_scaling: Dict[str, Any],
+        rope_scaling: dict[str, Any],
         rope_ext_factors: rx.Expr,
         rotary_dim: int,
         enable_disaggregation: bool,
@@ -580,10 +580,10 @@ def __init__(  # pylint: disable=too-many-locals
             The target to build the model to.
         """
         rope_scaling = _prepare_yarn_rope_scaling(rope_scaling, rope_theta)
-        attn_kind_single = attn_kind[0] if isinstance(attn_kind, List) else attn_kind
+        attn_kind_single = attn_kind[0] if isinstance(attn_kind, list) else attn_kind
         if attn_kind_single == "mha_sliding":
             attn_kind_single = "mha"
-        if isinstance(attn_kind, List):
+        if isinstance(attn_kind, list):
             attn_kind = [int(getattr(AttnKind, layer_kind.upper())) for layer_kind in attn_kind]
         else:
             attn_kind = [
@@ -858,9 +858,9 @@ def _rope(
     rotary_dim: int,
     theta: tir.Var,
     scale: tir.Var,
-    indices: Tuple[tir.Var, ...],
+    indices: tuple[tir.Var, ...],
     qkv_dtype: str,
-    rope_scaling: Dict[str, Any],
+    rope_scaling: dict[str, Any],
 ):
     d = indices[-1]
     cos_freq, sin_freq, var_map = switch_rope_freq_func(rope_scaling)(
@@ -922,7 +922,7 @@ def _get_seq_offset(pos, seq_id, length_info, sliding_window):
 
 
 def _attention_prefill_cpu(
-    h_kv, h_q, d, dtype, sliding_window: bool, rope_scaling: Dict[str, Any], page_size: int = 16
+    h_kv, h_q, d, dtype, sliding_window: bool, rope_scaling: dict[str, Any], page_size: int = 16
 ):
     global_symbol = "batch_prefill_paged_kv_cpu"
     if sliding_window:
@@ -1227,7 +1227,7 @@ def _attention_prefill(
     d,
     dtype,
     sliding_window: bool,
-    rope_scaling: Dict[str, Any],
+    rope_scaling: dict[str, Any],
     target: Target,
     page_size: int = 16,
 ):
@@ -1529,7 +1529,7 @@ def _attention_decode_cpu(
     head_dim,
     qkv_dtype,
     sliding_window: bool,
-    rope_scaling: Dict[str, Any],
+    rope_scaling: dict[str, Any],
     page_size: int = 16,
 ):
     H_qo = num_qo_heads
@@ -1685,7 +1685,7 @@ def _attention_decode(
     head_dim,
     qkv_dtype,
     sliding_window: bool,
-    rope_scaling: Dict[str, Any],
+    rope_scaling: dict[str, Any],
     target: Target,
     page_size: int = 16,
 ):
@@ -1977,7 +1977,7 @@ def merge_state_inplace_cpu(
 
 
 def _merge_state_inplace(
-    num_heads, head_dim, v_dtype, target: Target, global_symbol: Optional[str] = None
+    num_heads, head_dim, v_dtype, target: Target, global_symbol: str | None = None
 ):
     v_dtype_bytes = 2
     VEC_SIZE = min(max(8 // v_dtype_bytes, head_dim // 32), 4)
@@ -2305,7 +2305,7 @@ def batch_sequence_prefill_kv(  # pylint: disable=too-many-branches
     return sch.mod["main"].with_attr("tir.is_scheduled", True)
 
 
-def _attention_prefill_ragged_cpu(h_kv, h_q, d_qk, d_v, dtype, rope_scaling: Dict[str, Any]):
+def _attention_prefill_ragged_cpu(h_kv, h_q, d_qk, d_v, dtype, rope_scaling: dict[str, Any]):
     group_size = h_q // h_kv
 
     # fmt: off
@@ -2437,7 +2437,7 @@ def batch_prefill_ragged_kv(  # pylint: disable=too-many-branches
 
 
 def _attention_prefill_ragged(
-    h_kv, h_q, d_qk, d_v, dtype, rope_scaling: Dict[str, Any], target: Target
+    h_kv, h_q, d_qk, d_v, dtype, rope_scaling: dict[str, Any], target: Target
 ):
     # pylint: disable=line-too-long
     (
diff --git a/python/tvm/relax/frontend/nn/llm/position_embedding.py b/python/tvm/relax/frontend/nn/llm/position_embedding.py
index e2a7801adda4..a0bfbbc5782d 100644
--- a/python/tvm/relax/frontend/nn/llm/position_embedding.py
+++ b/python/tvm/relax/frontend/nn/llm/position_embedding.py
@@ -18,8 +18,9 @@
 """Operators for positional embeddings, e.g. RoPE."""
 
 import math
+from collections.abc import Callable
 from functools import partial
-from typing import Any, Callable, Dict, Optional, Tuple, Union
+from typing import Any
 
 from tvm import tir
 from tvm.relax.frontend.nn import Tensor, op
@@ -166,7 +167,7 @@ def rope_freq_longrope(  # pylint: disable=too-many-arguments
     dtype: str,
     max_position_embeddings: int,
     original_max_position_embeddings: int,
-    ext_factors: Optional[T.Buffer] = None,
+    ext_factors: T.Buffer | None = None,
 ):
     """Compute the inverse frequency of RoPE for longrope scaling."""
     scale = max_position_embeddings / original_max_position_embeddings
@@ -189,7 +190,7 @@ def yarn_find_correction_dim(
     num_rotations: int,
     d: tir.Var,
     max_position_embeddings: int,
-    inv_theta_log_scale: Optional[Union[float, tir.PrimExpr]] = None,
+    inv_theta_log_scale: float | tir.PrimExpr | None = None,
 ):
     """Inverse dim formula to find dim based on number of rotations"""
     return (
@@ -202,7 +203,7 @@ def yarn_find_correction_range(
     high_rot: int,
     d: tir.Var,
     max_position_embeddings: int,
-    inv_theta_log_scale: Optional[Union[float, tir.PrimExpr]] = None,
+    inv_theta_log_scale: float | tir.PrimExpr | None = None,
 ):
     """Find the correction range based on the number of rotations"""
     low = yarn_find_correction_dim(
@@ -218,13 +219,13 @@ def rope_freq_yarn(
     s: tir.Var,
     d: tir.Var,
     d_range: int,
-    theta: Union[float, tir.PrimExpr],
+    theta: float | tir.PrimExpr,
     dtype: str,
     original_max_position_embeddings: int,
     scaling_factor: float,
     beta_fast: int,
     beta_slow: int,
-    inv_theta_log_scale: Optional[Union[float, tir.PrimExpr]] = None,
+    inv_theta_log_scale: float | tir.PrimExpr | None = None,
 ):  # pylint: disable=too-many-arguments, too-many-locals
     """Compute the inverse frequency of RoPE for yarn RoPE scaling."""
 
@@ -252,7 +253,7 @@ def rope_freq_yarn(
     return cos_freq, sin_freq, {freq_var: freq}
 
 
-def switch_rope_freq_func(rope_scaling: Dict[str, Any]) -> Callable:
+def switch_rope_freq_func(rope_scaling: dict[str, Any]) -> Callable:
     """Return the RoPE inverse frequency computation function based
     on the given RoPE scaling.
     """
@@ -306,9 +307,9 @@ def llama_rope(  # pylint: disable=too-many-arguments
     scale: float,
     num_q_heads: int,
     num_kv_heads: int,
-    rope_scaling: Dict[str, Any],
-    rotary_dim: Optional[int] = None,
-) -> Tuple[Tensor, Tensor, Tensor]:
+    rope_scaling: dict[str, Any],
+    rotary_dim: int | None = None,
+) -> tuple[Tensor, Tensor, Tensor]:
     """Llama-style RoPE. Given a fused QKV tensor, it returns three tensors, Q, K, and V, where Q
     and K are rotated by RoPE while V remains unchanged.
 
@@ -445,8 +446,8 @@ def llama_rope_with_position_map(  # pylint: disable=too-many-arguments
     num_q_heads: int,
     num_kv_heads: int,
     dtype: str,
-    rope_scaling: Dict[str, Any],
-    rotary_dim: Optional[int] = None,
+    rope_scaling: dict[str, Any],
+    rotary_dim: int | None = None,
 ):
     """Return the TIR function that computes Llama-style RoPE with q position map.
 
@@ -493,7 +494,7 @@ def _rope(  # pylint: disable=too-many-arguments
         h: tir.Var,
         d: tir.Var,
         pos: tir.Var,
-        ext_factors: Optional[T.Buffer] = None,
+        ext_factors: T.Buffer | None = None,
     ):
         kwargs = {}
         if ext_factors:
@@ -671,8 +672,8 @@ def llama4_rope_with_position_map(  # pylint: disable=too-many-arguments
     num_q_heads: int,
     num_kv_heads: int,
     dtype: str,
-    rope_scaling: Dict[str, Any],
-    rotary_dim: Optional[int] = None,
+    rope_scaling: dict[str, Any],
+    rotary_dim: int | None = None,
 ):
     """Return the TIR function that computes Llama-style RoPE with q position map.
 
@@ -719,7 +720,7 @@ def _rope(  # pylint: disable=too-many-arguments
         h: tir.Var,
         d: tir.Var,
         pos: tir.Var,
-        ext_factors: Optional[T.Buffer] = None,
+        ext_factors: T.Buffer | None = None,
     ):
         kwargs = {}
         if ext_factors:
diff --git a/python/tvm/relax/frontend/nn/llm/tree_attn.py b/python/tvm/relax/frontend/nn/llm/tree_attn.py
index 433ce75928c4..abe672373630 100644
--- a/python/tvm/relax/frontend/nn/llm/tree_attn.py
+++ b/python/tvm/relax/frontend/nn/llm/tree_attn.py
@@ -20,7 +20,7 @@
 """Operators for tree attention."""
 
 import math
-from typing import Any, Dict, Tuple
+from typing import Any
 
 from tvm import s_tir, tir
 from tvm.runtime import DataType
@@ -43,9 +43,9 @@ def _rope(
     rotary_dim: int,
     theta: tir.Var,
     scale: tir.Var,
-    indices: Tuple[tir.Var, ...],
+    indices: tuple[tir.Var, ...],
     qkv_dtype: str,
-    rope_scaling: Dict[str, Any],
+    rope_scaling: dict[str, Any],
 ):
     d = indices[-1]
     cos_freq, sin_freq, var_map = switch_rope_freq_func(rope_scaling)(
@@ -91,7 +91,7 @@ def _declare_length_info(var_length_info, batch_size, sliding_window, elem_offse
     )
 
 
-def tree_attn_cpu(h_kv, h_q, d, dtype, rope_scaling: Dict[str, Any]):
+def tree_attn_cpu(h_kv, h_q, d, dtype, rope_scaling: dict[str, Any]):
     """Generate tree attention kernel for batched tree attention.
 
     Parameters
@@ -289,7 +289,7 @@ def batch_tree_attn(  # pylint: disable=too-many-branches,line-too-long
     return batch_tree_attn
 
 
-def tree_attn(h_kv, h_q, d, dtype, rope_scaling: Dict[str, Any], target: Target):  # pylint: disable=unused-argument
+def tree_attn(h_kv, h_q, d, dtype, rope_scaling: dict[str, Any], target: Target):  # pylint: disable=unused-argument
     """Generate tree attention kernel for batched tree attention.
 
     Parameters
@@ -651,7 +651,7 @@ def apply_to_md(sch, block):
     return sch.mod["main"].with_attr("tir.is_scheduled", True)
 
 
-def tree_attn_with_paged_kv_cache_cpu(h_kv, h_q, d, dtype, rope_scaling: Dict[str, Any]):
+def tree_attn_with_paged_kv_cache_cpu(h_kv, h_q, d, dtype, rope_scaling: dict[str, Any]):
     """Generate tree attention kernel for batched tree attention with paged key-value cache.
 
     Parameters
@@ -847,7 +847,7 @@ def tree_attn_paged_kv_cpu(
 
 
 def tree_attn_with_paged_kv_cache(
-    h_kv, h_q, d, dtype, rope_scaling: Dict[str, Any], target: Target
+    h_kv, h_q, d, dtype, rope_scaling: dict[str, Any], target: Target
 ):
     """Generate tree attention kernel for batched tree attention with paged key-value cache.
 
diff --git a/python/tvm/relax/frontend/nn/modules.py b/python/tvm/relax/frontend/nn/modules.py
index c234f8a02cbb..ee16f25d2881 100644
--- a/python/tvm/relax/frontend/nn/modules.py
+++ b/python/tvm/relax/frontend/nn/modules.py
@@ -18,7 +18,7 @@
 # ruff: noqa: RUF005
 """Builtin Modules."""
 
-from typing import List, Optional, Sequence, Union
+from collections.abc import Sequence
 
 from tvm import relax as rx
 from tvm import tir
@@ -33,23 +33,23 @@ class IOEffect(Effect):
     debug breakpoints, etc.
     """
 
-    effect: Optional[rx.Var]
+    effect: rx.Var | None
 
     def __init__(self):
         self.effect = None
 
-    def emit_init(self, name_hint, builder: rx.BlockBuilder) -> List[rx.DataflowVar]:
+    def emit_init(self, name_hint, builder: rx.BlockBuilder) -> list[rx.DataflowVar]:
         return [builder.emit(rx.op.null_value(), f"{name_hint}.io")]
 
-    def create(self, name_hint: str) -> List[rx.Var]:
+    def create(self, name_hint: str) -> list[rx.Var]:
         assert self.effect is None
         effect = rx.Var(f"{name_hint}.io", struct_info=rx.ObjectStructInfo())
         return [effect]
 
-    def set_state(self, state_vars: List[rx.Var]) -> None:
+    def set_state(self, state_vars: list[rx.Var]) -> None:
         (self.effect,) = state_vars
 
-    def finalize(self) -> List[rx.Var]:
+    def finalize(self) -> list[rx.Var]:
         result = self.effect
         self.effect = None
         return [result]
@@ -101,11 +101,11 @@ class Linear(Module):
 
     def __init__(
         self,
-        in_features: Union[int, str, tir.PrimExpr],
-        out_features: Union[int, str, tir.PrimExpr],
+        in_features: int | str | tir.PrimExpr,
+        out_features: int | str | tir.PrimExpr,
         bias: bool = True,
-        dtype: Optional[str] = None,
-        out_dtype: Optional[str] = None,
+        dtype: str | None = None,
+        out_dtype: str | None = None,
     ):
         super().__init__()
         self.in_features = in_features
@@ -140,7 +140,7 @@ def forward(self, x: Tensor) -> Tensor:
             x = x + self.bias
         return x
 
-    def to(self, dtype: Optional[str] = None) -> None:
+    def to(self, dtype: str | None = None) -> None:
         """
         Override to() such that we do not convert bias if there is `out_dtype`.
         Otherwise, we might run into dtype mismatch when computing `x + self.bias`
@@ -168,7 +168,7 @@ def __init__(
         dilation: int = 1,
         groups: int = 1,
         bias: bool = True,
-        dtype: Optional[str] = None,
+        dtype: str | None = None,
     ) -> None:
         super().__init__()
         self.in_channels = in_channels
@@ -220,13 +220,13 @@ def __init__(  # pylint: disable=too-many-arguments
         self,
         in_channels: int,
         out_channels: int,
-        kernel_size: Union[List[int], int],
+        kernel_size: list[int] | int,
         stride: int = 1,
         padding: int = 0,
         dilation: int = 1,
         groups: int = 1,
         bias: bool = True,
-        dtype: Optional[str] = None,
+        dtype: str | None = None,
         data_layout: str = "NCHW",
     ):
         super().__init__()
@@ -294,13 +294,13 @@ def __init__(  # pylint: disable=too-many-arguments
         self,
         in_channels: int,
         out_channels: int,
-        kernel_size: Union[List[int], int],
-        stride: Union[List[int], int] = 1,
-        padding: Union[List[int], int] = 0,
+        kernel_size: list[int] | int,
+        stride: list[int] | int = 1,
+        padding: list[int] | int = 0,
         dilation: int = 1,
         groups: int = 1,
         bias: bool = True,
-        dtype: Optional[str] = None,
+        dtype: str | None = None,
         data_layout: str = "NCDHW",
     ):
         super().__init__()
@@ -375,7 +375,7 @@ def __init__(
         dilation: int = 1,
         groups: int = 1,
         bias: bool = True,
-        dtype: Optional[str] = None,
+        dtype: str | None = None,
     ) -> None:
         super().__init__()
         self.in_channels = in_channels
@@ -434,9 +434,9 @@ class LayerNorm(Module):
     def __init__(
         self,
         normalized_shape: int,
-        eps: Optional[float] = 1e-5,
+        eps: float | None = 1e-5,
         elementwise_affine: bool = True,
-        dtype: Optional[str] = None,
+        dtype: str | None = None,
     ) -> None:
         super().__init__()
         self.normalized_shape = normalized_shape
@@ -480,10 +480,10 @@ class RMSNorm(Module):
     def __init__(
         self,
         hidden_size: int,
-        axes: Union[int, List[int]],
+        axes: int | list[int],
         epsilon: float = 1e-5,
         bias: bool = True,
-        dtype: Optional[str] = None,
+        dtype: str | None = None,
     ):
         super().__init__()
         self.epsilon = epsilon
@@ -525,7 +525,7 @@ def __init__(
         num_channels: int,
         eps: float = 1e-5,
         affine: bool = True,
-        dtype: Optional[str] = None,
+        dtype: str | None = None,
     ):
         super().__init__()
         self.num_groups = num_groups
@@ -538,7 +538,7 @@ def __init__(
             self.weight = None
             self.bias = None
 
-    def forward(self, x: Tensor, channel_axis: int = 1, axes: Optional[List[int]] = None):
+    def forward(self, x: Tensor, channel_axis: int = 1, axes: list[int] | None = None):
         """
         Forward method for group norm layer.
 
@@ -568,15 +568,15 @@ class KVCache(Effect):
     """
 
     init_seq_len: int
-    unit_shape: List[int]
+    unit_shape: list[int]
     dtype: str
-    cache: Optional[rx.Var]
+    cache: rx.Var | None
 
     def __init__(
         self,
         init_seq_len: int,
         unit_shape: Sequence[int],
-        dtype: Optional[str] = None,
+        dtype: str | None = None,
     ):
         if dtype is None:
             dtype = get_default_dtype()
@@ -612,7 +612,7 @@ def emit_init(self, name_hint: str, bb: rx.BlockBuilder):  # pylint: disable=arg
             )
         ]
 
-    def create(self, name_hint: str) -> List[rx.Var]:
+    def create(self, name_hint: str) -> list[rx.Var]:
         """
         Create the implicit inputs to a relax.Function that represents the KVCache effect.
 
@@ -629,10 +629,10 @@ def create(self, name_hint: str) -> List[rx.Var]:
         cache = rx.Var(name_hint, struct_info=rx.ObjectStructInfo())
         return [cache]
 
-    def set_state(self, state_vars: List[rx.Var]) -> None:
+    def set_state(self, state_vars: list[rx.Var]) -> None:
         (self.cache,) = state_vars
 
-    def finalize(self) -> List[rx.Var]:
+    def finalize(self) -> list[rx.Var]:
         """
         Finalize the KVCache effect as the implicit return value of a relax.Function.
 
@@ -645,7 +645,7 @@ def finalize(self) -> List[rx.Var]:
         self.cache = None
         return [result]
 
-    def to(self, dtype: Optional[str] = None) -> None:
+    def to(self, dtype: str | None = None) -> None:
         """
         Convert the KVCache effect to specific dtype.
 
@@ -714,9 +714,9 @@ class Embedding(Module):
 
     def __init__(
         self,
-        num: Union[int, str, tir.PrimExpr],
-        dim: Union[int, str, tir.PrimExpr],
-        dtype: Optional[str] = None,
+        num: int | str | tir.PrimExpr,
+        dim: int | str | tir.PrimExpr,
+        dtype: str | None = None,
     ):
         self.num = num
         self.dim = dim
@@ -758,9 +758,9 @@ def __init__(
         in_channels: int,
         time_embed_dim: int,
         act_fn: str = "silu",
-        out_dim: Optional[int] = None,
-        post_act_fn: Optional[str] = None,
-        cond_proj_dim: Optional[int] = None,
+        out_dim: int | None = None,
+        post_act_fn: str | None = None,
+        cond_proj_dim: int | None = None,
     ):
         self.linear_1 = Linear(in_channels, time_embed_dim)
 
@@ -785,7 +785,7 @@ def __init__(
             assert self.post_act == "silu", "Only SiLU post-activation supported."
             self.post_act = SiLU()
 
-    def forward(self, sample: Tensor, condition: Optional[Tensor] = None):
+    def forward(self, sample: Tensor, condition: Tensor | None = None):
         """
         Forward method for TimestepEmbedding layer.
 
@@ -864,11 +864,11 @@ class Attention(Module):
     def __init__(
         self,
         query_dim: int,
-        cross_attention_dim: Optional[int] = None,
+        cross_attention_dim: int | None = None,
         heads: int = 8,
         dim_head: int = 64,
         bias: bool = False,
-        norm_num_groups: Optional[int] = None,
+        norm_num_groups: int | None = None,
         out_bias: bool = True,
         scale_qk: bool = True,
     ):
@@ -900,8 +900,8 @@ def __init__(
     def forward(
         self,
         hidden_states: Tensor,
-        encoder_hidden_states: Optional[Tensor] = None,
-        attention_mask: Optional[Tensor] = None,
+        encoder_hidden_states: Tensor | None = None,
+        attention_mask: Tensor | None = None,
         **cross_attention_kwargs,
     ):
         """
diff --git a/python/tvm/relax/frontend/nn/op.py b/python/tvm/relax/frontend/nn/op.py
index 03b05492d48e..3c1fa442b007 100644
--- a/python/tvm/relax/frontend/nn/op.py
+++ b/python/tvm/relax/frontend/nn/op.py
@@ -20,7 +20,8 @@
 
 import inspect
 import math
-from typing import Any, Callable, Dict, List, Optional, Sequence, Tuple, TypeVar, Union
+from collections.abc import Callable, Sequence
+from typing import Any, TypeVar
 
 import numpy as np
 
@@ -33,7 +34,7 @@
 from ...block_builder import BlockBuilder
 from .core import Tensor, get_default_dtype, wrap_nested
 
-IntExpr = Union[int, _tir.PrimExpr]
+IntExpr = int | _tir.PrimExpr
 
 
 def unsqueeze(x: Tensor, dim: int, name: str = "unsqueeze") -> Tensor:
@@ -56,7 +57,7 @@ def unsqueeze(x: Tensor, dim: int, name: str = "unsqueeze") -> Tensor:
     return wrap_nested(_op.expand_dims(x._expr, dim), name)
 
 
-def concat(x: List[Tensor], dim: int, name: str = "concat") -> Tensor:
+def concat(x: list[Tensor], dim: int, name: str = "concat") -> Tensor:
     """Concatenate a list of tensors along an axis.
 
     Parameters
@@ -214,7 +215,7 @@ def chunk(x: Tensor, chunks: int, dim: int = 0, name: str = "chunk") -> Tensor:
 
 def sum(
     x: Tensor,
-    axis: Optional[Union[int, List[int]]] = None,
+    axis: int | list[int] | None = None,
     keepdims: bool = False,
     name: str = "sum",
 ) -> Tensor:
@@ -248,7 +249,7 @@ def sum(
 
 def max(
     x: Tensor,
-    axis: Optional[Union[int, List[int]]] = None,
+    axis: int | list[int] | None = None,
     keepdims: bool = False,
     name: str = "max",
 ) -> Tensor:
@@ -282,7 +283,7 @@ def max(
 
 def min(
     x: Tensor,
-    axis: Optional[Union[int, List[int]]] = None,
+    axis: int | list[int] | None = None,
     keepdims: bool = False,
     name: str = "min",
 ) -> Tensor:
@@ -314,7 +315,7 @@ def min(
     return wrap_nested(_op.min(x._expr, axis, keepdims), name)
 
 
-def matmul(a: Tensor, b: Tensor, out_dtype: Optional[str] = None, name: str = "matmul") -> Tensor:
+def matmul(a: Tensor, b: Tensor, out_dtype: str | None = None, name: str = "matmul") -> Tensor:
     """General matrix multiplication of two tensors, with broadcasting on batched dimensions.
 
     The semantics and output shape deduction rule is specified as
@@ -328,7 +329,7 @@ def matmul(a: Tensor, b: Tensor, out_dtype: Optional[str] = None, name: str = "m
     b : Tensor
         The second input tensor.
 
-    out_dtype: Optional[Union[str, DataType]]
+    out_dtype: Optional[str | DataType]
         The data type of the matmul result.
         When it is not specified, the output dtype will be the same as input dtype.
 
@@ -352,11 +353,11 @@ def matmul(a: Tensor, b: Tensor, out_dtype: Optional[str] = None, name: str = "m
 def conv1d(
     x: Tensor,
     weight: Tensor,
-    bias: Optional[Tensor] = None,
-    stride: Optional[Union[int, Tuple]] = 1,
-    padding: Optional[Union[int, Tuple, str]] = 0,
-    dilation: Optional[Union[int, Tuple]] = 1,
-    groups: Optional[int] = 1,
+    bias: Tensor | None = None,
+    stride: int | tuple | None = 1,
+    padding: int | tuple | str | None = 0,
+    dilation: int | tuple | None = 1,
+    groups: int | None = 1,
     name: str = "conv1d",
 ) -> Tensor:
     r"""1D convolution.
@@ -395,14 +396,14 @@ def conv1d(
     bias : Optional[Tensor]
         Optional bias tensor of shape [O].
 
-    strides : Optional[Union[int, Tuple]]
+    strides : Optional[int | Tuple]
         The strides of convolution. It is required to have length 1.
 
-    padding : Optional[Union[int, Tuple, str]]
+    padding : Optional[int | Tuple | str]
         The padding of convolution on both sides of inputs before convolution.
         It is required to have length either 1 or 2.
 
-    dilation : Optional[Union[int, Tuple]]
+    dilation : Optional[int | Tuple]
         Specifies the dilation rate to be used for dilated convolution.
         It is required to have length 1.
 
@@ -435,12 +436,12 @@ def conv1d(
 def conv2d(
     x: Tensor,
     weight: Tensor,
-    bias: Optional[Tensor] = None,
-    stride: Optional[Union[int, Tuple]] = 1,
-    padding: Optional[Union[int, Tuple, str]] = 0,
-    dilation: Optional[Union[int, Tuple]] = 1,
-    groups: Optional[int] = 1,
-    data_layout: Optional[str] = "NCHW",
+    bias: Tensor | None = None,
+    stride: int | tuple | None = 1,
+    padding: int | tuple | str | None = 0,
+    dilation: int | tuple | None = 1,
+    groups: int | None = 1,
+    data_layout: str | None = "NCHW",
     name: str = "conv2d",
 ) -> Tensor:
     """Applies a 2D convolution over an input image composed of sevaral input planes
@@ -456,14 +457,14 @@ def conv2d(
     bias : Optional[Tensor]
         Optional bias tensor of shape [O].
 
-    stride : Optional[Union[int, Tuple]]
+    stride : Optional[int | Tuple]
         The stride of the convolving kernel. Can be a single number
         or tuple of (sH, sW).
 
-    padding : Optional[[Union[int, Tuple]]]
+    padding : Optional[[int | Tuple]]
         Implicit paddings on both sides of the input.
 
-    dilation : Optional[Union[int, Tuple]]
+    dilation : Optional[int | Tuple]
         The spacing between kernel elements. Can be a single number of tuple (dH, dW).
 
     groups : Optional[int]
@@ -503,12 +504,12 @@ def conv2d(
 def conv3d(
     x: Tensor,
     weight: Tensor,
-    bias: Optional[Tensor] = None,
-    stride: Optional[Union[int, Tuple]] = 1,
-    padding: Optional[Union[int, Tuple, str]] = 0,
-    dilation: Optional[Union[int, Tuple]] = 1,
-    groups: Optional[int] = 1,
-    data_layout: Optional[str] = "NCDHW",
+    bias: Tensor | None = None,
+    stride: int | tuple | None = 1,
+    padding: int | tuple | str | None = 0,
+    dilation: int | tuple | None = 1,
+    groups: int | None = 1,
+    data_layout: str | None = "NCDHW",
     name: str = "conv3d",
 ) -> Tensor:
     """Applies a 3D convolution over an input image composed of sevaral input planes
@@ -524,14 +525,14 @@ def conv3d(
     bias : Optional[Tensor]
         Optional bias tensor of shape [O].
 
-    stride : Optional[Union[int, Tuple]]
+    stride : Optional[int | Tuple]
         The stride of the convolving kernel. Can be a single number
         or tuple of (sD, sH, sW).
 
-    padding : Optional[[Union[int, Tuple]]]
+    padding : Optional[[int | Tuple]]
         Implicit paddings on both sides of the input.
 
-    dilation : Optional[Union[int, Tuple]]
+    dilation : Optional[int | Tuple]
         The spacing between kernel elements. Can be a single number of tuple (dD, dH, dW).
 
     groups : Optional[int]
@@ -571,12 +572,12 @@ def conv3d(
 def conv1d_transpose(
     x: Tensor,
     weight: Tensor,
-    bias: Optional[Tensor] = None,
-    stride: Optional[Union[int, Tuple[int]]] = 1,
-    padding: Optional[Union[int, Tuple[int, ...]]] = 0,
-    output_padding: Optional[Union[int, Tuple[int]]] = 0,
-    dilation: Optional[Union[int, Tuple]] = 1,
-    groups: Optional[int] = 1,
+    bias: Tensor | None = None,
+    stride: int | tuple[int] | None = 1,
+    padding: int | tuple[int, ...] | None = 0,
+    output_padding: int | tuple[int] | None = 0,
+    dilation: int | tuple | None = 1,
+    groups: int | None = 1,
     name: str = "conv1d_transpose",
 ) -> Tensor:
     """1D transposed convolution operator.
@@ -625,7 +626,7 @@ def conv1d_transpose(
     out_layout : Optional[str]
         Layout of the output. If not specified, it is the same as data_layout
 
-    out_dtype : Optional[Union[str, DataType]]
+    out_dtype : Optional[str | DataType]
         Specifies the output data type for mixed precision conv2d.
 
     Returns
@@ -726,7 +727,7 @@ def broadcast_to(x: Tensor, shape: Sequence[IntExpr], name: str = "broadcast_to"
     return wrap_nested(_op.broadcast_to(x._expr, shape), name)
 
 
-def permute_dims(x: Tensor, axes: Optional[List[int]] = None, name: Optional[str] = None) -> Tensor:
+def permute_dims(x: Tensor, axes: list[int] | None = None, name: str | None = None) -> Tensor:
     """Permutes the dimensions of an array.
 
     Parameters
@@ -793,7 +794,7 @@ def reshape(x: Tensor, shape: Sequence[IntExpr], name="reshape") -> Tensor:
     return wrap_nested(_op.reshape(x._expr, shape), name)
 
 
-def repeat(x: Tensor, repeats: int, axis: Optional[int] = None, name="repeat") -> Tensor:
+def repeat(x: Tensor, repeats: int, axis: int | None = None, name="repeat") -> Tensor:
     """Repeats elements of an array.
 
     Parameters
@@ -854,7 +855,7 @@ def squeeze(x: Tensor, axis: int = -1, name: str = "squeeze") -> Tensor:
     return wrap_nested(_op.squeeze(x._expr, axis), name)
 
 
-def take(x: Tensor, indices: Tensor, axis: Optional[int] = None, name="take") -> Tensor:
+def take(x: Tensor, indices: Tensor, axis: int | None = None, name="take") -> Tensor:
     """Take elements from a tensor along an axis.
     Its semantic is mostly similar to `numpy.take`
     (https://numpy.org/doc/stable/reference/generated/numpy.take.html),
@@ -979,7 +980,7 @@ def silu(x: Tensor, name: str = "silu") -> Tensor:
     return wrap_nested(_op.nn.silu(x._expr), name)
 
 
-def gelu(x: Tensor, approximate: Optional[str] = None, name: str = "gelu") -> Tensor:
+def gelu(x: Tensor, approximate: str | None = None, name: str = "gelu") -> Tensor:
     r"""Applies the Gaussian Error Linear Units function
 
     .. math::
@@ -1228,9 +1229,9 @@ def floor(x: Tensor, name: str = "floor") -> Tensor:
 
 def arange(
     start: int,
-    end: Optional[int] = None,
+    end: int | None = None,
     step: int = 1,
-    dtype: Optional[str] = "float32",
+    dtype: str | None = "float32",
     name: str = "arange",
 ) -> Tensor:
     r"""Construct a tensor with evenly spaced elements.
@@ -1261,7 +1262,7 @@ def arange(
     return wrap_nested(_op.arange(start, end, step, dtype), name)
 
 
-def permute(x: Tensor, axes: Optional[List[int]], name: str = "permute") -> Tensor:
+def permute(x: Tensor, axes: list[int] | None, name: str = "permute") -> Tensor:
     """Permutes the dimensions of the input tensor.
 
     Parameters
@@ -1305,9 +1306,9 @@ def negative(x: Tensor, name: str = "neg") -> Tensor:
 
 def layer_norm(
     x: Tensor,
-    normalized_shape: Union[int, List[int]],
-    weight: Optional[Tensor] = None,
-    bias: Optional[Tensor] = None,
+    normalized_shape: int | list[int],
+    weight: Tensor | None = None,
+    bias: Tensor | None = None,
     eps: float = 1e-5,
     name: str = "layer_norm",
 ) -> Tensor:
@@ -1387,7 +1388,7 @@ def layer_norm(
 def rms_norm(
     x: Tensor,
     weight: Tensor,
-    axes: Union[int, List[int]],
+    axes: int | list[int],
     epsilon: float = 1e-5,
     name: str = "rms_norm",
 ) -> Tensor:
@@ -1429,11 +1430,11 @@ def rms_norm(
 def group_norm(
     x: Tensor,
     num_groups: int,
-    weight: Optional[Tensor],
-    bias: Optional[Tensor],
+    weight: Tensor | None,
+    bias: Tensor | None,
     eps: float = 1e-5,
     channel_axis: int = 1,
-    axes: Optional[List[int]] = None,
+    axes: list[int] | None = None,
     name: str = "group_norm",
 ) -> Tensor:
     r"""
@@ -1544,9 +1545,9 @@ def full(
     result : Tensor
         The result tensor.
     """
-    if isinstance(fill_value, (_tir.FloatImm, _tir.IntImm)):
+    if isinstance(fill_value, _tir.FloatImm | _tir.IntImm):
         fill_value = rx.const(fill_value.value, dtype=dtype)
-    elif isinstance(fill_value, (int, float)):
+    elif isinstance(fill_value, int | float):
         fill_value = rx.const(fill_value, dtype=dtype)
     else:
         fill_value = fill_value._expr
@@ -1640,10 +1641,10 @@ def empty(
 
 def split(
     ary: Tensor,
-    indices_or_sections: Union[int, Sequence[int]],
+    indices_or_sections: int | Sequence[int],
     axis: int = 0,
     name: str = "split",
-) -> Tuple[Tensor, ...]:
+) -> tuple[Tensor, ...]:
     """Split an array into multiple sub-arrays.
 
     Parameters
@@ -1667,7 +1668,7 @@ def split(
 
 def pad(
     x: Tensor,
-    pad: List[int],
+    pad: list[int],
     mode: str = "constant",
     value: float = 0.0,
     name: str = "pad",
@@ -1809,9 +1810,9 @@ def scaled_dot_product_attention(
     query: Tensor,
     key: Tensor,
     value: Tensor,
-    attn_mask: Optional[Tensor] = None,
-    is_causal: Optional[bool] = False,
-    scale: Optional[float] = None,
+    attn_mask: Tensor | None = None,
+    is_causal: bool | None = False,
+    scale: float | None = None,
     name: str = "scaled_dot_product_attention",
 ):
     """
@@ -1849,13 +1850,13 @@ def scaled_dot_product_attention(
 
 def interpolate(
     x: Tensor,
-    size: Optional[Union[int, Tuple[int]]] = None,
-    scale_factor: Optional[Union[float, Tuple[float]]] = None,
+    size: int | tuple[int] | None = None,
+    scale_factor: float | tuple[float] | None = None,
     mode: str = "nearest",
-    align_corners: Optional[bool] = None,
-    recompute_scale_factor: Optional[bool] = None,
-    antialias: Optional[bool] = None,
-    data_layout: Optional[str] = "NCHW",
+    align_corners: bool | None = None,
+    recompute_scale_factor: bool | None = None,
+    antialias: bool | None = None,
+    data_layout: str | None = "NCHW",
     name: str = "interpolate",
 ):
     """Resize a tensor using the specified mode.
@@ -1895,7 +1896,7 @@ def interpolate(
         for i, dim in enumerate(data_layout):
             # Only upscale spatial dimensions.
             if dim not in ["N", "C"]:
-                if isinstance(scale_factor, (list, tuple)):
+                if isinstance(scale_factor, list | tuple):
                     size.append(int(x.shape[i] * scale_factor[len(size)]))
                 else:
                     size.append(int(x.shape[i] * scale_factor))
@@ -1993,9 +1994,9 @@ def ccl_broadcast_from_worker0(x: Tensor, name="broadcast_from_worker"):
 def tensor_expr_op(
     tensor_expr_func: Callable,
     name_hint: str,
-    args: List[Union[Tensor, _tir.Var, int]],
+    args: list[Tensor | _tir.Var | int],
     *,
-    attrs: Optional[Dict[str, Any]] = None,
+    attrs: dict[str, Any] | None = None,
 ):
     """Build the given tensor_expr_func with te.
 
@@ -2007,7 +2008,7 @@ def tensor_expr_op(
     name_hint : str
         Name hint.
 
-    args: List[Union[Tensor, _tir.Var]]
+    args: List[Tensor | _tir.Var]
         Arguments passed to the function.
 
     attrs: Optional[Dict[str, Any]]
@@ -2035,13 +2036,13 @@ def _convert(arg):
     )
 
 
-OutType = TypeVar("OutType", bound=Union[Tensor, Sequence[Tensor]])
+OutType = TypeVar("OutType", bound=Tensor | Sequence[Tensor])
 
 
 def tensor_ir_op(
     func: _tir.PrimFunc,
     name_hint: str,
-    args: Union[Tensor, Sequence[Union[Tensor, rx.ShapeExpr, _tir.PrimExpr]]],
+    args: Tensor | Sequence[Tensor | rx.ShapeExpr | _tir.PrimExpr],
     out: OutType,
 ) -> OutType:
     """Create a `call_tir` binding with given PrimFunc
@@ -2054,7 +2055,7 @@ def tensor_ir_op(
     name_hint : str
         Name hint.
 
-    args : Union[Tensor, Sequence[Union[Tensor, rx.ShapeExpr, _tir.PrimExpr]]]
+    args : Union[Tensor, Sequence[Tensor | rx.ShapeExpr | _tir.PrimExpr]]
         The arguments to pass to the PrimFunc.
 
     out : Union[Tensor, List[Tensor]]
@@ -2068,13 +2069,13 @@ def tensor_ir_op(
     from tvm import relax as rx  # pylint: disable=import-outside-toplevel
 
     call_tir_args, tir_vars = [], []
-    if not isinstance(args, (tuple, list)):
+    if not isinstance(args, tuple | list):
         args = [args]
 
     for arg in args:
         if isinstance(arg, Tensor):
             call_tir_args.append(arg._expr)
-        elif isinstance(arg, (rx.ShapeExpr, _tir.PrimExpr)):
+        elif isinstance(arg, rx.ShapeExpr | _tir.PrimExpr):
             tir_vars.append(arg)
         else:
             raise TypeError(
@@ -2102,8 +2103,8 @@ def tensor_ir_op(
 def tensor_ir_inplace_op(
     func: _tir.PrimFunc,
     name_hint: str,
-    args: Union[Tensor, Sequence[Union[Tensor, rx.ShapeExpr, _tir.PrimExpr]]],
-    inplace_indices: Union[int, List[int]],
+    args: Tensor | Sequence[Tensor | rx.ShapeExpr | _tir.PrimExpr],
+    inplace_indices: int | list[int],
     out: OutType,
 ) -> OutType:
     """Create a `call_tir_inplace` binding with given PrimFunc
@@ -2116,7 +2117,7 @@ def tensor_ir_inplace_op(
     name_hint : str
         Name hint.
 
-    args : Union[Tensor, Sequence[Union[Tensor, rx.ShapeExpr, _tir.PrimExpr]]]
+    args : Union[Tensor, Sequence[Tensor | rx.ShapeExpr | _tir.PrimExpr]]
         The arguments to pass to the PrimFunc.
 
     inplace_indices : Union[int, List[int]]
@@ -2138,13 +2139,13 @@ def tensor_ir_inplace_op(
     from tvm import relax as rx  # pylint: disable=import-outside-toplevel
 
     call_tir_args, tir_vars = [], []
-    if not isinstance(args, (tuple, list)):
+    if not isinstance(args, tuple | list):
         args = [args]
 
     for arg in args:
         if isinstance(arg, Tensor):
             call_tir_args.append(arg._expr)
-        elif isinstance(arg, (rx.ShapeExpr, _tir.PrimExpr)):
+        elif isinstance(arg, rx.ShapeExpr | _tir.PrimExpr):
             tir_vars.append(arg)
         else:
             raise TypeError(
@@ -2170,7 +2171,7 @@ def tensor_ir_inplace_op(
 
 def extern(
     name: str,
-    args: Sequence[Union[Tensor, _tir.PrimExpr, int, float, str]],
+    args: Sequence[Tensor | _tir.PrimExpr | int | float | str],
     out: OutType,
 ) -> OutType:
     """Invoke an extern function during runtime. The extern function must be registered with the "
@@ -2181,7 +2182,7 @@ def extern(
     name : str
         The name of the extern function to call.
 
-    args : Sequence[Union[Tensor, _tir.PrimExpr, int, float, str]]
+    args : Sequence[Tensor | _tir.PrimExpr | int | float | str]
         The arguments to pass to the extern function.
 
     out : Union[Tensor, List[Tensor]]
@@ -2205,7 +2206,7 @@ def _convert(arg, name: str):
             return rx.StringImm(arg)
         if isinstance(arg, _tir.PrimExpr):
             return rx.PrimValue(arg)
-        if isinstance(arg, (tuple, list)):
+        if isinstance(arg, tuple | list):
             return rx.Tuple([_convert(e, f"{name}_{i}") for i, e in enumerate(arg)])
         raise TypeError(f"Unsupported input type: {type(arg)}")
 
@@ -2223,8 +2224,8 @@ def _convert(arg, name: str):
 
 def debug_func(
     name: str,
-    *args: Union[Tensor, _tir.PrimExpr, int, float, str],
-    _line_info: Optional[str] = None,
+    *args: Tensor | _tir.PrimExpr | int | float | str,
+    _line_info: str | None = None,
 ):
     """Call a debug function during runtime. The debug function must be registered with the
     following type signature:
@@ -2240,7 +2241,7 @@ def debug_func(lineno: str, arg_0, arg_1, ...) -> None:
     name : str
         The name of the debug function to call.
 
-    *args : Union[Tensor, _tir.PrimExpr, int, float, str]
+    *args : Tensor | _tir.PrimExpr | int | float | str
         The arguments to pass to the debug function.
     """
     # pylint: disable=import-outside-toplevel
@@ -2463,9 +2464,9 @@ def where(condition: Tensor, x1: Tensor, x2: Tensor, name: str = "where") -> Ten
 
 def cumsum(
     data: Tensor,
-    axis: Optional[int] = None,
-    dtype: Optional[str] = None,
-    exclusive: Optional[bool] = None,
+    axis: int | None = None,
+    dtype: str | None = None,
+    exclusive: bool | None = None,
     name: str = "cumsum",
 ) -> Tensor:
     """Numpy style cumsum op. Return the cumulative inclusive sum of the elements along
@@ -2633,7 +2634,7 @@ def topk(
 def multinomial_from_uniform(
     prob: Tensor,
     uniform_sample: Tensor,
-    sample_indices: Optional[Tensor] = None,
+    sample_indices: Tensor | None = None,
     dtype: str = "int64",
     name: str = "multinomial_from_uniform",
 ):
@@ -2711,7 +2712,7 @@ def sample_top_p_top_k_from_sorted_prob(
     top_p: Tensor,
     top_k: Tensor,
     uniform_sample: Tensor,
-    sample_indices: Optional[Tensor] = None,
+    sample_indices: Tensor | None = None,
 ):
     """Samples indices from a sorted probability tensor based on top_p and top_k criteria.
 
diff --git a/python/tvm/relax/frontend/nn/spec.py b/python/tvm/relax/frontend/nn/spec.py
index d0b0fe705cac..f3c7275be5d7 100644
--- a/python/tvm/relax/frontend/nn/spec.py
+++ b/python/tvm/relax/frontend/nn/spec.py
@@ -23,8 +23,8 @@
     from .core import Module as nn_module_class
 
 ArgSpecType = typing.Union["Int", "Tensor"]
-MethodSpecType = typing.Union["MethodSpec", typing.Dict[str, ArgSpecType]]
-ModuleSpecType = typing.Union["ModuleSpec", typing.Dict[str, MethodSpecType]]
+MethodSpecType = typing.Union["MethodSpec", dict[str, ArgSpecType]]
+ModuleSpecType = typing.Union["ModuleSpec", dict[str, MethodSpecType]]
 SpecAny = typing.Union["Object", "Int", "Tensor", "Tuple"]
 
 
@@ -41,10 +41,10 @@ def __repr__(self) -> str:
 class Tensor:  # pylint: disable=too-few-public-methods
     """A tensor input with static ndim and dtype, but can have symbolic shapes."""
 
-    shape: typing.List[typing.Union[int, str]]
+    shape: list[int | str]
     dtype: str
 
-    def __init__(self, shape: typing.Sequence[typing.Union[int, str]], dtype: str) -> None:
+    def __init__(self, shape: typing.Sequence[int | str], dtype: str) -> None:
         self.shape = list(shape)
         self.dtype = dtype
 
@@ -56,9 +56,9 @@ def __repr__(self) -> str:
 class Object:  # pylint: disable=too-few-public-methods
     """An non-tensor opaque frontend object."""
 
-    object_type: typing.Type
+    object_type: type
 
-    def __init__(self, object_type: typing.Type) -> None:
+    def __init__(self, object_type: type) -> None:
         self.object_type = object_type
 
     def __repr__(self) -> str:
@@ -69,14 +69,14 @@ class Tuple:  # pylint: disable=too-few-public-methods
     """A tuple input or a list input"""
 
     name: str
-    elements: typing.Union[typing.List[SpecAny], typing.Tuple[SpecAny, ...]]
+    elements: list[SpecAny] | tuple[SpecAny, ...]
 
     def __init__(
         self,
         name: str,
-        elements: typing.Union[typing.List[SpecAny], typing.Tuple[SpecAny, ...]],
+        elements: list[SpecAny] | tuple[SpecAny, ...],
     ) -> None:
-        assert isinstance(elements, (tuple, list)), f"Unsupported container type: {type(elements)}"
+        assert isinstance(elements, tuple | list), f"Unsupported container type: {type(elements)}"
         self.name = name
         self.elements = elements
 
@@ -88,16 +88,16 @@ class MethodSpec:
     """A spec for a compiled method"""
 
     method: typing.Callable
-    arg_names: typing.List[str]
-    arg_specs: typing.List[ArgSpecType]
+    arg_names: list[str]
+    arg_specs: list[ArgSpecType]
     param_mode: str  # "plain", "packed", "none"
     effect_mode: str  # "plain", "packed", "none"
 
     def __init__(  # pylint: disable=too-many-arguments
         self,
         method: typing.Callable,
-        arg_names: typing.List[str],
-        arg_specs: typing.List[ArgSpecType],
+        arg_names: list[str],
+        arg_specs: list[ArgSpecType],
         param_mode: str,
         effect_mode: str,
     ):
@@ -142,7 +142,7 @@ def from_raw(spec: MethodSpecType, method: typing.Callable) -> "MethodSpec":
         """
         if isinstance(spec, MethodSpec):
             return spec
-        config: typing.Dict[str, typing.Any] = spec.pop("$", {})  # type: ignore[assignment]
+        config: dict[str, typing.Any] = spec.pop("$", {})  # type: ignore[assignment]
         param_mode = config.get("param_mode", "plain")
         effect_mode = config.get("effect_mode", "plain")
         method_signature = inspect.signature(method)
@@ -154,9 +154,9 @@ def _convert_arg_spec(arg_spec, arg_name):
                 return Int()
             if isinstance(arg_spec, str) and arg_spec == "int":
                 return Int()
-            if isinstance(arg_spec, (Int, Tensor, Object)):
+            if isinstance(arg_spec, Int | Tensor | Object):
                 return arg_spec
-            if isinstance(arg_spec, (tuple, list, Tuple)):
+            if isinstance(arg_spec, tuple | list | Tuple):
                 return Tuple(
                     arg_name,
                     elements=type(arg_spec)(
@@ -182,7 +182,7 @@ def _convert_arg_spec(arg_spec, arg_name):
         )
 
     @staticmethod
-    def from_torch(args: typing.List[typing.Any], method: typing.Callable) -> "MethodSpec":
+    def from_torch(args: list[typing.Any], method: typing.Callable) -> "MethodSpec":
         """Converts a list of torch tensors to MethodSpec."""
         from .torch import (  # pylint: disable=import-outside-toplevel
             _method_spec_from_torch,
@@ -195,14 +195,14 @@ class ModuleSpec:
     """A spec for a compiled nn.Module"""
 
     module: "nn_module_class"
-    method_names: typing.List[str]
-    method_specs: typing.List[MethodSpec]
+    method_names: list[str]
+    method_specs: list[MethodSpec]
 
     def __init__(
         self,
         module: "nn_module_class",
-        method_names: typing.List[str],
-        method_specs: typing.List[MethodSpec],
+        method_names: list[str],
+        method_specs: list[MethodSpec],
     ) -> None:
         self.module = module
         self.method_names = method_names
@@ -237,7 +237,7 @@ def from_raw(spec: ModuleSpecType, module: "nn_module_class") -> "ModuleSpec":
         if isinstance(spec, ModuleSpec):
             return spec
         method_names = list(spec.keys())
-        method_specs: typing.List[MethodSpec] = []
+        method_specs: list[MethodSpec] = []
         for method_name in method_names:
             method_spec = spec[method_name]
             if isinstance(method_spec, MethodSpec):
diff --git a/python/tvm/relax/frontend/nn/subroutine.py b/python/tvm/relax/frontend/nn/subroutine.py
index d7d30aa65816..87f272071202 100644
--- a/python/tvm/relax/frontend/nn/subroutine.py
+++ b/python/tvm/relax/frontend/nn/subroutine.py
@@ -57,7 +57,7 @@ def _get_struct_info(arg):
         return arg.struct_info_
     elif isinstance(arg, nn.Tensor):
         return arg._expr.struct_info_
-    elif isinstance(arg, (tuple, list, ir.Array)):
+    elif isinstance(arg, tuple | list | ir.Array):
         return relax.TupleStructInfo([_get_struct_info(field) for field in arg])
     else:
         raise TypeError(f"Cannot find struct info for {arg} of type {type(arg)}")
diff --git a/python/tvm/relax/frontend/nn/torch.py b/python/tvm/relax/frontend/nn/torch.py
index 8b59233d1e28..45dc1a74a9e4 100644
--- a/python/tvm/relax/frontend/nn/torch.py
+++ b/python/tvm/relax/frontend/nn/torch.py
@@ -17,7 +17,8 @@
 """PyTorch integration with nn.Module"""
 
 import inspect
-from typing import Any, Callable, List
+from collections.abc import Callable
+from typing import Any
 
 import torch
 
@@ -35,14 +36,14 @@ class TorchModule:  # pylint: disable=too-few-public-methods
 
     spec: _spec.ModuleSpec
     vm: VirtualMachine  # pylint: disable=invalid-name
-    params: List[Tensor]
-    effects: List[Any]
+    params: list[Tensor]
+    effects: list[Any]
 
     def __init__(  # pylint: disable=invalid-name
         self,
         spec: _spec.ModuleSpec,
         vm: VirtualMachine,
-        params: List[Tensor],
+        params: list[Tensor],
     ):
         try:
             self.effects = vm["_initialize_effect"]()
@@ -86,7 +87,7 @@ def _closure(*args):
 
 
 def _tvm_to_torch(arg):
-    if isinstance(arg, (list, tuple, Array)):
+    if isinstance(arg, list | tuple | Array):
         return [_tvm_to_torch(i) for i in arg]
     if isinstance(arg, _tensor.Tensor):
         return torch.utils.dlpack.from_dlpack(arg)
@@ -117,7 +118,7 @@ def _torch_to_tvm(arg_name, arg_spec, arg_torch):
 
 
 def _method_spec_from_torch(
-    args_torch: List[Any],
+    args_torch: list[Any],
     method: Callable,
 ):
     def _as_spec(arg_torch):
diff --git a/python/tvm/relax/frontend/onnx/onnx_frontend.py b/python/tvm/relax/frontend/onnx/onnx_frontend.py
index a512e20e234b..b3c2d06eab07 100644
--- a/python/tvm/relax/frontend/onnx/onnx_frontend.py
+++ b/python/tvm/relax/frontend/onnx/onnx_frontend.py
@@ -41,7 +41,8 @@
 import operator
 import re
 import warnings
-from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+from collections.abc import Callable
+from typing import Any
 
 import numpy as _np
 import onnx.onnx_ml_pb2
@@ -56,7 +57,7 @@
 from ..common import autopad
 
 
-def get_type(elem_type: Union[str, int]) -> str:
+def get_type(elem_type: str | int) -> str:
     """Converts onnx integer datatype to numpy datatype"""
     # If a string was passed instead of a tensor type, it does not need
     # conversion and can be returned.
@@ -74,9 +75,9 @@ def get_type(elem_type: Union[str, int]) -> str:
 
 
 def get_constant(
-    var: Union[relax.Constant, relax.Var],
-    params: List[Dict[str, relax.Var]],
-) -> Union[relax.Constant, relax.Var]:
+    var: relax.Constant | relax.Var,
+    params: list[dict[str, relax.Var]],
+) -> relax.Constant | relax.Var:
     """Attempt to convert a variable to a constant if possible.
     This is the primary function meant to interact with params.
 
@@ -109,7 +110,7 @@ def get_constant(
         return var
 
 
-def get_value(token, value_dict: Dict[str, tvm.tir.SizeVar]) -> Union[int, tvm.tir.SizeVar]:
+def get_value(token, value_dict: dict[str, tvm.tir.SizeVar]) -> int | tvm.tir.SizeVar:
     """Converts to token to an integer value if it a constant, otherwise it generates a SizeVar
 
     Parameters
@@ -136,8 +137,8 @@ def get_value(token, value_dict: Dict[str, tvm.tir.SizeVar]) -> Union[int, tvm.t
 
 
 def parse_shape_name(
-    name: str, value_dict: Dict[str, tvm.tir.SizeVar]
-) -> Union[tir.PrimExpr, tvm.tir.SizeVar]:
+    name: str, value_dict: dict[str, tvm.tir.SizeVar]
+) -> tir.PrimExpr | tvm.tir.SizeVar:
     """Converts expressions in the shape dimension name to prim expressions.
 
     Parameters
@@ -187,8 +188,8 @@ def parse_shape_name(
 
 
 def get_info(
-    info_proto: onnx.onnx_ml_pb2.ValueInfoProto, value_dict: Dict[str, tvm.tir.SizeVar]
-) -> Tuple[str, List, str, List, Dict]:
+    info_proto: onnx.onnx_ml_pb2.ValueInfoProto, value_dict: dict[str, tvm.tir.SizeVar]
+) -> tuple[str, list, str, list, dict]:
     """Extract the shape from a ValueInfoProto.
 
     Parameters
@@ -235,8 +236,8 @@ def get_numpy(tensor_proto: onnx.onnx_ml_pb2.TensorProto) -> _np.ndarray:
 
 
 def get_prim_expr_list(
-    inputs: Union[relax.Constant, relax.ShapeExpr],
-) -> List[Union[int, tir.PrimExpr]]:
+    inputs: relax.Constant | relax.ShapeExpr,
+) -> list[int | tir.PrimExpr]:
     """Attempt to convert a variable to list of PrimExpr if possible.
 
     Parameters
@@ -319,7 +320,7 @@ def _impl_v13(cls, bb, inputs, attr, params):
 def _to_numpy(x):
     if isinstance(x, relax.PrimValue):
         x = x.value
-        if isinstance(x, (tir.IntImm, tir.FloatImm)):
+        if isinstance(x, tir.IntImm | tir.FloatImm):
             x = x.value
         return _np.array(x)
     else:
@@ -337,7 +338,7 @@ def base_impl(cls, bb, inputs, attr, params):
         """Base implementation for binary operations."""
         if cls.numpy_op is None or cls.relax_op is None:
             raise ValueError("Numpy and Relax operators must be defined for BinaryBase.")
-        if all([not isinstance(inp, (relax.expr.Call, relax.Var)) for inp in inputs]):
+        if all([not isinstance(inp, relax.expr.Call | relax.Var) for inp in inputs]):
             x = _to_numpy(inputs[0])
             y = _to_numpy(inputs[1])
             output = cls.numpy_op(x, y)  # pylint: disable=not-callable
@@ -511,7 +512,7 @@ def _impl_v13(cls, bb, inputs, attr, params):
         if all([isinstance(inp, relax.Constant) for inp in inputs]):
             output = inputs[0].data.numpy() == inputs[1].data.numpy()
             return relax.const(output, output.dtype)
-        elif all([isinstance(inp, (relax.Constant, relax.ShapeExpr)) for inp in inputs]):
+        elif all([isinstance(inp, relax.Constant | relax.ShapeExpr) for inp in inputs]):
             lhs = get_prim_expr_list(inputs[0])
             rhs = get_prim_expr_list(inputs[1])
             if len(lhs) != len(rhs):
@@ -855,7 +856,7 @@ class ScatterND(OnnxOpConverter):
     """Convert an onnx ScatterND node into an equivalent Relax expression."""
 
     @staticmethod
-    def _reduction_check(attr, valid_reductions: List[str]):
+    def _reduction_check(attr, valid_reductions: list[str]):
         reduction = attr.get("reduction", None)
         reduction = reduction or b"update"
         reduction = reduction.decode("utf-8")
@@ -994,7 +995,7 @@ def _impl_v16(cls, bb, inputs, attr, params):
             np_inputs = [inp.data.numpy() for inp in inputs]
             output = _np.where(*np_inputs)
             return relax.const(output, output.dtype)
-        if all([isinstance(inp, (relax.Constant, relax.ShapeExpr)) for inp in inputs]):
+        if all([isinstance(inp, relax.Constant | relax.ShapeExpr) for inp in inputs]):
             condition, x, y = [get_prim_expr_list(inp) for inp in inputs]
             if len(condition) != len(x) or len(condition) != len(y):
                 raise ValueError("Cannot broadcast condition to x and y")
@@ -1431,7 +1432,7 @@ def _impl_v13(cls, bb, inputs, attr, params):
 
         # If data is constant, perform computation directly.
         if isinstance(data, relax.Constant):
-            if isinstance(axis, (tuple, type(None))):
+            if isinstance(axis, tuple | type(None)):
                 out_data = _np.squeeze(data.data.numpy(), axis)
             else:
                 raise NotImplementedError("Squeeze with symbolic axes not supported")
@@ -1857,7 +1858,7 @@ def _impl_v13(cls, bb, inputs, attr, params):
         if not all(
             [
                 (
-                    isinstance(param, (relax.Constant, relax.ShapeExpr, relax.PrimValue))
+                    isinstance(param, relax.Constant | relax.ShapeExpr | relax.PrimValue)
                     or param is None
                 )
                 for param in [starts, ends, axes, steps]
@@ -1886,7 +1887,7 @@ def _impl_v13(cls, bb, inputs, attr, params):
             assert all(len(i) == 1 for i in [starts, ends, steps])
             sliced_values = shape_data[starts[0] : ends[0] : steps[0]]
 
-            if all([isinstance(val, (tir.IntImm, int)) for val in sliced_values]):
+            if all([isinstance(val, tir.IntImm | int) for val in sliced_values]):
                 return relax.const([x.value for x in sliced_values], "int64")
             else:
                 return relax.ShapeExpr(sliced_values)
@@ -1894,7 +1895,7 @@ def _impl_v13(cls, bb, inputs, attr, params):
         # If all `starts`, `ends`, and `steps` are constant, use strict mode
         # Otherwise, we assume the slice is inbound.
         assume_inbound = not all(
-            [isinstance(param, (tir.IntImm, int)) for param in [*starts, *ends, *steps]]
+            [isinstance(param, tir.IntImm | int) for param in [*starts, *ends, *steps]]
         )
 
         # Converting PrimExpr to PrimValue since relax.op.strided_slice does not accept PrimExpr
@@ -2774,7 +2775,7 @@ def _impl_v17(cls, bb, inputs, attr, params):
             if gamma_shape != beta_shape:
                 raise ValueError("gamma and beta shapes do not match")
 
-        axis = list(axis) if isinstance(axis, (list, tuple)) else [axis]
+        axis = list(axis) if isinstance(axis, list | tuple) else [axis]
         if len(axis) < len(gamma_shape):
             axis.extend(range(axis[-1] + 1, axis[-1] + 1 + len(gamma_shape) - len(axis)))
 
@@ -4041,16 +4042,16 @@ class ONNXGraphImporter:
 
     def __init__(
         self,
-        shape_dict: Dict[str, List],
-        dtype_dict: Union[str, Dict[str, str]],
+        shape_dict: dict[str, list],
+        dtype_dict: str | dict[str, str],
         keep_params_in_input: bool = False,
         sanitize: bool = True,
     ):
-        self._nodes: Dict[str, relax.Expr] = {}
-        self._inputs: Dict[str, relax.Var] = {}
+        self._nodes: dict[str, relax.Expr] = {}
+        self._inputs: dict[str, relax.Var] = {}
         self._num_input: int = 0
         self._shape = shape_dict.copy() if shape_dict else {}
-        self._input_names: List[str] = []
+        self._input_names: list[str] = []
         self._dtype = dtype_dict
         self.opset: int = None
         self._name_supply = NameSupply()
@@ -4153,7 +4154,7 @@ def _sanitize_name(self, name: str) -> str:
             warnings.warn(f"Renaming name {name} to {new_name}")
         return new_name
 
-    def _new_var(self, var_name: str, shape: List, dtype: str = "float32"):
+    def _new_var(self, var_name: str, shape: list, dtype: str = "float32"):
         """Creates a new Relax variable."""
         return relax.Var(
             name_hint=var_name, struct_info=relax.TensorStructInfo(shape=shape, dtype=dtype)
@@ -4296,7 +4297,7 @@ def _parse_array(self, tensor_proto: onnx.onnx_ml_pb2.TensorProto) -> tvm.runtim
         np_array = get_numpy(tensor_proto).reshape(tuple(tensor_proto.dims))
         return tvm.runtime.tensor(np_array)
 
-    def _parse_attr(self, attr_proto: onnx.onnx_ml_pb2.AttributeProto) -> Dict[str, Any]:
+    def _parse_attr(self, attr_proto: onnx.onnx_ml_pb2.AttributeProto) -> dict[str, Any]:
         """Convert a list of AttributeProto to a dict, with names as keys."""
         attrs = {}
         for a in attr_proto:
@@ -4324,8 +4325,8 @@ def _parse_attr(self, attr_proto: onnx.onnx_ml_pb2.AttributeProto) -> Dict[str,
     def _convert_operator(
         self,
         op_name: str,
-        inputs: List[relax.Expr],
-        attrs: Dict,
+        inputs: list[relax.Expr],
+        attrs: dict,
         opset: int,
     ) -> relax.Expr:
         """Convert ONNX operator into a Relax operator.
@@ -4359,9 +4360,9 @@ def _convert_operator(
 
 def from_onnx(
     model: onnx.onnx_ml_pb2.GraphProto,
-    shape_dict: Optional[Dict[str, List]] = None,
-    dtype_dict: Optional[Union[str, Dict[str, str]]] = "float32",
-    opset: Optional[int] = None,
+    shape_dict: dict[str, list] | None = None,
+    dtype_dict: str | dict[str, str] | None = "float32",
+    opset: int | None = None,
     keep_params_in_input: bool = False,
     sanitize_input_names: bool = True,
 ) -> IRModule:
diff --git a/python/tvm/relax/frontend/stablehlo/stablehlo_translator.py b/python/tvm/relax/frontend/stablehlo/stablehlo_translator.py
index f3bb2a57faa4..45f2989539b5 100644
--- a/python/tvm/relax/frontend/stablehlo/stablehlo_translator.py
+++ b/python/tvm/relax/frontend/stablehlo/stablehlo_translator.py
@@ -18,7 +18,8 @@
 
 """StableHLO frontend of Relax."""
 
-from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+from collections.abc import Callable
+from typing import Any
 
 import tvm
 from tvm import relax, tir
@@ -33,7 +34,7 @@ class StableHLOImporter:
     def __init__(self) -> None:
         from jaxlib import mlir
 
-        self._nodes: Dict[Union[str, mlir.ir.Operation], relax.Expr] = {}
+        self._nodes: dict[str | mlir.ir.Operation, relax.Expr] = {}
         self.block_builder: relax.BlockBuilder = None
         self.create_convert_map()
 
@@ -73,7 +74,7 @@ def _convert_data_type(input_type):
         else:
             raise NotImplementedError(f"input_type {input_type} is not handled yet")
 
-    def _attr2value(self, node) -> Union[Any, List[Any]]:
+    def _attr2value(self, node) -> Any | list[Any]:
         import numpy as np
         from jaxlib import mlir
 
@@ -107,7 +108,7 @@ def _retrieve_operands(self, node):
             return self._nodes[node]
         if isinstance(node, tuple):
             return tuple(self._retrieve_operands(x) for x in node)
-        if isinstance(node, (list, mlir.ir.OpOperandList)):
+        if isinstance(node, list | mlir.ir.OpOperandList):
             return [self._retrieve_operands(x) for x in node]
         if isinstance(node, dict):
             return {self._retrieve_operands(k): self._retrieve_operands(v) for k, v in node.items()}
@@ -118,7 +119,7 @@ def _retrieve_operands(self, node):
             return self._retrieve_operands(node.owner)
         return node
 
-    def get_shape(self, inpt_type) -> List[Any]:
+    def get_shape(self, inpt_type) -> list[Any]:
         """Get the shape from Type like tensor<?x?xf32>"""
         from jaxlib import mlir
 
@@ -325,7 +326,7 @@ def _return(self, node: mlir.ir.Operation) -> relax.Expr:
     def create_convert_map(self):
         from jaxlib import mlir
 
-        self.convert_map: Dict[str, Callable[[mlir.ir.Operation], relax.Var]] = {
+        self.convert_map: dict[str, Callable[[mlir.ir.Operation], relax.Var]] = {
             "stablehlo.add": self._add,
             "stablehlo.broadcast_in_dim": self._broadcast_in_dim,
             "stablehlo.constant": self._const,
@@ -351,7 +352,7 @@ def create_convert_map(self):
             "stablehlo.return": self._return,
         }
 
-    def from_stablehlo(self, model, input_info: List[Tuple[Tuple[int], str]]) -> tvm.IRModule:
+    def from_stablehlo(self, model, input_info: list[tuple[tuple[int], str]]) -> tvm.IRModule:
         """Convert a StableHLO Module to a Relax program.
 
         Parameters
@@ -395,7 +396,7 @@ def from_stablehlo(self, model, input_info: List[Tuple[Tuple[int], str]]) -> tvm
             with self.block_builder.dataflow():
                 block = model.body.operations[0].regions[0].blocks[0]
                 for operation in block.operations:
-                    if isinstance(operation, (mlir.dialects.func.ReturnOp, stablehlo.ReturnOp)):
+                    if isinstance(operation, mlir.dialects.func.ReturnOp | stablehlo.ReturnOp):
                         operation = operation.operands[0].owner
                         # TODO (yongwww): handle multiple outputs
                         output = self.block_builder.emit_output(self._nodes[operation])
@@ -416,7 +417,7 @@ def from_stablehlo(self, model, input_info: List[Tuple[Tuple[int], str]]) -> tvm
 
 def from_stablehlo(
     stablehlo_module,
-    input_info: Optional[List[Tuple[Tuple[int], str]]] = None,
+    input_info: list[tuple[tuple[int], str]] | None = None,
 ) -> tvm.IRModule:
     """Convert a StableHLO Module to a Relax program
 
diff --git a/python/tvm/relax/frontend/torch/base_fx_graph_translator.py b/python/tvm/relax/frontend/torch/base_fx_graph_translator.py
index 5fe0ef17d4e6..2615a68ee34c 100644
--- a/python/tvm/relax/frontend/torch/base_fx_graph_translator.py
+++ b/python/tvm/relax/frontend/torch/base_fx_graph_translator.py
@@ -22,8 +22,8 @@
 
 import abc
 import math
+from collections.abc import Callable
 from functools import reduce
-from typing import Callable, Dict, List, Optional, Tuple, Union
 
 import tvm
 from tvm import relax, tir
@@ -39,16 +39,16 @@ def __init__(self) -> None:
         import torch  # type: ignore
         from torch import fx
 
-        self.env: Dict[fx.Node, relax.Expr] = {}
-        self.params: Dict[torch.Tensor, relax.Expr] = {}
+        self.env: dict[fx.Node, relax.Expr] = {}
+        self.params: dict[torch.Tensor, relax.Expr] = {}
         self.block_builder: relax.BlockBuilder = None
-        self.convert_map: Dict[Union[torch.nn.Module, str], Callable[[fx.Node], relax.Var]] = (
+        self.convert_map: dict[torch.nn.Module | str, Callable[[fx.Node], relax.Var]] = (
             self.create_convert_map()
         )
 
     ########## Utilities ##########
 
-    def update_convert_map(self, custom_convert_map: Dict[str, Callable]):
+    def update_convert_map(self, custom_convert_map: dict[str, Callable]):
         """Update self.convert_map with custom convert map
 
         Parameters
@@ -60,7 +60,7 @@ def update_convert_map(self, custom_convert_map: Dict[str, Callable]):
         self.convert_map.update(custom_convert_map)
 
     @staticmethod
-    def _convert_data_type(input_type: Union[str, torch.dtype], env: Optional[Dict] = None):
+    def _convert_data_type(input_type: str | torch.dtype, env: dict | None = None):
         """converts the PyTorch scalar type input_type to a TVM dtype."""
         import torch  # type: ignore
 
@@ -121,7 +121,7 @@ def shape_of(tensor):
         raise ValueError(f"Unsupported type: {type(tensor)}")
 
     @staticmethod
-    def _promote_common_dtype(lhs_dtype: Optional[str], rhs_dtype: Optional[str]) -> Optional[str]:
+    def _promote_common_dtype(lhs_dtype: str | None, rhs_dtype: str | None) -> str | None:
         """Return the promoted dtype following PyTorch rules, or None if unsupported."""
         import torch  # type: ignore
 
@@ -185,7 +185,7 @@ def _retrieve_args(self, node):
         else:
             return node
 
-    def _check_unsupported_func_type(self, nodes: List[fx.Node]):
+    def _check_unsupported_func_type(self, nodes: list[fx.Node]):
         missing_func_types = list(
             {
                 node.target.__name__
@@ -210,7 +210,7 @@ def _celu(self, node: fx.Node) -> relax.Var:
         alpha = node.args[1] if len(node.args) > 1 else node.kwargs.get("alpha", 1.0)
         dtype = x.struct_info.dtype
 
-        if isinstance(alpha, (int, float)):
+        if isinstance(alpha, int | float):
             alpha = relax.const(alpha, dtype)
         else:
             if not isinstance(alpha, relax.Var):
@@ -243,7 +243,7 @@ def _clamp(self, node: fx.Node) -> relax.Expr:
         a_max = math.inf if a_max is None else a_max
 
         # Handle the case where a_min is a tensor
-        if not isinstance(a_min, (int, float)):
+        if not isinstance(a_min, int | float):
             from torch import fx
 
             if isinstance(a_min, fx.Node):
@@ -258,7 +258,7 @@ def _clamp(self, node: fx.Node) -> relax.Expr:
             a_min = -math.inf
 
         # Handle the case where a_max is a tensor
-        if not isinstance(a_max, (int, float)):
+        if not isinstance(a_max, int | float):
             from torch import fx
 
             if isinstance(a_max, fx.Node):
@@ -283,7 +283,7 @@ def _clamp_min(self, node: fx.Node) -> relax.Expr:
         a_min = -math.inf if a_min is None else a_min
 
         # Handle the case where a_min is a tensor
-        if not isinstance(a_min, (int, float)):
+        if not isinstance(a_min, int | float):
             from torch import fx
 
             if isinstance(a_min, fx.Node):
@@ -308,7 +308,7 @@ def _clamp_max(self, node: fx.Node) -> relax.Expr:
         a_max = math.inf if a_max is None else a_max
 
         # Handle the case where a_max is a tensor
-        if not isinstance(a_max, (int, float)):
+        if not isinstance(a_max, int | float):
             from torch import fx
 
             if isinstance(a_max, fx.Node):
@@ -329,7 +329,7 @@ def _elu(self, node: fx.Node) -> relax.Var:
         alpha = node.args[1] if len(node.args) > 1 else node.kwargs.get("alpha", 1.0)
         dtype = x.struct_info.dtype
 
-        if isinstance(alpha, (int, float)):
+        if isinstance(alpha, int | float):
             alpha = relax.const(-alpha, dtype)
         else:
             if not isinstance(alpha, relax.Var):
@@ -521,7 +521,7 @@ def _div(self, node: fx.Node) -> relax.Var:
         inp_2 = args[1]
 
         # Handle scalar cases
-        if isinstance(inp_2, (int, float)):
+        if isinstance(inp_2, int | float):
             inp_2 = relax.const(inp_2)
 
         # Get rounding_mode from node kwargs
@@ -560,7 +560,7 @@ def _rsub(self, node: fx.Node) -> relax.Var:
         lhs = args[0]
         rhs = args[1]
 
-        if isinstance(rhs, (int, float)):
+        if isinstance(rhs, int | float):
             rhs = relax.const(rhs)
 
         return self.block_builder.emit(relax.op.subtract(rhs, lhs))
@@ -689,11 +689,11 @@ def _addmm(self, node: fx.Node) -> relax.Var:
     def _avg_pool1d_impl(
         self,
         x: relax.Expr,
-        kernel_size: Union[int, Tuple[int]] = 1,
-        stride: Optional[Union[int, Tuple[int]]] = None,
-        padding: Optional[int] = 0,
-        ceil_mode: Optional[bool] = False,
-        count_include_pad: Optional[bool] = True,
+        kernel_size: int | tuple[int] = 1,
+        stride: int | tuple[int] | None = None,
+        padding: int | None = 0,
+        ceil_mode: bool | None = False,
+        count_include_pad: bool | None = True,
     ) -> relax.Var:
         # Expand to 3D by adding batch dim if input is 2D
         x_ndim = x.struct_info.ndim
@@ -731,11 +731,11 @@ def _avg_pool1d(self, node: fx.Node) -> relax.Var:
     def _avg_pool2d_impl(
         self,
         x: relax.Expr,
-        kernel_size: Union[int, Tuple[int, int]] = (1, 1),
-        stride: Optional[Union[int, Tuple[int, int]]] = None,
-        padding: Optional[int] = 0,
-        ceil_mode: Optional[bool] = False,
-        count_include_pad: Optional[bool] = True,
+        kernel_size: int | tuple[int, int] = (1, 1),
+        stride: int | tuple[int, int] | None = None,
+        padding: int | None = 0,
+        ceil_mode: bool | None = False,
+        count_include_pad: bool | None = True,
     ) -> relax.Var:
         # Expand to 4D by adding batch dim if input is 3D
         x_ndim = x.struct_info.ndim
@@ -772,11 +772,11 @@ def _avg_pool2d(self, node: fx.Node) -> relax.Var:
     def _avg_pool3d_impl(
         self,
         x: relax.Expr,
-        kernel_size: Union[int, Tuple[int, int, int]] = (1, 1, 1),
-        stride: Optional[Union[int, Tuple[int, int, int]]] = None,
-        padding: Optional[int] = 0,
-        ceil_mode: Optional[bool] = False,
-        count_include_pad: Optional[bool] = True,
+        kernel_size: int | tuple[int, int, int] = (1, 1, 1),
+        stride: int | tuple[int, int, int] | None = None,
+        padding: int | None = 0,
+        ceil_mode: bool | None = False,
+        count_include_pad: bool | None = True,
     ) -> relax.Var:
         # Expand to 5D by adding batch dim if input is 4D
         x_ndim = x.struct_info.ndim
@@ -837,12 +837,12 @@ def _conv_transpose1d_impl(
         self,
         x: relax.Expr,
         weight: relax.Expr,
-        bias: Optional[relax.Expr],
-        strides: Optional[Tuple],
-        padding: Optional[Tuple],
-        dilation: Optional[Tuple],
-        groups: Optional[Tuple],
-        output_padding: Optional[Tuple],
+        bias: relax.Expr | None,
+        strides: tuple | None,
+        padding: tuple | None,
+        dilation: tuple | None,
+        groups: tuple | None,
+        output_padding: tuple | None,
     ) -> relax.Var:
         conv1d_transpose = self.block_builder.emit(
             relax.op.nn.conv1d_transpose(
@@ -891,12 +891,12 @@ def _conv_transpose2d_impl(
         self,
         x: relax.Expr,
         weight: relax.Expr,
-        bias: Optional[relax.Expr],
-        strides: Optional[Tuple],
-        padding: Optional[Tuple],
-        dilation: Optional[Tuple],
-        groups: Optional[Tuple],
-        output_padding: Optional[Tuple],
+        bias: relax.Expr | None,
+        strides: tuple | None,
+        padding: tuple | None,
+        dilation: tuple | None,
+        groups: tuple | None,
+        output_padding: tuple | None,
     ) -> relax.Var:
         conv2d_transpose = self.block_builder.emit(
             relax.op.nn.conv2d_transpose(
@@ -945,11 +945,11 @@ def _conv1d_impl(
         self,
         x: relax.Expr,
         weight: relax.Expr,
-        bias: Optional[relax.Expr],
-        strides: Optional[Tuple],
-        padding: Optional[Tuple],
-        dilation: Optional[Tuple],
-        groups: Optional[Tuple],
+        bias: relax.Expr | None,
+        strides: tuple | None,
+        padding: tuple | None,
+        dilation: tuple | None,
+        groups: tuple | None,
     ) -> relax.Var:
         conv1d = self.block_builder.emit(
             relax.op.nn.conv1d(
@@ -994,11 +994,11 @@ def _conv2d_impl(
         self,
         x: relax.Expr,
         weight: relax.Expr,
-        bias: Optional[relax.Expr],
-        strides: Optional[Tuple],
-        padding: Optional[Tuple],
-        dilation: Optional[Tuple],
-        groups: Optional[Tuple],
+        bias: relax.Expr | None,
+        strides: tuple | None,
+        padding: tuple | None,
+        dilation: tuple | None,
+        groups: tuple | None,
     ):
         conv2d = self.block_builder.emit(
             relax.op.nn.conv2d(
@@ -1043,11 +1043,11 @@ def _conv3d_impl(
         self,
         x: relax.Expr,
         weight: relax.Expr,
-        bias: Optional[relax.Expr],
-        strides: Optional[Tuple],
-        padding: Optional[Tuple],
-        dilation: Optional[Tuple],
-        groups: Optional[Tuple],
+        bias: relax.Expr | None,
+        strides: tuple | None,
+        padding: tuple | None,
+        dilation: tuple | None,
+        groups: tuple | None,
     ):
         conv3d = self.block_builder.emit(
             relax.op.nn.conv3d(
@@ -1166,7 +1166,7 @@ def _cross_entropy_loss(
         self,
         preds: relax.Expr,
         targets: relax.Expr,
-        weights: Optional[relax.Expr],
+        weights: relax.Expr | None,
         reduction: str,
         ignore_index: int,
     ) -> relax.Expr:
@@ -1185,7 +1185,7 @@ def _einsum(self, node: fx.Node) -> relax.Var:
         import torch  # type: ignore
 
         args = self.retrieve_args(node)
-        operands = args[1] if isinstance(args[1], (torch.Size, tuple, list)) else args[1:]
+        operands = args[1] if isinstance(args[1], torch.Size | tuple | list) else args[1:]
         return self.block_builder.emit(relax.op.einsum(operands, args[0]))
 
     def _embedding_impl(
@@ -1209,7 +1209,7 @@ def _layer_norm_impl(self, x, gamma, beta, eps, normalized_shape) -> relax.Var:
         import numpy as np  # type: ignore
         from torch.fx.immutable_collections import immutable_list
 
-        if isinstance(normalized_shape, (immutable_list, tuple)):
+        if isinstance(normalized_shape, immutable_list | tuple):
             normalized_shape = tuple(normalized_shape)
         else:
             try:
@@ -1270,11 +1270,11 @@ def _linear(self, node: fx.Node) -> relax.Var:
     def _max_pool1d_impl(
         self,
         x: relax.Expr,
-        kernel_size: Union[int, Tuple[int]] = 1,
-        stride: Optional[Union[int, Tuple[int]]] = None,
-        padding: Optional[int] = 0,
-        dilation: Optional[int] = 1,
-        ceil_mode: Optional[bool] = False,
+        kernel_size: int | tuple[int] = 1,
+        stride: int | tuple[int] | None = None,
+        padding: int | None = 0,
+        dilation: int | None = 1,
+        ceil_mode: bool | None = False,
     ) -> relax.Var:
         # Expand to 3D by adding batch dim if input is 2D
         x_ndim = x.struct_info.ndim
@@ -1314,11 +1314,11 @@ def _max_pool1d(self, node: fx.Node) -> relax.Var:
     def _max_pool2d_impl(
         self,
         x: relax.Expr,
-        kernel_size: Union[int, Tuple[int, int]] = (1, 1),
-        stride: Optional[Union[int, Tuple[int, int]]] = None,
-        padding: Optional[int] = 0,
-        dilation: Optional[int] = 1,
-        ceil_mode: Optional[bool] = False,
+        kernel_size: int | tuple[int, int] = (1, 1),
+        stride: int | tuple[int, int] | None = None,
+        padding: int | None = 0,
+        dilation: int | None = 1,
+        ceil_mode: bool | None = False,
     ) -> relax.Var:
         # Expand to 4D by adding batch dim if input is 3D
         x_ndim = x.struct_info.ndim
@@ -1358,11 +1358,11 @@ def _max_pool2d(self, node: fx.Node) -> relax.Var:
     def _max_pool3d_impl(
         self,
         x: relax.Expr,
-        kernel_size: Union[int, Tuple[int, int, int]] = (1, 1, 1),
-        stride: Optional[Union[int, Tuple[int, int, int]]] = None,
-        padding: Optional[int] = 0,
-        dilation: Optional[int] = 1,
-        ceil_mode: Optional[bool] = False,
+        kernel_size: int | tuple[int, int, int] = (1, 1, 1),
+        stride: int | tuple[int, int, int] | None = None,
+        padding: int | None = 0,
+        dilation: int | None = 1,
+        ceil_mode: bool | None = False,
     ) -> relax.Var:
         # Expand to 5D by adding batch dim if input is 4D
         x_ndim = x.struct_info.ndim
@@ -1639,7 +1639,7 @@ def _sum(self, node: fx.Node) -> relax.Var:
         args = self.retrieve_args(node)
         x = args[0]
         dim = args[1] if len(node.args) > 1 else node.kwargs.get("dim", None)
-        if isinstance(dim, (list, tuple)) and len(dim) == 0:
+        if isinstance(dim, list | tuple) and len(dim) == 0:
             dim = None
         keepdim = args[2] if len(node.args) > 2 else node.kwargs.get("keepdim", False)
         return self.block_builder.emit(relax.op.sum(x, dim, keepdims=keepdim))
@@ -1791,7 +1791,7 @@ def _flatten(self, node: fx.Node) -> relax.Var:
     def _flip(self, node: fx.Node) -> relax.Var:
         x = self.env[node.args[0]]
         dims = node.args[1] if len(node.args) > 1 else node.kwargs.get("dims", None)
-        if isinstance(dims, (list, tuple)) and len(dims) > 0:
+        if isinstance(dims, list | tuple) and len(dims) > 0:
             dims = dims[0]
         elif not isinstance(dims, int):
             raise TypeError(f"flip expects an integer axis, but got {type(dims)}: {dims}")
@@ -1816,7 +1816,7 @@ def _index_put(self, node: fx.Node) -> relax.Var:
         if not isinstance(accumulate, bool):
             raise TypeError(f"'accumulate' must be a boolean value, got {type(accumulate)}")
 
-        if isinstance(indices, (list, tuple)):
+        if isinstance(indices, list | tuple):
             # In PyTorch index_put, None means "select all elements" for that dimension
             non_none_indices = [(i, idx) for i, idx in enumerate(indices) if idx is not None]
 
@@ -1942,13 +1942,13 @@ def _normalize_bound(bound):
             max_index_val = 9223372036854775807
 
             def _adjust(val):
-                if isinstance(val, (int, tir.IntImm)):
+                if isinstance(val, int | tir.IntImm):
                     int_val = int(val)
                     if int_val >= max_index_val:
                         return input_shape[axis]
                     if int_val < 0:
                         return input_shape[axis] + int_val
-                    if isinstance(input_shape[axis], (int, tir.IntImm)) and int_val > int(
+                    if isinstance(input_shape[axis], int | tir.IntImm) and int_val > int(
                         input_shape[axis]
                     ):
                         return input_shape[axis]
@@ -1977,7 +1977,7 @@ def _permute(self, node: fx.Node) -> relax.Var:
 
         args = self.retrieve_args(node)
         x = args[0]
-        dims = args[1] if isinstance(args[1], (torch.Size, tuple, list)) else args[1:]
+        dims = args[1] if isinstance(args[1], torch.Size | tuple | list) else args[1:]
         return self.block_builder.emit(relax.op.permute_dims(x, dims))
 
     def _repeat(self, node: fx.Node) -> relax.Var:
@@ -1985,7 +1985,7 @@ def _repeat(self, node: fx.Node) -> relax.Var:
 
         args = self.retrieve_args(node)
         x = args[0]
-        dims = args[1] if isinstance(args[1], (torch.Size, tuple, list)) else args[1:]
+        dims = args[1] if isinstance(args[1], torch.Size | tuple | list) else args[1:]
         return self.block_builder.emit(relax.op.tile(x, dims))
 
     def _roll(self, node: fx.Node) -> relax.Var:
@@ -2040,17 +2040,17 @@ def roll_single_dim(tensor: relax.Var, shift: int, dim: int) -> relax.Var:
         # Handle dims=None (flatten -> roll -> reshape)
         if dims is None:
             flattened = self.block_builder.emit(relax.op.reshape(input_tensor, (-1,)))
-            shift_scalar = to_int(shifts[0] if isinstance(shifts, (list, tuple)) else shifts)
+            shift_scalar = to_int(shifts[0] if isinstance(shifts, list | tuple) else shifts)
             rolled = roll_single_dim(flattened, shift_scalar, 0)
             return self.block_builder.emit(relax.op.reshape(rolled, original_shape))
 
         # Normalize shifts and dims
-        if isinstance(shifts, (list, tuple)):
+        if isinstance(shifts, list | tuple):
             shifts = [to_int(s) for s in shifts]
         else:
             shifts = [to_int(shifts)]
 
-        if isinstance(dims, (list, tuple)):
+        if isinstance(dims, list | tuple):
             dims = [to_int(d) for d in dims]
         else:
             dims = [to_int(dims)]
@@ -2072,7 +2072,7 @@ def _reshape(self, node: fx.Node) -> relax.Var:
 
         args = self.retrieve_args(node)
         x = args[0]
-        dims = args[1] if isinstance(args[1], (torch.Size, tuple, list)) else args[1:]
+        dims = args[1] if isinstance(args[1], torch.Size | tuple | list) else args[1:]
 
         # Skip identity reshape
         current_shape = self.shape_of(x)
@@ -2117,7 +2117,7 @@ def _split(self, node: fx.Node) -> relax.Var:
         x = self.env[node.args[0]]
         split_size = node.args[1]
         dim = node.args[2] if len(node.args) > 2 else node.kwargs.get("dim", 0)
-        if isinstance(split_size, (list, tuple)):
+        if isinstance(split_size, list | tuple):
             n_section = []
             for s in split_size[:-1]:
                 cum_sum = 0 if not n_section else n_section[-1]
@@ -2135,7 +2135,7 @@ def _squeeze(self, node: fx.Node) -> relax.Var:
 
         # If dims is a list, filter out axes where dimension is not 1
         # This is needed because PyTorch decomposition may pass all axes
-        if isinstance(dim, (list, tuple)) and len(dim) > 0:
+        if isinstance(dim, list | tuple) and len(dim) > 0:
             shape = self.shape_of(x)
             # Filter to only include axes where the dimension is 1
             valid_dims = []
@@ -2165,7 +2165,7 @@ def _tile(self, node: fx.Node) -> relax.Var:
 
         args = self.retrieve_args(node)
         x = args[0]
-        dims = args[1] if isinstance(args[1], (torch.Size, tuple, list)) else args[1:]
+        dims = args[1] if isinstance(args[1], torch.Size | tuple | list) else args[1:]
         return self.block_builder.emit(relax.op.tile(x, dims))
 
     def _topk(self, node: fx.Node) -> relax.Var:
@@ -2311,7 +2311,7 @@ def _full(self, node: fx.Node) -> relax.Var:
         import torch
 
         args = self.retrieve_args(node)
-        size = relax.ShapeExpr(args[0] if isinstance(args[0], (list, tuple)) else (args[0],))
+        size = relax.ShapeExpr(args[0] if isinstance(args[0], list | tuple) else (args[0],))
         dtype = self._convert_data_type(
             node.kwargs.get("dtype", torch.get_default_dtype()), self.env
         )
@@ -2331,7 +2331,7 @@ def _full_like(self, node: fx.Node) -> relax.Var:
 
         x_dtype = x.struct_info.dtype
         fill_dtype = None
-        if isinstance(value, (int, float)) and (math.isinf(value) or math.isnan(value)):
+        if isinstance(value, int | float) and (math.isinf(value) or math.isnan(value)):
             if not ("float" in x_dtype or "bfloat16" in x_dtype):
                 fill_dtype = "float32"
 
@@ -2351,7 +2351,7 @@ def _inplace_masked_fill(self, node: fx.Node) -> relax.Var:
 
         x_dtype = x.struct_info.dtype
         fill_dtype = None
-        if isinstance(value, (int, float)) and (math.isinf(value) or math.isnan(value)):
+        if isinstance(value, int | float) and (math.isinf(value) or math.isnan(value)):
             if not ("float" in x_dtype or "bfloat16" in x_dtype):
                 fill_dtype = "float32"
 
@@ -2396,7 +2396,7 @@ def _masked_fill(self, node: fx.Node) -> relax.Var:
 
         x_dtype = x.struct_info.dtype
         fill_dtype = None
-        if isinstance(value, (int, float)) and (math.isinf(value) or math.isnan(value)):
+        if isinstance(value, int | float) and (math.isinf(value) or math.isnan(value)):
             if not ("float" in x_dtype or "bfloat16" in x_dtype):
                 fill_dtype = "float32"
 
@@ -2431,8 +2431,8 @@ def _masked_select(self, node: fx.Node) -> relax.Var:
     def _new_ones(self, node: fx.Node) -> relax.Var:
         args = self.retrieve_args(node)
         self_var = args[0]
-        size = args[1] if isinstance(args[1], (list, tuple)) else args[1:]
-        if not isinstance(size, (list, tuple)):
+        size = args[1] if isinstance(args[1], list | tuple) else args[1:]
+        if not isinstance(size, list | tuple):
             size = (size,)
         size = relax.ShapeExpr(size)
         return self.block_builder.emit(
@@ -2448,7 +2448,7 @@ def _new_zeros(self, node: fx.Node) -> relax.Var:
         input_tensor = args[0]
         size = (
             args[1]
-            if isinstance(args[1], (list, tuple))
+            if isinstance(args[1], list | tuple)
             else (args[1],)
             if len(args[1:]) == 1
             else args[1:]
@@ -2466,7 +2466,7 @@ def _ones(self, node: fx.Node) -> relax.Var:
         import torch
 
         args = self.retrieve_args(node)
-        size = relax.ShapeExpr(args[0] if isinstance(args[0], (list, tuple)) else (args[0],))
+        size = relax.ShapeExpr(args[0] if isinstance(args[0], list | tuple) else (args[0],))
         dtype = self._convert_data_type(
             node.kwargs.get("dtype", torch.get_default_dtype()), self.env
         )
@@ -2505,7 +2505,7 @@ def _getitem(self, node: fx.Node) -> relax.Var:
         import torch
 
         x = self.env[node.args[0]]
-        if isinstance(x, (list, tuple, relax.ShapeExpr, relax.Tuple)):
+        if isinstance(x, list | tuple | relax.ShapeExpr | relax.Tuple):
             return x[node.args[1]]
         elif isinstance(x, relax.Var):
             if isinstance(x.struct_info, relax.TupleStructInfo):
@@ -2514,7 +2514,7 @@ def _getitem(self, node: fx.Node) -> relax.Var:
             assert isinstance(x.struct_info, relax.TensorStructInfo)
             if isinstance(node.args[1], int):
                 return x
-            if not isinstance(node.args[1], (list, tuple)):
+            if not isinstance(node.args[1], list | tuple):
                 indices = [node.args[1]]
             else:
                 indices = node.args[1]
@@ -2529,7 +2529,7 @@ def _getitem(self, node: fx.Node) -> relax.Var:
             shape = self.shape_of(x)
             non_ellipsis_cnt = 0
             for index in indices:
-                if isinstance(index, (int, slice, torch.fx.Node)):
+                if isinstance(index, int | slice | torch.fx.Node):
                     non_ellipsis_cnt += 1
             for index in indices:
                 if isinstance(index, int):
@@ -2620,5 +2620,5 @@ def _zeros_like(self, node: fx.node) -> relax.Var:
     @abc.abstractmethod
     def create_convert_map(
         self,
-    ) -> Dict[Union[torch.nn.Module, str], Callable[[fx.Node], relax.Var]]:
+    ) -> dict[torch.nn.Module | str, Callable[[fx.Node], relax.Var]]:
         """Create convert map"""
diff --git a/python/tvm/relax/frontend/torch/dynamo.py b/python/tvm/relax/frontend/torch/dynamo.py
index 292d13dda7a0..d21a74acf1c7 100644
--- a/python/tvm/relax/frontend/torch/dynamo.py
+++ b/python/tvm/relax/frontend/torch/dynamo.py
@@ -21,7 +21,6 @@
 """PyTorch Dynamo backend of Relax."""
 
 import functools
-from typing import Optional
 
 import tvm
 from tvm.relax import build as relax_build
@@ -36,7 +35,7 @@ def device_from_inputs(example_inputs):
     return None
 
 
-def relax_dynamo(pipeline: Optional[tvm.transform.Pass] = None):
+def relax_dynamo(pipeline: tvm.transform.Pass | None = None):
     """A helper function to create a relax backend.
 
     Parameters
diff --git a/python/tvm/relax/frontend/torch/exported_program_translator.py b/python/tvm/relax/frontend/torch/exported_program_translator.py
index 39595a9f0043..4c841b00b56d 100644
--- a/python/tvm/relax/frontend/torch/exported_program_translator.py
+++ b/python/tvm/relax/frontend/torch/exported_program_translator.py
@@ -21,8 +21,8 @@
 """PyTorch ExportedProgram of Relax."""
 
 from collections import ChainMap, OrderedDict
+from collections.abc import Callable
 from functools import partial
-from typing import Callable, Dict, List, Optional, Tuple
 
 import torch
 from torch import fx
@@ -298,7 +298,7 @@ def _upsample_impl(
         if size is None:
             shape = self.shape_of(x)
             assert isinstance(shape, relax.ShapeExpr)
-            if isinstance(scale_factor, (tuple, list)):
+            if isinstance(scale_factor, tuple | list):
                 assert len(scale_factor) == len(shape) - 2
                 size = tuple(
                     int(shape[i].value * scale_factor[i - 2]) for i in range(2, len(shape))
@@ -995,7 +995,7 @@ def _randn(self, node: fx.Node) -> relax.Var:
         import numpy as np
 
         args = self.retrieve_args(node)
-        size = args[0] if isinstance(args[0], (list, tuple)) else (args[0],)
+        size = args[0] if isinstance(args[0], list | tuple) else (args[0],)
         dtype = self._convert_data_type(
             node.kwargs.get("dtype", torch.get_default_dtype()), self.env
         )
@@ -1014,7 +1014,7 @@ def _randn_like(self, node: fx.Node) -> relax.Var:
 
     def _zeros(self, node: fx.Node) -> relax.Var:
         args = self.retrieve_args(node)
-        size = relax.ShapeExpr(args[0] if isinstance(args[0], (list, tuple)) else (args[0],))
+        size = relax.ShapeExpr(args[0] if isinstance(args[0], list | tuple) else (args[0],))
         dtype = self._convert_data_type(
             node.kwargs.get("dtype", torch.get_default_dtype()), self.env
         )
@@ -1166,8 +1166,8 @@ def _as_strided(self, node: fx.Node) -> relax.Var:
         assert storage_offset == 0, "as_strided with non-zero storage_offset is not supported yet"
 
         # Only handle view-like cases where the provided strides align with a contiguous layout.
-        can_check = all(isinstance(dim, (int, tvm.tir.IntImm)) for dim in size) and all(
-            isinstance(st, (int, tvm.tir.IntImm)) for st in stride
+        can_check = all(isinstance(dim, int | tvm.tir.IntImm) for dim in size) and all(
+            isinstance(st, int | tvm.tir.IntImm) for st in stride
         )
         if can_check:
             expected_stride = []
@@ -1196,7 +1196,7 @@ def _symbolic_comparison(self, _: fx.Node) -> relax.Expr:
 
     def create_convert_map(
         self,
-    ) -> Dict[str, Callable[[fx.Node], relax.Var]]:
+    ) -> dict[str, Callable[[fx.Node], relax.Var]]:
         import operator
 
         return {
@@ -1531,15 +1531,15 @@ def create_convert_map(
         }
 
     def _process_derived_symbol(
-        self, symbol, torch_symbol_to_relax_var: Dict[str, tvm.tir.Var]
-    ) -> Tuple[str, Optional[tvm.tir.PrimExpr]]:
+        self, symbol, torch_symbol_to_relax_var: dict[str, tvm.tir.Var]
+    ) -> tuple[str, tvm.tir.PrimExpr | None]:
         """Process a sympy symbol to generate a descriptive name and TIR expression."""
         import sympy
 
         if isinstance(symbol, sympy.Symbol):
             return str(symbol), None
 
-        if not isinstance(symbol, (sympy.Add, sympy.Mul)):
+        if not isinstance(symbol, sympy.Add | sympy.Mul):
             return str(symbol), None
 
         tir_expr = None
@@ -1577,11 +1577,11 @@ def _process_derived_symbol(
 
     def create_input_vars(
         self, exported_program: torch.export.ExportedProgram
-    ) -> Tuple[Dict[str, relax.Var], Dict[str, relax.Var], Dict[str, Tuple[int, Optional[int]]]]:
+    ) -> tuple[dict[str, relax.Var], dict[str, relax.Var], dict[str, tuple[int, int | None]]]:
         """Create relax input vars."""
         parameters_buffers_constants = OrderedDict()
         user_inputs = OrderedDict()
-        torch_symbol_to_relax_var: Dict[str, tvm.tir.Var] = {}
+        torch_symbol_to_relax_var: dict[str, tvm.tir.Var] = {}
         range_constraints = {}
 
         if hasattr(exported_program, "range_constraints"):
@@ -1655,9 +1655,7 @@ def from_exported_program(
         keep_params_as_input: bool,
         unwrap_unit_return_tuple: bool,
         no_bind_return_tuple: bool,
-        custom_convert_map: Optional[
-            Dict[str, Callable[[fx.Node, BaseFXGraphImporter], relax.Var]]
-        ],
+        custom_convert_map: dict[str, Callable[[fx.Node, BaseFXGraphImporter], relax.Var]] | None,
     ) -> tvm.IRModule:
         """Convert a PyTorch ExportedProgram to a Relax program."""
 
@@ -1695,7 +1693,7 @@ def from_exported_program(
             if upper_bounds:
                 func_attrs["tir_var_upper_bound"] = upper_bounds
 
-        nodes: List[fx.Node] = exported_program.graph.nodes
+        nodes: list[fx.Node] = exported_program.graph.nodes
 
         # Find all the missing function types
         self._check_unsupported_func_type(nodes)
@@ -1716,7 +1714,7 @@ def from_exported_program(
                     elif node.op == "output":
                         args = self.retrieve_args(node)
                         assert len(args) == 1
-                        assert isinstance(args[0], (tuple, relax.Tuple))
+                        assert isinstance(args[0], tuple | relax.Tuple)
 
                         if unwrap_unit_return_tuple and len(args[0]) == 1:
                             output = self.block_builder.emit_output(args[0][0])
@@ -1774,9 +1772,9 @@ def from_exported_program(
     keep_params_as_input: bool = False,
     unwrap_unit_return_tuple: bool = False,
     no_bind_return_tuple: bool = False,
-    custom_convert_map: Optional[
-        Dict[str, Callable[[fx.Node, BaseFXGraphImporter], relax.Var]]
-    ] = None,
+    custom_convert_map: (
+        dict[str, Callable[[fx.Node, BaseFXGraphImporter], relax.Var]] | None
+    ) = None,
     run_ep_decomposition: bool = True,
 ) -> tvm.IRModule:
     """Convert a PyTorch ExportedProgram to a Relax program
diff --git a/python/tvm/relax/frontend/torch/fx_translator.py b/python/tvm/relax/frontend/torch/fx_translator.py
index 5fd414d3f409..e7fcb0c202b9 100644
--- a/python/tvm/relax/frontend/torch/fx_translator.py
+++ b/python/tvm/relax/frontend/torch/fx_translator.py
@@ -19,8 +19,8 @@
 # pylint: disable=import-outside-toplevel
 """PyTorch FX frontend of Relax."""
 
+from collections.abc import Callable
 from functools import partial, reduce
-from typing import Callable, Dict, List, Optional, Tuple, Union
 
 import tvm
 from tvm import relax
@@ -38,7 +38,7 @@ def __init__(self, default_image_layout: str = "NCHW") -> None:
         import torch  # type: ignore
 
         super().__init__()
-        self.named_modules: Dict[str, torch.Module] = None
+        self.named_modules: dict[str, torch.Module] = None
         self.default_image_layout = default_image_layout
 
     ########## Utilities ##########
@@ -758,7 +758,7 @@ def _getattr(self, node: fx.Node) -> relax.Var:
                 return self.shape_of(self.env[node.args[0]])
         return getattr(self.env[node.args[0]], node.args[1])
 
-    def create_input_vars(self, input_info: List[Tuple[Tuple[int], str]]) -> List[relax.Var]:
+    def create_input_vars(self, input_info: list[tuple[tuple[int], str]]) -> list[relax.Var]:
         inputs = list()
         for idx, (shape, dtype) in enumerate(input_info):
             inputs.append(
@@ -770,7 +770,7 @@ def create_input_vars(self, input_info: List[Tuple[Tuple[int], str]]) -> List[re
 
     def create_convert_map(
         self,
-    ) -> Dict[Union[torch.nn.Module, str], Callable[[fx.Node], relax.Var]]:
+    ) -> dict[torch.nn.Module | str, Callable[[fx.Node], relax.Var]]:
         import operator
 
         from torch import nn
@@ -1041,11 +1041,11 @@ def create_convert_map(
     def from_fx(
         self,
         model,
-        input_info: List[Tuple[Tuple[int], str]],
+        input_info: list[tuple[tuple[int], str]],
         keep_params_as_input: bool,
         unwrap_unit_return_tuple: bool,
         no_bind_return_tuple: bool,
-        custom_convert_map: Optional[dict] = None,
+        custom_convert_map: dict | None = None,
     ) -> tvm.IRModule:
         """Convert a PyTorch FX GraphModule to a Relax program."""
         from torch import fx
@@ -1106,7 +1106,7 @@ def from_fx(
                         assert len(args) == 1
 
                         # return tuple
-                        if isinstance(args[0], (tuple, list, relax.Tuple)):
+                        if isinstance(args[0], tuple | list | relax.Tuple):
                             # unit tuple
                             if unwrap_unit_return_tuple and len(args[0]) == 1:
                                 output = self.block_builder.emit_output(args[0][0])
@@ -1150,12 +1150,12 @@ def from_fx(
 
 def from_fx(
     model,
-    input_info: List[Tuple[Tuple[int], str]],
+    input_info: list[tuple[tuple[int], str]],
     *,
     keep_params_as_input: bool = False,
     unwrap_unit_return_tuple: bool = False,
     no_bind_return_tuple: bool = False,
-    custom_convert_map: Optional[dict] = None,
+    custom_convert_map: dict | None = None,
     default_image_layout: str = "NCHW",
 ) -> tvm.IRModule:
     """Convert a PyTorch FX GraphModule to a Relax program
diff --git a/python/tvm/relax/op/_op_gradient.py b/python/tvm/relax/op/_op_gradient.py
index 42aabec9d858..3cab1ece9e96 100644
--- a/python/tvm/relax/op/_op_gradient.py
+++ b/python/tvm/relax/op/_op_gradient.py
@@ -19,7 +19,6 @@
 
 import functools
 import operator
-from typing import List
 
 from tvm import relax
 from tvm.arith import Analyzer
@@ -131,7 +130,7 @@ def add_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of add.
 
     Forward Form:
@@ -152,7 +151,7 @@ def subtract_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of subtract.
 
     Forward Form:
@@ -173,7 +172,7 @@ def multiply_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of multiply.
 
     Forward Form:
@@ -195,7 +194,7 @@ def divide_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of divide.
 
     Forward Form:
@@ -217,7 +216,7 @@ def power_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of power.
 
     Forward Form:
@@ -242,7 +241,7 @@ def maximum_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of maximum.
 
     Forward Form:
@@ -266,7 +265,7 @@ def minimum_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of minimum.
 
     Forward Form:
@@ -295,7 +294,7 @@ def equal_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     return [no_grad(orig_call.args[0]), no_grad(orig_call.args[1])]
 
 
@@ -305,7 +304,7 @@ def greater_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     return [no_grad(orig_call.args[0]), no_grad(orig_call.args[1])]
 
 
@@ -315,7 +314,7 @@ def greater_equal_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     return [no_grad(orig_call.args[0]), no_grad(orig_call.args[1])]
 
 
@@ -325,7 +324,7 @@ def less_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     return [no_grad(orig_call.args[0]), no_grad(orig_call.args[1])]
 
 
@@ -335,7 +334,7 @@ def less_equal_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     return [no_grad(orig_call.args[0]), no_grad(orig_call.args[1])]
 
 
@@ -345,7 +344,7 @@ def not_equal_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     return [no_grad(orig_call.args[0]), no_grad(orig_call.args[1])]
 
 
@@ -360,7 +359,7 @@ def zeros_like_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     return [no_grad(orig_call.args[0])]
 
 
@@ -370,7 +369,7 @@ def ones_like_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     return [no_grad(orig_call.args[0])]
 
 
@@ -380,7 +379,7 @@ def full_like_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     return [no_grad(orig_call.args[0]), no_grad(orig_call.args[1])]
 
 
@@ -390,7 +389,7 @@ def zeros_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     return [no_grad(orig_call.args[0])]
 
 
@@ -400,7 +399,7 @@ def ones_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     return [no_grad(orig_call.args[0])]
 
 
@@ -410,7 +409,7 @@ def full_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     return [no_grad(orig_call.args[0]), no_grad(orig_call.args[1])]
 
 
@@ -423,7 +422,7 @@ def triu_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of triu.
 
     Forward Form:
@@ -445,7 +444,7 @@ def abs_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of abs.
 
     Forward Form:
@@ -466,7 +465,7 @@ def cos_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of cos.
 
     Forward Form:
@@ -484,7 +483,7 @@ def exp_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of exp.
 
     Forward Form:
@@ -502,7 +501,7 @@ def log_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of log.
 
     Forward Form:
@@ -520,7 +519,7 @@ def negative_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of negative.
 
     Forward Form:
@@ -538,7 +537,7 @@ def sigmoid_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of sigmoid.
 
     Forward Form:
@@ -557,7 +556,7 @@ def sin_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of sin.
 
     Forward Form:
@@ -575,7 +574,7 @@ def sqrt_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of sqrt.
 
     Forward Form:
@@ -595,7 +594,7 @@ def tanh_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of tanh.
 
     Forward Form:
@@ -617,7 +616,7 @@ def sum_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of sum.
 
     Forward Form:
@@ -641,7 +640,7 @@ def mean_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of mean.
 
     Forward Form:
@@ -668,7 +667,7 @@ def variance_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of variance.
 
     Forward Form:
@@ -700,7 +699,7 @@ def permute_dims_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of permute_dims.
 
     Forward Form:
@@ -725,7 +724,7 @@ def concat_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of concat.
 
     Forward Form:
@@ -737,7 +736,7 @@ def concat_grad(
     axis = orig_call.attrs.axis
     assert axis is not None
     axis = int(axis)
-    split_indices: List[PrimExpr] = []
+    split_indices: list[PrimExpr] = []
     sinfo = orig_call.args[0].struct_info
     assert isinstance(sinfo, relax.TupleStructInfo)
     for i in range(len(sinfo.fields) - 1):
@@ -757,7 +756,7 @@ def split_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of split.
 
     Forward Form:
@@ -777,7 +776,7 @@ def expand_dims_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of expand_dims.
 
     Forward Form:
@@ -795,7 +794,7 @@ def reshape_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of reshape.
 
     Forward Form:
@@ -818,7 +817,7 @@ def cumsum_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of cumsum.
 
     Forward Form:
@@ -852,7 +851,7 @@ def broadcast_to_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of broadcast_to.
 
     Forward Form:
@@ -878,7 +877,7 @@ def take_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of take.
 
     Forward Form:
@@ -907,7 +906,7 @@ def where_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of where.
 
     Forward Form:
@@ -939,7 +938,7 @@ def matmul_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of matmul.
 
     Forward Form:
@@ -999,7 +998,7 @@ def astype_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of astype.
 
     Forward Form:
@@ -1020,7 +1019,7 @@ def relu_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of relu.
 
     Forward Form:
@@ -1040,7 +1039,7 @@ def silu_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of silu.
 
     Forward Form:
@@ -1061,7 +1060,7 @@ def softmax_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of softmax.
 
     Forward Form:
@@ -1079,7 +1078,7 @@ def log_softmax_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of log_softmax.
 
     Forward Form:
@@ -1098,7 +1097,7 @@ def cross_entropy_with_logits_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of cross_entropy_with_logits.
 
     Forward Form:
@@ -1169,7 +1168,7 @@ def conv2d_grad(
     orig_call: Call,
     output_grad: Var,
     ctx: BlockBuilder,
-) -> List[Expr]:
+) -> list[Expr]:
     """Gradient of conv2d. Now only supports `NCHW` data layout and `OIHW` kernel layout.
 
     Forward Form:
diff --git a/python/tvm/relax/op/base.py b/python/tvm/relax/op/base.py
index d3db6655a207..10a4eb76b310 100644
--- a/python/tvm/relax/op/base.py
+++ b/python/tvm/relax/op/base.py
@@ -17,7 +17,7 @@
 # ruff: noqa: F821
 """The base Relax operators."""
 
-from typing import Callable, Dict, List, Optional, Tuple, Union
+from collections.abc import Callable
 
 import tvm
 import tvm.runtime
@@ -35,7 +35,7 @@
 
 def register_gradient(
     op_name: str,
-    fgradient: Optional[Callable[[Var, Call, Var, "BlockBuilder"], List[Expr]]] = None,
+    fgradient: Callable[[Var, Call, Var, "BlockBuilder"], list[Expr]] | None = None,
     level: int = 10,
 ):
     """Register operator gradient function for a relax operator.
@@ -93,8 +93,8 @@ def _wrap_inline_arg_tuple(args) -> Expr:
 def call_tir(
     gvar: GlobalVar,
     args: Expr,
-    out_sinfo: Union[TensorStructInfo, List[TensorStructInfo]],
-    tir_vars: Optional[Union[ShapeExpr, Tuple[PrimExpr], List[PrimExpr]]] = None,
+    out_sinfo: TensorStructInfo | list[TensorStructInfo],
+    tir_vars: ShapeExpr | tuple[PrimExpr] | list[PrimExpr] | None = None,
 ) -> Call:
     """
     Call a tir.prim_func and return the output.
@@ -125,7 +125,7 @@ def call_tir(
     if not isinstance(out_sinfo, list):
         out_sinfo = [out_sinfo]
 
-    if isinstance(tir_vars, (list, tuple)):
+    if isinstance(tir_vars, list | tuple):
         tir_vars = ShapeExpr(tir_vars)
 
     return _ffi_api.call_tir(gvar, args, out_sinfo, tir_vars)  # type: ignore
@@ -135,10 +135,10 @@ def call_tir(
 def call_tir_with_grad(
     gvar: GlobalVar,
     args: Expr,
-    out_sinfo: Union[TensorStructInfo, List[TensorStructInfo]],
+    out_sinfo: TensorStructInfo | list[TensorStructInfo],
     te_grad_name: str,
-    te_grad_kwargs: Optional[Dict[str, Object]] = None,
-    tir_vars: Optional[Union[ShapeExpr, Tuple[PrimExpr], List[PrimExpr]]] = None,
+    te_grad_kwargs: dict[str, Object] | None = None,
+    tir_vars: ShapeExpr | tuple[PrimExpr] | list[PrimExpr] | None = None,
 ) -> Call:
     """
     Call a tir.prim_func and return the output. This intrinsic will bind a te gradient function
@@ -179,7 +179,7 @@ def call_tir_with_grad(
     if not isinstance(out_sinfo, list):
         out_sinfo = [out_sinfo]
 
-    if isinstance(tir_vars, (list, tuple)):
+    if isinstance(tir_vars, list | tuple):
         tir_vars = ShapeExpr(tir_vars)
 
     if te_grad_kwargs is None:
@@ -194,9 +194,9 @@ def call_tir_with_grad(
 def call_tir_inplace(
     gvar: GlobalVar,
     args: Expr,
-    inplace_indices: Union[int, List[int]],
-    out_sinfo: Union[TensorStructInfo, List[TensorStructInfo]],
-    tir_vars: Optional[Union[ShapeExpr, Tuple[PrimExpr], List[PrimExpr]]] = None,
+    inplace_indices: int | list[int],
+    out_sinfo: TensorStructInfo | list[TensorStructInfo],
+    tir_vars: ShapeExpr | tuple[PrimExpr] | list[PrimExpr] | None = None,
 ) -> Call:
     """
     Call a TIR PrimFunc and return the result, doing the specified computations in-place
@@ -249,7 +249,7 @@ def call_tir_inplace(
     if not isinstance(out_sinfo, list):
         out_sinfo = [out_sinfo]
 
-    if isinstance(tir_vars, (list, tuple)):
+    if isinstance(tir_vars, list | tuple):
         tir_vars = ShapeExpr(tir_vars)
 
     return _ffi_api.call_tir_inplace(  # type: ignore
@@ -263,9 +263,9 @@ def call_tir_inplace(
 
 @args_converter.auto
 def call_dps_packed(
-    func: Union[str, Expr],
+    func: str | Expr,
     args: Expr,
-    out_sinfo: Union[TensorStructInfo, List[TensorStructInfo]],
+    out_sinfo: TensorStructInfo | list[TensorStructInfo],
 ) -> Call:
     """
     Call a destination-passing-style packed function and return the output.
@@ -307,7 +307,7 @@ def call_dps_packed(
 def call_py_func(
     func_name: str,
     args: Expr,
-    out_sinfo: Union[TensorStructInfo, List[TensorStructInfo]],
+    out_sinfo: TensorStructInfo | list[TensorStructInfo],
 ) -> Call:
     """
     Call a Python function and return the output.
@@ -341,10 +341,10 @@ def call_py_func(
 
 @args_converter.auto
 def call_builtin_with_ctx(
-    func: Union[str, Expr],
+    func: str | Expr,
     args: Expr,
     *,
-    sinfo_args: Optional[Union[StructInfo, List[StructInfo]]] = None,
+    sinfo_args: StructInfo | list[StructInfo] | None = None,
 ) -> Call:
     """Call a builtin function func.
 
@@ -367,7 +367,7 @@ def call_builtin_with_ctx(
     if isinstance(func, str):
         func = ExternFunc(func)
 
-    if sinfo_args is not None and not isinstance(sinfo_args, (list, tuple)):
+    if sinfo_args is not None and not isinstance(sinfo_args, list | tuple):
         sinfo_args = [sinfo_args]
 
     return _ffi_api.call_builtin_with_ctx(  # type: ignore
@@ -407,7 +407,7 @@ def make_closure(
 def invoke_closure(
     closure: Expr,
     args: Expr,
-    sinfo_args: Union[List[StructInfo], StructInfo],
+    sinfo_args: list[StructInfo] | StructInfo,
 ) -> Call:
     """
     Invoke a closure.
@@ -429,7 +429,7 @@ def invoke_closure(
         A call to `invoke_closure`.
     """
 
-    if not isinstance(sinfo_args, (list, tuple)):
+    if not isinstance(sinfo_args, list | tuple):
         sinfo_args = [sinfo_args]
 
     return _ffi_api.invoke_closure(closure, args, sinfo_args)  # type: ignore
@@ -496,7 +496,7 @@ def relax_print(format_str: str, *format_args: tvm.Object) -> None:
         py_print(format_str.format(*val_strs))
 
 
-def print(*values: List[Expr], format: Union[str, Expr] = "") -> Expr:
+def print(*values: list[Expr], format: str | Expr = "") -> Expr:
     """Print op to print the values
 
     Parameters
@@ -547,7 +547,7 @@ def relax_assert_op(condition: tvm.Object, format_str: str, *format_args: tvm.Ob
             f"The format string argument to assert must be a string, given {type(format_str)})"
         )
 
-    if isinstance(condition, (bool, int)):
+    if isinstance(condition, bool | int):
         val = condition
     elif isinstance(condition, tvm.runtime.Tensor):
         # may happen if the original program had unknown shape or dtype for the tensor's type
@@ -579,9 +579,9 @@ def relax_assert_op(condition: tvm.Object, format_str: str, *format_args: tvm.Ob
 
 
 def assert_op(
-    condition: Union[Expr, PrimExpr],
-    format_args: Optional[Union[Expr, List[Expr]]] = None,
-    format: Union[str, Expr] = "",
+    condition: Expr | PrimExpr,
+    format_args: Expr | list[Expr] | None = None,
+    format: str | Expr = "",
 ) -> Expr:
     """
     Create a call to Relax's assert_op operation (`assert` is reserved in Python,
@@ -679,10 +679,10 @@ def shape_to_tensor(expr: Expr) -> Expr:
 
 @args_converter.auto
 def call_inplace_packed(
-    func: Union[str, ExternFunc, GlobalVar],
+    func: str | ExternFunc | GlobalVar,
     *args: Expr,
-    inplace_indices: Union[int, List[int]],
-    sinfo_args: Union[StructInfo, List[StructInfo]],
+    inplace_indices: int | list[int],
+    sinfo_args: StructInfo | list[StructInfo],
 ) -> Expr:
     """
     Construct a call to a packed function that consumes some of its arguments "in-place"
@@ -745,9 +745,9 @@ def call_inplace_packed(
 
 @args_converter.auto
 def call_pure_packed(
-    func: Union[str, ExternFunc, GlobalVar],
+    func: str | ExternFunc | GlobalVar,
     *args: Expr,
-    sinfo_args: Union[StructInfo, List[StructInfo]],
+    sinfo_args: StructInfo | list[StructInfo],
 ) -> Expr:
     """
     Construct a call to a packed function that should be treated as pure,
@@ -811,7 +811,7 @@ def call_pure_packed(
 def invoke_pure_closure(
     closure: Expr,
     args: Expr,
-    sinfo_args: Union[List[StructInfo], StructInfo],
+    sinfo_args: list[StructInfo] | StructInfo,
 ) -> Call:
     """
     Invoke a closure and indicate to the compiler that it is pure.
@@ -839,7 +839,7 @@ def invoke_pure_closure(
         A call to `invoke_pure_closure`.
     """
 
-    if not isinstance(sinfo_args, (list, tuple)):
+    if not isinstance(sinfo_args, list | tuple):
         sinfo_args = [sinfo_args]
 
     return _ffi_api.invoke_pure_closure(closure, args, sinfo_args)  # type: ignore
diff --git a/python/tvm/relax/op/builtin/builtin.py b/python/tvm/relax/op/builtin/builtin.py
index b0d04ac74f2a..47fab9c7a5c9 100644
--- a/python/tvm/relax/op/builtin/builtin.py
+++ b/python/tvm/relax/op/builtin/builtin.py
@@ -15,8 +15,6 @@
 # specific language governing permissions and limitations
 """The builtin Relax operators."""
 
-from typing import Union
-
 from ...expr import Call, DataTypeImm, Expr, PrimValue, StringImm
 from ...utils import args_converter
 from . import _ffi_api
@@ -25,9 +23,9 @@
 @args_converter.auto
 def alloc_tensor(
     shape: Expr,
-    dtype: Union[str, Expr],
-    runtime_device_index: Union[int, Expr],
-    storage_scope: Union[str, Expr] = "global",
+    dtype: str | Expr,
+    runtime_device_index: int | Expr,
+    storage_scope: str | Expr = "global",
 ) -> Call:
     """Construct a Call to allocate a tensor with specific shape, dtype, runtime_device_index.
 
diff --git a/python/tvm/relax/op/create.py b/python/tvm/relax/op/create.py
index 1e07338a3501..9d28ed92f9c5 100644
--- a/python/tvm/relax/op/create.py
+++ b/python/tvm/relax/op/create.py
@@ -16,21 +16,19 @@
 # under the License.
 """Creation operators."""
 
-from typing import Optional, Tuple, Union
-
 from tvm import DataType, DataTypeCode
 from tvm.ir.expr import PrimExpr
 
 from ..expr import Expr, PrimValue, ShapeExpr
 from . import _ffi_api
 
-PrimExprLike = Union[int, PrimExpr]
+PrimExprLike = int | PrimExpr
 
 
 def full(
-    shape: Union[Tuple[PrimExprLike], Expr],
+    shape: tuple[PrimExprLike] | Expr,
     fill_value: Expr,
-    dtype: Optional[Union[str, DataType]] = None,
+    dtype: str | DataType | None = None,
 ) -> Expr:
     """Fill array with scalar value.
 
@@ -42,7 +40,7 @@ def full(
     fill_value : relax.Expr
         The value to fill. Must be a scalar tensor.
 
-    dtype : Optional[Union[str, DataType]]
+    dtype : Optional[str | DataType]
         The data type of the created tensor.
         If dtype is not given, it will by default use the dtype of fill_value.
 
@@ -54,7 +52,7 @@ def full(
     return _ffi_api.full(shape, fill_value, dtype)  # type: ignore
 
 
-def full_like(x: Expr, fill_value: Expr, dtype: Optional[Union[str, DataType]] = None) -> Expr:
+def full_like(x: Expr, fill_value: Expr, dtype: str | DataType | None = None) -> Expr:
     """Construct a tensor such that
     - its shape is the same as the input data tensor's shape,
     - its value is filled with the input scalar fill value.
@@ -68,7 +66,7 @@ def full_like(x: Expr, fill_value: Expr, dtype: Optional[Union[str, DataType]] =
     fill_value : relax.Expr
         The value to fill. Must be a scalar tensor.
 
-    dtype : Optional[Union[str, DataType]]
+    dtype : Optional[str | DataType]
         The data type of the created tensor.
         If dtype is not given, it will by default use the dtype of the input tensor.
 
@@ -80,7 +78,7 @@ def full_like(x: Expr, fill_value: Expr, dtype: Optional[Union[str, DataType]] =
     return _ffi_api.full_like(x, fill_value, dtype)  # type: ignore
 
 
-def ones(shape: Union[Tuple[PrimExprLike], Expr], dtype: Union[str, DataType]) -> Expr:
+def ones(shape: tuple[PrimExprLike] | Expr, dtype: str | DataType) -> Expr:
     """Construct a tensor of all ones, with the input shape and dtype.
 
     Parameters
@@ -88,7 +86,7 @@ def ones(shape: Union[Tuple[PrimExprLike], Expr], dtype: Union[str, DataType]) -
     shape : Union[Tuple[PrimExprLike], Expr]
         The shape of the created tensor.
 
-    dtype : Union[str, DataType]
+    dtype : str | DataType
         The data type of the created tensor.
 
     Returns
@@ -96,12 +94,12 @@ def ones(shape: Union[Tuple[PrimExprLike], Expr], dtype: Union[str, DataType]) -
     result : relax.Expr
         The result tensor.
     """
-    if isinstance(shape, (tuple, list)):
+    if isinstance(shape, tuple | list):
         shape = ShapeExpr(shape)
     return _ffi_api.ones(shape, dtype)  # type: ignore
 
 
-def ones_like(x: Expr, dtype: Optional[Union[str, DataType]] = None) -> Expr:
+def ones_like(x: Expr, dtype: str | DataType | None = None) -> Expr:
     """Construct a tensor with all ones, with shape of the input tensor shape.
 
     Parameters
@@ -110,7 +108,7 @@ def ones_like(x: Expr, dtype: Optional[Union[str, DataType]] = None) -> Expr:
         The input tensor, which provides the shape, and dtype
         when the `dtype` field is not specified.
 
-    dtype : Optional[Union[str, DataType]]
+    dtype : Optional[str | DataType]
         The data type of the created tensor.
         If dtype is not given, it will by default use the dtype of the input tensor.
 
@@ -122,7 +120,7 @@ def ones_like(x: Expr, dtype: Optional[Union[str, DataType]] = None) -> Expr:
     return _ffi_api.ones_like(x, dtype)  # type: ignore
 
 
-def zeros(shape: Union[Tuple[PrimExprLike], Expr], dtype: Union[str, DataType]) -> Expr:
+def zeros(shape: tuple[PrimExprLike] | Expr, dtype: str | DataType) -> Expr:
     """Construct a tensor of all zeros, with the input shape and dtype.
 
     Parameters
@@ -130,7 +128,7 @@ def zeros(shape: Union[Tuple[PrimExprLike], Expr], dtype: Union[str, DataType])
     shape : Union[Tuple[PrimExprLike], Expr]
         The shape of the created tensor.
 
-    dtype : Union[str, DataType]
+    dtype : str | DataType
         The data type of the created tensor.
 
     Returns
@@ -138,12 +136,12 @@ def zeros(shape: Union[Tuple[PrimExprLike], Expr], dtype: Union[str, DataType])
     result : relax.Expr
         The result tensor.
     """
-    if isinstance(shape, (tuple, list)):
+    if isinstance(shape, tuple | list):
         shape = ShapeExpr(shape)
     return _ffi_api.zeros(shape, dtype)  # type: ignore
 
 
-def zeros_like(x: Expr, dtype: Optional[Union[str, DataType]] = None) -> Expr:
+def zeros_like(x: Expr, dtype: str | DataType | None = None) -> Expr:
     """Construct a tensor with all zeros, with shape of the input tensor shape.
 
     Parameters
@@ -152,7 +150,7 @@ def zeros_like(x: Expr, dtype: Optional[Union[str, DataType]] = None) -> Expr:
         The input tensor, which provides the shape, and dtype
         when the `dtype` field is not specified.
 
-    dtype : Optional[Union[str, DataType]]
+    dtype : Optional[str | DataType]
         The data type of the created tensor.
         If dtype is not given, it will by default use the dtype of the input tensor.
 
@@ -165,27 +163,27 @@ def zeros_like(x: Expr, dtype: Optional[Union[str, DataType]] = None) -> Expr:
 
 
 def eye(
-    n: Union[PrimExprLike, PrimValue],
-    m: Optional[Union[PrimExprLike, PrimValue]] = None,
-    k: Union[PrimExprLike, PrimValue] = 0,
-    dtype: Union[str, DataType] = "float32",
+    n: PrimExprLike | PrimValue,
+    m: PrimExprLike | PrimValue | None = None,
+    k: PrimExprLike | PrimValue = 0,
+    dtype: str | DataType = "float32",
 ) -> Expr:
     """Construct a 2-D tensor with ones on the diagonal and zeros elsewhere.
 
     Parameters
     ----------
-    n : Union[PrimExprLike, PrimValue]
+    n : PrimExprLike | PrimValue
         Number of rows in the output.
 
-    m : Optional[Union[PrimExprLike, PrimValue]]
+    m : Optional[PrimExprLike | PrimValue]
         Number of columns in the output. If None, defaults to n.
 
-    k : Union[PrimExprLike, PrimValue]
+    k : PrimExprLike | PrimValue
         Index of the diagonal: 0 (the default) refers to the main diagonal,
         a positive value refers to an upper diagonal, and a negative value
         to a lower diagonal.
 
-    dtype : Union[str, DataType]
+    dtype : str | DataType
         The data type of the created tensor.
 
     Returns
@@ -202,8 +200,8 @@ def eye(
 
 def eye_like(
     x: Expr,
-    k: Union[PrimExprLike, PrimValue] = 0,
-    dtype: Optional[Union[str, DataType]] = None,
+    k: PrimExprLike | PrimValue = 0,
+    dtype: str | DataType | None = None,
 ) -> Expr:
     """Return a 2-D tensor with ones on the diagonal and zeros elsewhere,
     with the same shape as the input tensor.
@@ -214,12 +212,12 @@ def eye_like(
         The input tensor, which provides the shape, and dtype
         when the `dtype` field is not specified.
 
-    k : Union[PrimExprLike, PrimValue]
+    k : PrimExprLike | PrimValue
         Index of the diagonal: 0 (the default) refers to the main diagonal,
         a positive value refers to an upper diagonal, and a negative value
         to a lower diagonal.
 
-    dtype : Optional[Union[str, DataType]]
+    dtype : Optional[str | DataType]
         The data type of the created tensor.
         If dtype is not given, it will by default use the dtype of the input tensor.
 
@@ -233,26 +231,26 @@ def eye_like(
 
 
 def arange(
-    start: Union[PrimExprLike, PrimValue],
-    end: Optional[Union[PrimExprLike, PrimValue]] = None,
-    step: Union[PrimExprLike, PrimValue] = 1,
-    dtype: Optional[Union[str, DataType]] = None,
+    start: PrimExprLike | PrimValue,
+    end: PrimExprLike | PrimValue | None = None,
+    step: PrimExprLike | PrimValue = 1,
+    dtype: str | DataType | None = None,
 ) -> Expr:
     """Construct a tensor with evenly spaced elements.
 
     Parameters
     ----------
-    start : Union[PrimExprLike,PrimValue]
+    start : PrimExprLike | PrimValue
         The start of the interval.
 
-    end : Optional[Union[PrimExprLike,PrimValue]]
+    end : Optional[PrimExprLike | PrimValue]
         The end of the interval. If not given, it will be set to start,
         and start will be set to 0.
 
-    step : Union[PrimExprLike,PrimValue]
+    step : PrimExprLike | PrimValue
         The step size.
 
-    dtype : Optional[Union[str, DataType]]
+    dtype : Optional[str | DataType]
         The data type of the created tensor.
 
     Returns
@@ -317,7 +315,7 @@ def hamming_window(window_size, periodic, alpha, beta, dtype):
     return _ffi_api.hamming_window(window_size, periodic, alpha, beta, dtype)
 
 
-def tril(x: Expr, k: Union[int, PrimExpr, Expr] = 0) -> Expr:
+def tril(x: Expr, k: int | PrimExpr | Expr = 0) -> Expr:
     """Return the lower triangular part of a matrix or a batch of matrices.
 
     Parameters
@@ -343,7 +341,7 @@ def tril(x: Expr, k: Union[int, PrimExpr, Expr] = 0) -> Expr:
     return _ffi_api.tril(x, k)  # type: ignore
 
 
-def triu(x: Expr, k: Union[int, PrimExpr, Expr] = 0) -> Expr:
+def triu(x: Expr, k: int | PrimExpr | Expr = 0) -> Expr:
     """Return the upper triangular part of a matrix or a batch of matrices.
 
     Parameters
diff --git a/python/tvm/relax/op/datatype.py b/python/tvm/relax/op/datatype.py
index ec01d3fbf361..8c1aa1d91ea2 100644
--- a/python/tvm/relax/op/datatype.py
+++ b/python/tvm/relax/op/datatype.py
@@ -16,15 +16,13 @@
 # under the License.
 """Datatype operators."""
 
-from typing import Union
-
 from tvm import DataType
 
 from ..expr import Expr
 from . import _ffi_api
 
 
-def astype(x: Expr, dtype: Union[str, DataType]) -> Expr:
+def astype(x: Expr, dtype: str | DataType) -> Expr:
     """Cast input tensor to the given data type.
 
     Parameters
@@ -43,7 +41,7 @@ def astype(x: Expr, dtype: Union[str, DataType]) -> Expr:
     return _ffi_api.astype(x, dtype)  # type: ignore
 
 
-def wrap_param(data: Expr, dtype: Union[str, DataType] = "float32") -> Expr:
+def wrap_param(data: Expr, dtype: str | DataType = "float32") -> Expr:
     """Cast input tensor which is model param to data type if the dtype of the input data is not
     the same as the given dtype.
     Parameters
diff --git a/python/tvm/relax/op/distributed/distributed.py b/python/tvm/relax/op/distributed/distributed.py
index 9909fc659a92..b17708223d8b 100644
--- a/python/tvm/relax/op/distributed/distributed.py
+++ b/python/tvm/relax/op/distributed/distributed.py
@@ -17,8 +17,6 @@
 # pylint: disable=redefined-builtin
 """Operators for distributed Relax."""
 
-from typing import List, Optional, Tuple, Union
-
 from tvm.ir import PrimExpr
 from tvm.relax.distributed import DTensorStructInfo
 from tvm.relax.distributed.struct_info import DeviceMesh, Placement
@@ -72,8 +70,8 @@ def redistribute(input: Expr, device_mesh: DeviceMesh, placement: Placement) ->
 def call_tir_local_view(
     gvar: GlobalVar,
     args: Expr,
-    out_sinfo: Union[DTensorStructInfo, List[DTensorStructInfo]],
-    tir_vars: Optional[Union[ShapeExpr, Tuple[PrimExpr], List[PrimExpr]]] = None,
+    out_sinfo: DTensorStructInfo | list[DTensorStructInfo],
+    tir_vars: ShapeExpr | tuple[PrimExpr] | list[PrimExpr] | None = None,
 ) -> Call:
     """
     Call a tir.prim_func and return the output. The prim_func should be a worker-local function
@@ -107,7 +105,7 @@ def call_tir_local_view(
     if not isinstance(out_sinfo, list):
         out_sinfo = [out_sinfo]
 
-    if isinstance(tir_vars, (list, tuple)):
+    if isinstance(tir_vars, list | tuple):
         tir_vars = ShapeExpr(tir_vars)
 
     return _ffi_api.call_tir_local_view(gvar, args, out_sinfo, tir_vars)  # type: ignore
diff --git a/python/tvm/relax/op/grad/grad.py b/python/tvm/relax/op/grad/grad.py
index 06e3f924377e..45ceda6fdbcc 100644
--- a/python/tvm/relax/op/grad/grad.py
+++ b/python/tvm/relax/op/grad/grad.py
@@ -22,8 +22,6 @@
 struct_info of the input.
 """
 
-from typing import Optional, Tuple
-
 from ...expr import Expr
 from . import _ffi_api
 
@@ -101,7 +99,7 @@ def nll_loss_backward(
     output_grad: Expr,
     predictions: Expr,
     targets: Expr,
-    weights: Optional[Expr] = None,
+    weights: Expr | None = None,
     reduction: str = "mean",
     ignore_index: int = -100,
 ) -> Expr:
@@ -126,14 +124,14 @@ def nll_loss_backward(
 def max_pool2d_backward(
     output_grad: Expr,
     data: Expr,
-    pool_size: Tuple[int, int] = (1, 1),
-    strides: Tuple[int, int] = (1, 1),
-    padding: Tuple[int, int, int, int] = (0, 0, 0, 0),
-    dilation: Tuple[int, int] = (1, 1),
+    pool_size: tuple[int, int] = (1, 1),
+    strides: tuple[int, int] = (1, 1),
+    padding: tuple[int, int, int, int] = (0, 0, 0, 0),
+    dilation: tuple[int, int] = (1, 1),
     ceil_mode: bool = False,
     count_include_pad: bool = False,
     layout: str = "NCHW",
-    out_layout: Optional[str] = None,
+    out_layout: str | None = None,
 ) -> Expr:
     """Backward operator of relax.nn.max_pool2d. All parameters except output_grad is the same as
     relax.nn.max_pool2d. Returns the gradient w.r.t. data.
@@ -165,14 +163,14 @@ def max_pool2d_backward(
 def avg_pool2d_backward(
     output_grad: Expr,
     data: Expr,
-    pool_size: Tuple[int, int] = (1, 1),
-    strides: Tuple[int, int] = (1, 1),
-    padding: Tuple[int, int, int, int] = (0, 0, 0, 0),
-    dilation: Tuple[int, int] = (1, 1),
+    pool_size: tuple[int, int] = (1, 1),
+    strides: tuple[int, int] = (1, 1),
+    padding: tuple[int, int, int, int] = (0, 0, 0, 0),
+    dilation: tuple[int, int] = (1, 1),
     ceil_mode: bool = False,
     count_include_pad: bool = False,
     layout: str = "NCHW",
-    out_layout: Optional[str] = None,
+    out_layout: str | None = None,
 ) -> Expr:
     """Backward operator of relax.nn.avg_pool2d. All parameters except output_grad is the same as
     relax.nn.avg_pool2d. Returns the gradient w.r.t. data.
@@ -201,7 +199,7 @@ def avg_pool2d_backward(
     )
 
 
-def take_backward(output_grad: Expr, x: Expr, indices: Expr, axis: Optional[int] = None) -> Expr:
+def take_backward(output_grad: Expr, x: Expr, indices: Expr, axis: int | None = None) -> Expr:
     """Backward operator of relax.take. All parameters except output_grad is the same as
     relax.take. Returns the gradient w.r.t. x.
 
diff --git a/python/tvm/relax/op/image/image.py b/python/tvm/relax/op/image/image.py
index cc1ca142870b..5e48bb65221a 100644
--- a/python/tvm/relax/op/image/image.py
+++ b/python/tvm/relax/op/image/image.py
@@ -16,21 +16,19 @@
 # under the License.
 """Image operators."""
 
-from typing import Optional, Tuple, Union
-
 from tvm import DataType
 from tvm.ir.expr import PrimExpr
 
 from ...expr import Expr, ShapeExpr
 from . import _ffi_api
 
-PrimExprLike = Union[int, PrimExpr]
+PrimExprLike = int | PrimExpr
 
 
 def resize2d(
     data: Expr,
-    size: Union[Expr, PrimExprLike, Tuple[PrimExprLike]],
-    roi: Optional[Union[float, Tuple[float]]] = None,
+    size: Expr | PrimExprLike | tuple[PrimExprLike],
+    roi: float | tuple[float] | None = None,
     layout: str = "NCHW",
     method: str = "linear",
     coordinate_transformation_mode: str = "half_pixel",
@@ -38,7 +36,7 @@ def resize2d(
     cubic_alpha: float = -0.75,
     cubic_exclude: int = 0,
     extrapolation_value: float = 0.0,
-    out_dtype: Optional[Union[str, DataType]] = None,
+    out_dtype: str | DataType | None = None,
 ) -> Expr:
     """Image resize2d operator.
 
@@ -91,7 +89,7 @@ def resize2d(
     extrapolation_value: float
         Fill value to use when roi is outside of the image
 
-    out_dtype : Optional[Union[str, DataType]]
+    out_dtype : Optional[str | DataType]
         The dtype of the output tensor.
         It it is not specified, the output will have the same dtype as input if not specified.
 
@@ -104,14 +102,14 @@ def resize2d(
         roi = (0.0, 0.0, 0.0, 0.0)  # type: ignore
     elif isinstance(roi, float):
         roi = (roi, roi, roi, roi)  # type: ignore
-    elif isinstance(roi, (tuple, list)):
+    elif isinstance(roi, tuple | list):
         roi = tuple(val if isinstance(val, float) else float(val) for val in roi)
     else:
         raise NotImplementedError(f"Unsupported roi type {type(roi)}")
 
-    if isinstance(size, (int, PrimExpr)):
+    if isinstance(size, int | PrimExpr):
         size = (size, size)
-    if isinstance(size, (tuple, list)):
+    if isinstance(size, tuple | list):
         if len(size) == 1:
             size = ShapeExpr([size[0], size[0]])
         else:
diff --git a/python/tvm/relax/op/index.py b/python/tvm/relax/op/index.py
index 640777ed2ae5..1edb807db840 100644
--- a/python/tvm/relax/op/index.py
+++ b/python/tvm/relax/op/index.py
@@ -16,18 +16,16 @@
 # under the License.
 """Indexing operators."""
 
-from typing import Optional, Union
-
 from tvm.ir.expr import PrimExpr
 
 from .. import args_converter
 from ..expr import Expr
 from . import _ffi_api
 
-PrimExprLike = Union[int, PrimExpr]
+PrimExprLike = int | PrimExpr
 
 
-def take(x: Expr, indices: Expr, axis: Optional[int] = None, mode: str = "fast") -> Expr:
+def take(x: Expr, indices: Expr, axis: int | None = None, mode: str = "fast") -> Expr:
     """Take elements from a tensor along an axis.
     Its semantic is mostly similar to `numpy.take`
     (https://numpy.org/doc/stable/reference/generated/numpy.take.html),
@@ -66,7 +64,7 @@ def strided_slice(
     axes: Expr,
     begin: Expr,
     end: Expr,
-    strides: Optional[Expr] = None,
+    strides: Expr | None = None,
     assume_inbound: bool = False,
 ) -> Expr:
     """Strided slice of a tensor.
diff --git a/python/tvm/relax/op/linear_algebra.py b/python/tvm/relax/op/linear_algebra.py
index 323c2bb22920..8cc9e4def8f4 100644
--- a/python/tvm/relax/op/linear_algebra.py
+++ b/python/tvm/relax/op/linear_algebra.py
@@ -17,8 +17,6 @@
 # pylint: disable=invalid-name
 """Relax linear algebra operators"""
 
-from typing import Optional, Union
-
 from tvm import DataType
 
 from ..expr import Expr
@@ -27,7 +25,7 @@
 from .manipulate import permute_dims
 
 
-def matmul(x1: Expr, x2: Expr, out_dtype: Optional[Union[str, DataType]] = None) -> Expr:
+def matmul(x1: Expr, x2: Expr, out_dtype: str | DataType | None = None) -> Expr:
     """General matrix multiplication of two tensors, with broadcasting on batched dimensions.
 
     The semantics and output shape deduction rule is specified as
@@ -56,8 +54,8 @@ def matmul(x1: Expr, x2: Expr, out_dtype: Optional[Union[str, DataType]] = None)
 def linear(
     data: Expr,
     weight: Expr,
-    bias: Optional[Expr] = None,
-    out_dtype: Optional[Union[str, DataType]] = None,
+    bias: Expr | None = None,
+    out_dtype: str | DataType | None = None,
 ) -> Expr:
     """Applies a linear transformation to the incoming data: y = xA^T + b
 
@@ -108,7 +106,7 @@ def einsum(operands, subscripts):
     result : relax.Expr
         The output from the einsum op.
     """
-    if isinstance(operands, (list, tuple)):
+    if isinstance(operands, list | tuple):
         operands = RxTuple(operands)
 
     return _ffi_api.einsum(operands, subscripts)  # type: ignore
diff --git a/python/tvm/relax/op/manipulate.py b/python/tvm/relax/op/manipulate.py
index cd15d844b705..25333bf20d54 100644
--- a/python/tvm/relax/op/manipulate.py
+++ b/python/tvm/relax/op/manipulate.py
@@ -16,7 +16,7 @@
 # under the License.
 """Manipulation operators."""
 
-from typing import Callable, List, Optional, Tuple, Union
+from collections.abc import Callable
 
 from tvm.ir.expr import PrimExpr
 from tvm.tir import FloatImm, IndexMap, IntImm
@@ -25,10 +25,10 @@
 from ..expr import Tuple as RxTuple
 from . import _ffi_api
 
-PrimExprLike = Union[int, PrimExpr]
+PrimExprLike = int | PrimExpr
 
 
-def broadcast_to(x: Expr, shape: Union[Tuple[PrimExprLike], Expr]) -> Expr:
+def broadcast_to(x: Expr, shape: tuple[PrimExprLike] | Expr) -> Expr:
     """Broadcasts a tensor to a specified shape.
 
     Parameters
@@ -44,12 +44,12 @@ def broadcast_to(x: Expr, shape: Union[Tuple[PrimExprLike], Expr]) -> Expr:
     result : relax.Expr
         The broadcasted tensor.
     """
-    if isinstance(shape, (tuple, list)):
+    if isinstance(shape, tuple | list):
         shape = ShapeExpr(shape)
     return _ffi_api.broadcast_to(x, shape)  # type: ignore
 
 
-def concat(tensors: Union[Expr, List[Expr]], axis: Optional[int] = 0) -> Expr:
+def concat(tensors: Expr | list[Expr], axis: int | None = 0) -> Expr:
     """Concatenate the input tensors along the given axis.
 
     Parameters
@@ -67,12 +67,12 @@ def concat(tensors: Union[Expr, List[Expr]], axis: Optional[int] = 0) -> Expr:
     result: relax.Expr
         The concatenated tensor.
     """
-    if isinstance(tensors, (list, tuple)):
+    if isinstance(tensors, list | tuple):
         tensors = RxTuple(tensors)
     return _ffi_api.concat(tensors, axis)  # type: ignore
 
 
-def expand_dims(x: Expr, axis: Union[int, List[int]]) -> Expr:
+def expand_dims(x: Expr, axis: int | list[int]) -> Expr:
     """Insert new axes at the positions given by `axis`.
 
     Parameters
@@ -113,10 +113,10 @@ def flatten(x: Expr) -> Expr:
 
 def layout_transform(
     x: Expr,
-    index_map: Union[Callable, IndexMap],
-    pad_value: Optional[Union[int, float, PrimValue]] = None,
-    axis_separators: Optional[Union[int, IndexMap.AXIS_SEPARATOR]] = None,
-    input_axis_separators: Optional[Union[int, IndexMap.AXIS_SEPARATOR]] = None,
+    index_map: Callable | IndexMap,
+    pad_value: int | float | PrimValue | None = None,
+    axis_separators: int | str | None = None,  # str for IndexMap.AXIS_SEPARATOR
+    input_axis_separators: int | str | None = None,  # str for IndexMap.AXIS_SEPARATOR
 ):
     """Modifies the layout of a tensor.
 
@@ -125,14 +125,14 @@ def layout_transform(
     x : relax.Expr
         The input tensor to the operator.
 
-    index_map : Union[Callable, IndexMap]
+    index_map : Callable | IndexMap
         The transformation to apply.
 
-    pad_value : Optional[Union[int, float, PrimValue]]
+    pad_value : Optional[int | float | PrimValue]
         The value used for padding if the transformation results in implicit padding.
         If not specified, any value can be used.
 
-    axis_separators : Optional[Union[int, IndexMap.AXIS_SEPARATOR]]
+    axis_separators : Optional[int | IndexMap.AXIS_SEPARATOR]
         The axis_separators for index_map to create non flat buffers.
 
     Returns
@@ -153,7 +153,7 @@ def layout_transform(
     elif not isinstance(pad_value, PrimValue):
         if "int" in x_dtype and isinstance(pad_value, int):
             pad_value = IntImm(x_dtype, pad_value)
-        elif "float" in x_dtype and (isinstance(pad_value, (int, float))):
+        elif "float" in x_dtype and (isinstance(pad_value, int | float)):
             pad_value = FloatImm(x_dtype, float(pad_value))
         pad_value = PrimValue(pad_value)
 
@@ -168,7 +168,7 @@ def layout_transform(
     )
 
 
-def permute_dims(x: Expr, axes: Optional[List[int]] = None) -> Expr:
+def permute_dims(x: Expr, axes: list[int] | None = None) -> Expr:
     """Permutes the dimensions of an array.
 
     Parameters
@@ -187,7 +187,7 @@ def permute_dims(x: Expr, axes: Optional[List[int]] = None) -> Expr:
     return _ffi_api.permute_dims(x, axes)  # type: ignore
 
 
-def reshape(x: Expr, shape: Union[Tuple[PrimExprLike], Expr]) -> Expr:
+def reshape(x: Expr, shape: tuple[PrimExprLike] | Expr) -> Expr:
     """Reshape the input array.
 
     ``-1`` infers the dimension of the output shape by using the remainder of
@@ -224,7 +224,7 @@ def reshape(x: Expr, shape: Union[Tuple[PrimExprLike], Expr]) -> Expr:
 
 def split(
     x: Expr,
-    indices_or_sections: Union[int, List[PrimExprLike]],
+    indices_or_sections: int | list[PrimExprLike],
     axis: int = 0,
 ) -> Expr:
     """Split input tensor along axis by sections or indices.
@@ -257,7 +257,7 @@ def split(
     return _ffi_api.split(x, indices_or_sections, axis)  # type: ignore
 
 
-def squeeze(x: Expr, axis: Optional[Union[int, List[int]]] = None) -> Expr:
+def squeeze(x: Expr, axis: int | list[int] | None = None) -> Expr:
     """Squeeze axes in the array.
 
     Parameters
@@ -280,7 +280,7 @@ def squeeze(x: Expr, axis: Optional[Union[int, List[int]]] = None) -> Expr:
     return _ffi_api.squeeze(x, axis)  # type: ignore
 
 
-def stack(tensors: Union[Expr, List[Expr]], axis: int = 0) -> Expr:
+def stack(tensors: Expr | list[Expr], axis: int = 0) -> Expr:
     """Stack the input tensors along a new axis.
 
     Parameters
@@ -299,7 +299,7 @@ def stack(tensors: Union[Expr, List[Expr]], axis: int = 0) -> Expr:
         The stacked tensor with an additional dimension compared to the input tensors.
 
     """
-    if isinstance(tensors, (list, tuple)):
+    if isinstance(tensors, list | tuple):
         tensors = RxTuple(tensors)
     return _ffi_api.stack(tensors, axis)  # type: ignore
 
@@ -325,7 +325,7 @@ def collapse_sum_like(data: Expr, collapse_target: Expr) -> Expr:
     return _ffi_api.collapse_sum_like(data, collapse_target)  # type: ignore
 
 
-def collapse_sum_to(data: Expr, shape: Union[Tuple[PrimExprLike], Expr]) -> Expr:
+def collapse_sum_to(data: Expr, shape: tuple[PrimExprLike] | Expr) -> Expr:
     """Return a summation of data to the given shape.
 
     collapse_sum_to is intended as the backward operator of tvm.relax.op.broadcast_to and
@@ -352,12 +352,12 @@ def collapse_sum_to(data: Expr, shape: Union[Tuple[PrimExprLike], Expr]) -> Expr
     result : relax.Expr
         The result tensor of the given shape after summation.
     """
-    if isinstance(shape, (tuple, list)):
+    if isinstance(shape, tuple | list):
         shape = ShapeExpr(shape)
     return _ffi_api.collapse_sum_to(data, shape)  # type: ignore
 
 
-def repeat(data: Expr, repeats: int, axis: Optional[int] = None) -> Expr:
+def repeat(data: Expr, repeats: int, axis: int | None = None) -> Expr:
     """Repeats elements of an array.
 
     Parameters
@@ -390,7 +390,7 @@ def repeat(data: Expr, repeats: int, axis: Optional[int] = None) -> Expr:
     return _ffi_api.repeat(data, repeats, axis)  # type: ignore
 
 
-def tile(data: Expr, repeats: Union[int, Tuple[int], List[int]]) -> Expr:
+def tile(data: Expr, repeats: int | tuple[int] | list[int]) -> Expr:
     """Construct an array by repeating data the number of times given by repeats.
 
     If repeats has length l, and data has dimension d, the result will have dimension of max(l, d).
@@ -533,7 +533,7 @@ def gather_nd(data: Expr, indices: Expr, batch_dims: int = 0) -> Expr:
     return _ffi_api.gather_nd(data, indices, batch_dims)  # type: ignore
 
 
-def index_tensor(data: Expr, indices: Union[Expr, List[Expr]]) -> Expr:
+def index_tensor(data: Expr, indices: Expr | list[Expr]) -> Expr:
     """Advanced-tensor indexing (NumPy/PyTorch-style).
 
     Given k index tensors ``indices = (I0, I1, …, Ik-1)`` this
@@ -591,14 +591,14 @@ def index_tensor(data: Expr, indices: Union[Expr, List[Expr]]) -> Expr:
         # z.shape == (2,3)
 
     """
-    if isinstance(indices, (list, tuple)):
+    if isinstance(indices, list | tuple):
         indices = RxTuple(indices)
     return _ffi_api.index_tensor(data, indices)  # type: ignore
 
 
 def index_put(
     data: Expr,
-    indices: Union[Expr, Tuple[Expr]],
+    indices: Expr | tuple[Expr],
     values: Expr,
     accumulate: bool = False,
 ) -> Expr:
@@ -643,12 +643,12 @@ def index_put(
             [0.0, 3.0, 0.0],
         ]
     """
-    if isinstance(indices, (list, tuple)):
+    if isinstance(indices, list | tuple):
         indices = RxTuple(indices)
     return _ffi_api.index_put(data, indices, values, accumulate)  # type: ignore
 
 
-def meshgrid(tensors: Union[Expr, List[Expr]], indexing: Optional[str] = "ij") -> Expr:
+def meshgrid(tensors: Expr | list[Expr], indexing: str | None = "ij") -> Expr:
     """Generate coordinate grids from input tensors.
 
     Parameters
@@ -666,7 +666,7 @@ def meshgrid(tensors: Union[Expr, List[Expr]], indexing: Optional[str] = "ij") -
     result : relax.Expr
         A Tuple of tensors representing the coordinate grids.
     """
-    if isinstance(tensors, (list, tuple)):
+    if isinstance(tensors, list | tuple):
         tensors = RxTuple(tensors)
     return _ffi_api.meshgrid(tensors, indexing)
 
diff --git a/python/tvm/relax/op/memory/memory.py b/python/tvm/relax/op/memory/memory.py
index 6806c5800146..ac8343c310e0 100644
--- a/python/tvm/relax/op/memory/memory.py
+++ b/python/tvm/relax/op/memory/memory.py
@@ -15,8 +15,6 @@
 # specific language governing permissions and limitations
 """Relax memory primitives."""
 
-from typing import Union
-
 from ...expr import Call, DataTypeImm, Expr, PrimValue, StringImm
 from ...utils import args_converter
 from . import _ffi_api
@@ -25,9 +23,9 @@
 @args_converter.auto
 def alloc_storage(
     size: Expr,
-    virtual_device_index: Union[int, Expr],
-    storage_scope: Union[str, Expr],
-    dtype: Union[str, Expr],
+    virtual_device_index: int | Expr,
+    storage_scope: str | Expr,
+    dtype: str | Expr,
 ) -> Call:
     """Construct a Call to allocate a storage with specific size, virtual_device_index,
     storage_scope and dtype.
@@ -64,10 +62,10 @@ def alloc_storage(
 @args_converter.auto
 def alloc_tensor(
     storage: Expr,
-    offset: Union[int, Expr],
+    offset: int | Expr,
     shape: Expr,
-    dtype: Union[str, Expr],
-    runtime_device_ind: Union[int, Expr] = PrimValue(0),
+    dtype: str | Expr,
+    runtime_device_ind: int | Expr = PrimValue(0),
 ) -> Call:
     """Construct a Call to allocate a tensor on a certain storage starting from the given offset.
 
diff --git a/python/tvm/relax/op/memory/view.py b/python/tvm/relax/op/memory/view.py
index b1d91926735c..e4fdd71df4b7 100644
--- a/python/tvm/relax/op/memory/view.py
+++ b/python/tvm/relax/op/memory/view.py
@@ -26,21 +26,21 @@
 
 """
 
-from typing import Optional, Sequence, Union
+from collections.abc import Sequence
 
 from tvm.relax import DataTypeImm, Expr, PrimValue, ShapeExpr
 from tvm.tir import PrimExpr
 
 from . import _ffi_api
 
-PrimExprLike = Union[int, PrimExpr]
+PrimExprLike = int | PrimExpr
 
 
 def view(
     data: Expr,
-    shape: Optional[Union[Sequence[PrimExprLike], Expr]] = None,
-    dtype: Optional[Expr] = None,
-    relative_byte_offset: Optional[Expr] = None,
+    shape: Sequence[PrimExprLike] | Expr | None = None,
+    dtype: Expr | None = None,
+    relative_byte_offset: Expr | None = None,
 ) -> Expr:
     """Provide a view into an existing tensor
 
diff --git a/python/tvm/relax/op/nn/nn.py b/python/tvm/relax/op/nn/nn.py
index f4c8340460b7..e467494f2b71 100644
--- a/python/tvm/relax/op/nn/nn.py
+++ b/python/tvm/relax/op/nn/nn.py
@@ -16,8 +16,6 @@
 # under the License.
 """Relax Neural Network (NN) operators"""
 
-from typing import List, Optional, Tuple, Union
-
 from tvm import DataType, relax
 from tvm.tir import FloatImm
 
@@ -28,14 +26,14 @@
 def conv1d(
     data: Expr,
     weight: Expr,
-    strides: Union[int, Tuple[int]] = 1,
-    padding: Union[int, Tuple[int, ...]] = 0,
-    dilation: Union[int, Tuple[int]] = 1,
+    strides: int | tuple[int] = 1,
+    padding: int | tuple[int, ...] = 0,
+    dilation: int | tuple[int] = 1,
     groups: int = 1,
     data_layout: str = "NCW",
     kernel_layout: str = "OIW",
-    out_layout: Optional[str] = None,
-    out_dtype: Optional[Union[str, DataType]] = None,
+    out_layout: str | None = None,
+    out_dtype: str | DataType | None = None,
 ) -> Expr:
     r"""1D convolution.
 
@@ -126,14 +124,14 @@ def conv1d(
 def conv2d(
     data: Expr,
     weight: Expr,
-    strides: Union[int, Tuple[int, int]] = (1, 1),
-    padding: Union[int, Tuple[int, ...]] = (0, 0),
-    dilation: Union[int, Tuple[int, int]] = (1, 1),
+    strides: int | tuple[int, int] = (1, 1),
+    padding: int | tuple[int, ...] = (0, 0),
+    dilation: int | tuple[int, int] = (1, 1),
     groups: int = 1,
     data_layout: str = "NCHW",
     kernel_layout: str = "OIHW",
-    out_layout: Optional[str] = None,
-    out_dtype: Optional[Union[str, DataType]] = None,
+    out_layout: str | None = None,
+    out_dtype: str | DataType | None = None,
 ) -> Expr:
     r"""2D convolution.
 
@@ -224,14 +222,14 @@ def conv2d(
 def conv3d(
     data: Expr,
     weight: Expr,
-    strides: Union[int, Tuple[int, int]] = (1, 1, 1),
-    padding: Union[int, Tuple[int, ...]] = (0, 0, 0),
-    dilation: Union[int, Tuple[int, int]] = (1, 1, 1),
+    strides: int | tuple[int, int] = (1, 1, 1),
+    padding: int | tuple[int, ...] = (0, 0, 0),
+    dilation: int | tuple[int, int] = (1, 1, 1),
     groups: int = 1,
     data_layout: str = "NCDHW",
     kernel_layout: str = "OIDHW",
-    out_layout: Optional[str] = None,
-    out_dtype: Optional[Union[str, DataType]] = None,
+    out_layout: str | None = None,
+    out_dtype: str | DataType | None = None,
 ) -> Expr:
     r"""3D convolution.
 
@@ -324,15 +322,15 @@ def conv3d(
 def conv1d_transpose(
     data: Expr,
     weight: Expr,
-    strides: Union[int, Tuple[int]] = 1,
-    padding: Union[int, Tuple[int, ...]] = 0,
-    output_padding: Union[int, Tuple[int]] = 0,
-    dilation: Union[int, Tuple[int]] = 1,
+    strides: int | tuple[int] = 1,
+    padding: int | tuple[int, ...] = 0,
+    output_padding: int | tuple[int] = 0,
+    dilation: int | tuple[int] = 1,
     groups: int = 1,
     data_layout: str = "NCW",
     kernel_layout: str = "IOW",
-    out_layout: Optional[str] = None,
-    out_dtype: Optional[Union[str, DataType]] = None,
+    out_layout: str | None = None,
+    out_dtype: str | DataType | None = None,
 ) -> Expr:
     r"""1D transposed convolution operator.
 
@@ -415,15 +413,15 @@ def conv1d_transpose(
 def conv2d_transpose(
     data: Expr,
     weight: Expr,
-    strides: Union[int, Tuple[int, int]] = (1, 1),
-    padding: Union[int, Tuple[int, ...]] = (0, 0),
-    output_padding: Union[int, Tuple[int, int]] = (0, 0),
-    dilation: Union[int, Tuple[int, int]] = (1, 1),
+    strides: int | tuple[int, int] = (1, 1),
+    padding: int | tuple[int, ...] = (0, 0),
+    output_padding: int | tuple[int, int] = (0, 0),
+    dilation: int | tuple[int, int] = (1, 1),
     groups: int = 1,
     data_layout: str = "NCHW",
     kernel_layout: str = "IOHW",
-    out_layout: Optional[str] = None,
-    out_dtype: Optional[Union[str, DataType]] = None,
+    out_layout: str | None = None,
+    out_dtype: str | DataType | None = None,
 ) -> Expr:
     r"""Two dimensional transposed convolution operator.
 
@@ -516,9 +514,9 @@ def conv2d_transpose(
 
 def pad(
     data: Expr,
-    pad_width: Union[List[int], Tuple[int, ...]],
-    pad_mode: Optional[str] = "constant",
-    pad_value: Optional[float] = 0.0,
+    pad_width: list[int] | tuple[int, ...],
+    pad_mode: str | None = "constant",
+    pad_value: float | None = 0.0,
 ):
     r"""Padding
 
@@ -585,14 +583,14 @@ def pixel_shuffle(data: Expr, upscale_factor: int):
 
 def max_pool1d(
     data: Expr,
-    pool_size: Union[int, Tuple[int, int]] = (1,),
-    strides: Union[int, Tuple[int, int]] = (1,),
-    padding: Union[int, Tuple[int, ...]] = (0, 0),
-    dilation: Union[int, Tuple[int, int]] = (1,),
+    pool_size: int | tuple[int, int] = (1,),
+    strides: int | tuple[int, int] = (1,),
+    padding: int | tuple[int, ...] = (0, 0),
+    dilation: int | tuple[int, int] = (1,),
     ceil_mode: bool = False,
     count_include_pad: bool = False,
     layout: str = "NCW",
-    out_layout: Optional[str] = None,
+    out_layout: str | None = None,
 ) -> Expr:
     r"""1D maximum pooling operator.
 
@@ -666,14 +664,14 @@ def max_pool1d(
 
 def max_pool2d(
     data: Expr,
-    pool_size: Union[int, Tuple[int, int]] = (1, 1),
-    strides: Union[int, Tuple[int, int]] = (1, 1),
-    padding: Union[int, Tuple[int, ...]] = (0, 0),
-    dilation: Union[int, Tuple[int, int]] = (1, 1),
+    pool_size: int | tuple[int, int] = (1, 1),
+    strides: int | tuple[int, int] = (1, 1),
+    padding: int | tuple[int, ...] = (0, 0),
+    dilation: int | tuple[int, int] = (1, 1),
     ceil_mode: bool = False,
     count_include_pad: bool = False,
     layout: str = "NCHW",
-    out_layout: Optional[str] = None,
+    out_layout: str | None = None,
 ) -> Expr:
     r"""2D maximum pooling operator.
 
@@ -754,14 +752,14 @@ def max_pool2d(
 
 def max_pool3d(
     data: Expr,
-    pool_size: Union[int, Tuple[int, int]] = (1, 1, 1),
-    strides: Union[int, Tuple[int, int]] = (1, 1, 1),
-    padding: Union[int, Tuple[int, ...]] = (0, 0, 0),
-    dilation: Union[int, Tuple[int, int]] = (1, 1, 1),
+    pool_size: int | tuple[int, int] = (1, 1, 1),
+    strides: int | tuple[int, int] = (1, 1, 1),
+    padding: int | tuple[int, ...] = (0, 0, 0),
+    dilation: int | tuple[int, int] = (1, 1, 1),
     ceil_mode: bool = False,
     count_include_pad: bool = False,
     layout: str = "NCDHW",
-    out_layout: Optional[str] = None,
+    out_layout: str | None = None,
 ) -> Expr:
     r"""3D maximum pooling operator.
 
@@ -836,14 +834,14 @@ def max_pool3d(
 
 def avg_pool1d(
     data: Expr,
-    pool_size: Union[int, Tuple[int, int]] = (1,),
-    strides: Union[int, Tuple[int, int]] = (1,),
-    padding: Union[int, Tuple[int, ...]] = (0, 0),
-    dilation: Union[int, Tuple[int, int]] = (1,),
+    pool_size: int | tuple[int, int] = (1,),
+    strides: int | tuple[int, int] = (1,),
+    padding: int | tuple[int, ...] = (0, 0),
+    dilation: int | tuple[int, int] = (1,),
     ceil_mode: bool = False,
     count_include_pad: bool = True,
     layout: str = "NCW",
-    out_layout: Optional[str] = None,
+    out_layout: str | None = None,
 ) -> Expr:
     r"""1D average pooling operator.
 
@@ -916,14 +914,14 @@ def avg_pool1d(
 
 def avg_pool2d(
     data: Expr,
-    pool_size: Union[int, Tuple[int, int]] = (1, 1),
-    strides: Union[int, Tuple[int, int]] = (1, 1),
-    padding: Union[int, Tuple[int, ...]] = (0, 0),
-    dilation: Union[int, Tuple[int, int]] = (1, 1),
+    pool_size: int | tuple[int, int] = (1, 1),
+    strides: int | tuple[int, int] = (1, 1),
+    padding: int | tuple[int, ...] = (0, 0),
+    dilation: int | tuple[int, int] = (1, 1),
     ceil_mode: bool = False,
     count_include_pad: bool = False,
     layout: str = "NCHW",
-    out_layout: Optional[str] = None,
+    out_layout: str | None = None,
 ) -> Expr:
     r"""2D average pooling operator.
 
@@ -1004,14 +1002,14 @@ def avg_pool2d(
 
 def avg_pool3d(
     data: Expr,
-    pool_size: Union[int, Tuple[int, int]] = (1, 1, 1),
-    strides: Union[int, Tuple[int, int]] = (1, 1, 1),
-    padding: Union[int, Tuple[int, ...]] = (0, 0, 0),
-    dilation: Union[int, Tuple[int, int]] = (1, 1, 1),
+    pool_size: int | tuple[int, int] = (1, 1, 1),
+    strides: int | tuple[int, int] = (1, 1, 1),
+    padding: int | tuple[int, ...] = (0, 0, 0),
+    dilation: int | tuple[int, int] = (1, 1, 1),
     ceil_mode: bool = False,
     count_include_pad: bool = True,
     layout: str = "NCDHW",
-    out_layout: Optional[str] = None,
+    out_layout: str | None = None,
 ) -> Expr:
     r"""2D average pooling operator.
 
@@ -1086,9 +1084,9 @@ def avg_pool3d(
 
 def adaptive_avg_pool1d(
     data: Expr,
-    output_size: Optional[Union[int, Tuple[int]]] = None,
+    output_size: int | tuple[int] | None = None,
     layout: str = "NCW",
-    out_layout: Optional[str] = None,
+    out_layout: str | None = None,
 ) -> Expr:
     r"""1D adaptive average pooling operator. This operator is experimental.
 
@@ -1139,9 +1137,9 @@ def adaptive_avg_pool1d(
 
 def adaptive_avg_pool2d(
     data: Expr,
-    output_size: Optional[Union[int, Tuple[int, int]]] = None,
+    output_size: int | tuple[int, int] | None = None,
     layout: str = "NCHW",
-    out_layout: Optional[str] = None,
+    out_layout: str | None = None,
 ) -> Expr:
     r"""2D adaptive average pooling operator. This operator is experimental.
 
@@ -1195,9 +1193,9 @@ def adaptive_avg_pool2d(
 
 def adaptive_avg_pool3d(
     data: Expr,
-    output_size: Optional[Union[int, Tuple[int, int]]] = None,
+    output_size: int | tuple[int, int] | None = None,
     layout: str = "NCDHW",
-    out_layout: Optional[str] = None,
+    out_layout: str | None = None,
 ) -> Expr:
     r"""3D adaptive average pooling operator. This operator is experimental.
 
@@ -1640,7 +1638,7 @@ def layer_norm(
     data: Expr,
     gamma: Expr,
     beta: Expr,
-    axes: Union[int, List[int]],
+    axes: int | list[int],
     epsilon: float = 1e-5,
     center: bool = True,
     scale: bool = True,
@@ -1703,7 +1701,7 @@ def group_norm(
     beta: Expr,
     num_groups: int,
     channel_axis: int,
-    axes: Union[int, List[int]],
+    axes: int | list[int],
     epsilon: float = 1e-5,
     center: bool = True,
     scale: bool = True,
@@ -1760,7 +1758,7 @@ def instance_norm(
     gamma: Expr,
     beta: Expr,
     channel_axis: int,
-    axes: List[int],
+    axes: list[int],
     epsilon: float = 1e-5,
     center: bool = True,
     scale: bool = True,
@@ -1813,7 +1811,7 @@ def instance_norm(
 def rms_norm(
     data: Expr,
     weight: Expr,
-    axes: Union[int, List[int]] = -1,
+    axes: int | list[int] = -1,
     epsilon: float = 1e-5,
 ) -> Expr:
     r"""
@@ -1905,7 +1903,7 @@ def cross_entropy_with_logits(predictions: Expr, labels: Expr) -> Expr:
 def nll_loss(
     predictions: Expr,
     targets: Expr,
-    weights: Optional[Expr] = None,
+    weights: Expr | None = None,
     reduction: str = "mean",
     ignore_index: int = -100,
 ) -> Expr:
@@ -1951,10 +1949,10 @@ def attention(
     query: Expr,
     key: Expr,
     value: Expr,
-    bias: Optional[Expr] = None,
-    scale: Optional[FloatImm] = None,
-    causal_mask: Optional[str] = None,
-    window_size: Optional[int] = None,
+    bias: Expr | None = None,
+    scale: FloatImm | None = None,
+    causal_mask: str | None = None,
+    window_size: int | None = None,
 ) -> Expr:
     r"""Computes fused multi head attention.
 
@@ -2042,10 +2040,10 @@ def attention_bias(
     query: Expr,
     key: Expr,
     value: Expr,
-    bias: Optional[Expr] = None,
-    scale: Optional[FloatImm] = None,
-    causal_mask: Optional[str] = None,
-    window_size: Optional[int] = None,
+    bias: Expr | None = None,
+    scale: FloatImm | None = None,
+    causal_mask: str | None = None,
+    window_size: int | None = None,
 ) -> Expr:
     r"""Computes fused multi head attention.
 
@@ -2139,11 +2137,11 @@ def attention_var_len(
     values: Expr,
     seqstart_q: Expr,
     max_seqlen_q: Expr,
-    seqstart_k: Optional[Expr] = None,
-    max_seqlen_k: Optional[Expr] = None,
-    scale: Optional[FloatImm] = None,
-    causal_mask: Optional[str] = None,
-    window_size: Optional[int] = None,
+    seqstart_k: Expr | None = None,
+    max_seqlen_k: Expr | None = None,
+    scale: FloatImm | None = None,
+    causal_mask: str | None = None,
+    window_size: int | None = None,
 ) -> Expr:
     """Computes fused multi head attention over batched sequences of variable lengths.
 
diff --git a/python/tvm/relax/op/search.py b/python/tvm/relax/op/search.py
index 95cfe60da9f5..8428400a4537 100644
--- a/python/tvm/relax/op/search.py
+++ b/python/tvm/relax/op/search.py
@@ -17,8 +17,6 @@
 # pylint: disable=invalid-name
 """Search operators."""
 
-from typing import Optional
-
 from ..expr import Expr
 from . import _ffi_api
 
@@ -53,7 +51,7 @@ def where(condition: Expr, x1: Expr, x2: Expr) -> Expr:
     return _ffi_api.where(condition, x1, x2)  # type: ignore
 
 
-def argmax(x: Expr, axis: Optional[int] = None, keepdims: bool = False) -> Expr:
+def argmax(x: Expr, axis: int | None = None, keepdims: bool = False) -> Expr:
     """Computes the argmax of tensor elements over given axis.
 
     Parameters
@@ -79,7 +77,7 @@ def argmax(x: Expr, axis: Optional[int] = None, keepdims: bool = False) -> Expr:
     return _ffi_api.argmax(x, axis, keepdims)  # type: ignore
 
 
-def argmin(x: Expr, axis: Optional[int] = None, keepdims: bool = False) -> Expr:
+def argmin(x: Expr, axis: int | None = None, keepdims: bool = False) -> Expr:
     """Computes the argmin of tensor elements over given axis.
 
     Parameters
diff --git a/python/tvm/relax/op/set.py b/python/tvm/relax/op/set.py
index dd72fae26628..010b3d41adbf 100644
--- a/python/tvm/relax/op/set.py
+++ b/python/tvm/relax/op/set.py
@@ -17,8 +17,6 @@
 # pylint: disable=import-outside-toplevel, redefined-builtin, unused-argument
 """Set operators."""
 
-from typing import Optional, Union
-
 import numpy as np  # type: ignore
 
 import tvm
@@ -29,11 +27,11 @@
 
 def unique(
     x: Expr,
-    sorted: Union[bool, Expr] = True,
-    return_index: Union[bool, Expr] = False,
-    return_inverse: Union[bool, Expr] = False,
-    return_counts: Union[bool, Expr] = False,
-    axis: Optional[Union[int, Expr]] = None,
+    sorted: bool | Expr = True,
+    return_index: bool | Expr = False,
+    return_inverse: bool | Expr = False,
+    return_counts: bool | Expr = False,
+    axis: int | Expr | None = None,
 ) -> Expr:
     """Find the unique elements in a given tensor.
     In addition, it optionally returns
@@ -93,7 +91,7 @@ def numpy_unique(
     return_index: int,
     return_inverse: int,
     return_counts: int,
-    axis: Optional[int] = None,
+    axis: int | None = None,
 ) -> tvm.runtime.tensor:
     """Returns the unique elements of the input tensor.
 
diff --git a/python/tvm/relax/op/statistical.py b/python/tvm/relax/op/statistical.py
index 210184fe5652..ecbfe385eb71 100644
--- a/python/tvm/relax/op/statistical.py
+++ b/python/tvm/relax/op/statistical.py
@@ -17,15 +17,13 @@
 # pylint: disable=redefined-builtin
 """Statistical operators."""
 
-from typing import List, Optional, Union
-
 from tvm import DataType
 
 from ..expr import Expr
 from . import _ffi_api
 
 
-def max(x: Expr, axis: Optional[Union[int, List[int]]] = None, keepdims: bool = False) -> Expr:
+def max(x: Expr, axis: int | list[int] | None = None, keepdims: bool = False) -> Expr:
     """Computes the max of tensor elements over given axes.
 
     Parameters
@@ -53,7 +51,7 @@ def max(x: Expr, axis: Optional[Union[int, List[int]]] = None, keepdims: bool =
     return _ffi_api.max(x, axis, keepdims)  # type: ignore
 
 
-def mean(x: Expr, axis: Optional[Union[int, List[int]]] = None, keepdims: bool = False) -> Expr:
+def mean(x: Expr, axis: int | list[int] | None = None, keepdims: bool = False) -> Expr:
     """Computes the mean of tensor elements over given axes.
 
     Parameters
@@ -81,7 +79,7 @@ def mean(x: Expr, axis: Optional[Union[int, List[int]]] = None, keepdims: bool =
     return _ffi_api.mean(x, axis, keepdims)  # type: ignore
 
 
-def min(x: Expr, axis: Optional[Union[int, List[int]]] = None, keepdims: bool = False) -> Expr:
+def min(x: Expr, axis: int | list[int] | None = None, keepdims: bool = False) -> Expr:
     """Computes the min of tensor elements over given axes.
 
     Parameters
@@ -109,7 +107,7 @@ def min(x: Expr, axis: Optional[Union[int, List[int]]] = None, keepdims: bool =
     return _ffi_api.min(x, axis, keepdims)  # type: ignore
 
 
-def prod(x: Expr, axis: Optional[Union[int, List[int]]] = None, keepdims: bool = False) -> Expr:
+def prod(x: Expr, axis: int | list[int] | None = None, keepdims: bool = False) -> Expr:
     """Computes the product of tensor elements over given axes.
 
     Parameters
@@ -137,7 +135,7 @@ def prod(x: Expr, axis: Optional[Union[int, List[int]]] = None, keepdims: bool =
     return _ffi_api.prod(x, axis, keepdims)  # type: ignore
 
 
-def std(x: Expr, axis: Optional[Union[int, List[int]]] = None, keepdims: bool = False) -> Expr:
+def std(x: Expr, axis: int | list[int] | None = None, keepdims: bool = False) -> Expr:
     """Computes the standard deviation of tensor elements over given axes.
 
     Parameters
@@ -165,7 +163,7 @@ def std(x: Expr, axis: Optional[Union[int, List[int]]] = None, keepdims: bool =
     return _ffi_api.std(x, axis, keepdims)  # type: ignore
 
 
-def sum(x: Expr, axis: Optional[Union[int, List[int]]] = None, keepdims: bool = False) -> Expr:
+def sum(x: Expr, axis: int | list[int] | None = None, keepdims: bool = False) -> Expr:
     """Computes the sum of tensor elements over given axes.
 
     Parameters
@@ -195,8 +193,8 @@ def sum(x: Expr, axis: Optional[Union[int, List[int]]] = None, keepdims: bool =
 
 def cumprod(
     data: Expr,
-    axis: Optional[int] = None,
-    dtype: Optional[Union[str, DataType]] = None,
+    axis: int | None = None,
+    dtype: str | DataType | None = None,
     exclusive: bool = False,
 ):
     """Numpy style cumprod op. Return the cumulative product of the elements along
@@ -257,8 +255,8 @@ def cumprod(
 
 def cumsum(
     data: Expr,
-    axis: Optional[int] = None,
-    dtype: Optional[Union[str, DataType]] = None,
+    axis: int | None = None,
+    dtype: str | DataType | None = None,
     exclusive: bool = False,
 ):
     """Numpy style cumsum op. Return the cumulative inclusive sum of the elements along
@@ -317,7 +315,7 @@ def cumsum(
     return _ffi_api.cumsum(data, axis, dtype, exclusive)  # type: ignore
 
 
-def variance(x: Expr, axis: Optional[Union[int, List[int]]] = None, keepdims: bool = False) -> Expr:
+def variance(x: Expr, axis: int | list[int] | None = None, keepdims: bool = False) -> Expr:
     """Computes the variance of tensor elements over given axes.
 
     Parameters
@@ -345,7 +343,7 @@ def variance(x: Expr, axis: Optional[Union[int, List[int]]] = None, keepdims: bo
     return _ffi_api.variance(x, axis, keepdims)  # type: ignore
 
 
-def median(x: Expr, axis: Optional[Union[int, List[int]]] = None, keepdims: bool = False) -> Expr:
+def median(x: Expr, axis: int | list[int] | None = None, keepdims: bool = False) -> Expr:
     """Computes the median of tensor elements over given axes.
 
     Parameters
diff --git a/python/tvm/relax/op/vm/vm.py b/python/tvm/relax/op/vm/vm.py
index 6354359e1106..de1015d08d8b 100644
--- a/python/tvm/relax/op/vm/vm.py
+++ b/python/tvm/relax/op/vm/vm.py
@@ -15,8 +15,6 @@
 # specific language governing permissions and limitations
 """Relax vm primitives."""
 
-from typing import Union
-
 from ...expr import Call, DataTypeImm, Expr, PrimValue, StringImm, Tuple
 from ...utils import args_converter
 from . import _ffi_api
@@ -25,9 +23,9 @@
 @args_converter.auto
 def alloc_storage(
     shape: Expr,
-    runtime_device_index: Union[int, Expr],
-    dtype: Union[str, Expr],
-    storage_scope: Union[str, StringImm] = "global",
+    runtime_device_index: int | Expr,
+    dtype: str | Expr,
+    storage_scope: str | StringImm = "global",
 ) -> Call:
     """Construct a Call to allocate a storage with specific size,
     runtime_device_index, and dtype.
@@ -64,10 +62,10 @@ def alloc_storage(
 @args_converter.auto
 def alloc_tensor(
     storage: Expr,
-    offset: Union[int, Expr],
+    offset: int | Expr,
     shape: Expr,
-    dtype: Union[str, Expr],
-    runtime_device_ind: Union[int, Expr] = PrimValue(0),
+    dtype: str | Expr,
+    runtime_device_ind: int | Expr = PrimValue(0),
 ) -> Call:
     """Construct a Call to allocate a tensor on a certain storage starting from the given offset.
 
@@ -136,7 +134,7 @@ def call_tir_dyn(func: Expr, args: Tuple) -> Call:
     result : Call
         A relax Call to call_tir_dyn.
     """
-    if isinstance(args, (list, tuple)):
+    if isinstance(args, list | tuple):
         args = Tuple(args)
 
     return _ffi_api.call_tir_dyn(func, args)  # type: ignore
diff --git a/python/tvm/relax/pipeline.py b/python/tvm/relax/pipeline.py
index b94227fa59a4..49675cc43056 100644
--- a/python/tvm/relax/pipeline.py
+++ b/python/tvm/relax/pipeline.py
@@ -22,7 +22,6 @@
 """
 
 # pylint: disable=unused-argument
-from typing import Optional, Union
 
 import tvm
 from tvm.s_tir import meta_schedule as ms
@@ -110,10 +109,10 @@ def _pipeline(mod: tvm.ir.IRModule, _ctx: tvm.transform.PassContext) -> tvm.ir.I
 
 def static_shape_tuning_pipeline(
     total_trials: int,
-    target: Union[str, tvm.target.Target],
+    target: str | tvm.target.Target,
     work_dir: str = "tuning_logs",
     cpu_weight_prepack: bool = False,
-    max_trials_per_task: Optional[int] = None,
+    max_trials_per_task: int | None = None,
 ):
     """Tune the static shape model and store the log to database.
 
diff --git a/python/tvm/relax/relax_to_pyfunc_converter.py b/python/tvm/relax/relax_to_pyfunc_converter.py
index 2905d6fb4be0..bd15a85610f0 100644
--- a/python/tvm/relax/relax_to_pyfunc_converter.py
+++ b/python/tvm/relax/relax_to_pyfunc_converter.py
@@ -22,7 +22,7 @@
 """
 
 import traceback
-from typing import Any, Dict, List, Optional, Union
+from typing import Any, Optional, Union
 
 import numpy  # pylint: disable=unused-import
 import torch
@@ -55,7 +55,7 @@ def __init__(self, ir_module: IRModule):
         self._op_cache = {}
 
     def _create_fallback_tensor(
-        self, shape_hint: Optional[List[int]] = None, dtype: str = "float32"
+        self, shape_hint: list[int] | None = None, dtype: str = "float32"
     ) -> torch.Tensor:
         """Create a fallback tensor with reasonable default shape."""
         if shape_hint:
@@ -65,7 +65,7 @@ def _create_fallback_tensor(
             # Use a small default shape
             return torch.zeros(1, dtype=getattr(torch, dtype))
 
-    def convert(self, relax_function_names: Union[str, List[str]]) -> IRModule:
+    def convert(self, relax_function_names: str | list[str]) -> IRModule:
         """Convert specified Relax functions to Python functions.
 
         Args:
@@ -150,7 +150,7 @@ def converted_function(*args, **_kwargs):
         return converted_function
 
     @staticmethod
-    def _get_op_map() -> Dict[str, str]:
+    def _get_op_map() -> dict[str, str]:
         """Get the mapping from Relax operators to PyTorch operators."""
         return {
             # Binary operations
@@ -362,9 +362,9 @@ class RelaxExpressionConverter:
 
     def __init__(
         self,
-        operator_map: Dict[str, str],
+        operator_map: dict[str, str],
         ir_module: IRModule = None,
-        op_cache: Optional[Dict[str, str]] = None,
+        op_cache: dict[str, str] | None = None,
     ):
         """Initialize the expression converter.
 
@@ -374,14 +374,14 @@ def __init__(
             op_cache: Shared cache for operator mappings to avoid repeated lookups
         """
         self.operator_map = operator_map
-        self.variable_map: Dict[str, Any] = {}
-        self.current_params: List[relax.Var] = []
+        self.variable_map: dict[str, Any] = {}
+        self.current_params: list[relax.Var] = []
         self.ir_module = ir_module
         # Use shared operator cache or create new one
         self._op_cache = op_cache if op_cache is not None else {}
 
     def _create_fallback_tensor(
-        self, shape_hint: Optional[List[int]] = None, dtype: str = "float32"
+        self, shape_hint: list[int] | None = None, dtype: str = "float32"
     ) -> torch.Tensor:
         """Create a fallback tensor with reasonable default shape."""
         if shape_hint:
@@ -389,7 +389,7 @@ def _create_fallback_tensor(
         else:
             return torch.zeros(1, dtype=getattr(torch, dtype))
 
-    def convert_expr(self, expr: relax.Expr, args: List[Any]) -> Any:
+    def convert_expr(self, expr: relax.Expr, args: list[Any]) -> Any:
         """Convert a Relax expression to Python/PyTorch equivalent."""
         if isinstance(expr, relax.Var):
             return self._convert_var(expr, args)
@@ -411,7 +411,7 @@ def convert_expr(self, expr: relax.Expr, args: List[Any]) -> Any:
             # Fallback for unknown expression types
             return f"<unknown_expr: {type(expr).__name__}>"
 
-    def _convert_var(self, var: relax.Var, args: List[Any]) -> Any:
+    def _convert_var(self, var: relax.Var, args: list[Any]) -> Any:
         """Convert a Relax variable to Python equivalent."""
         if hasattr(var, "name_hint"):
             var_name = var.name_hint
@@ -445,7 +445,7 @@ def _convert_var(self, var: relax.Var, args: List[Any]) -> Any:
             return self._create_fallback_tensor()
         return self._create_fallback_tensor()
 
-    def _convert_call(self, call: relax.Call, args: List[Any]) -> Any:
+    def _convert_call(self, call: relax.Call, args: list[Any]) -> Any:
         """Convert a Relax call to Python/PyTorch equivalent."""
         op = call.op
 
@@ -462,7 +462,7 @@ def _convert_call(self, call: relax.Call, args: List[Any]) -> Any:
         else:
             return self._create_fallback_tensor()
 
-    def _convert_function_call(self, call: relax.Call, args: List[Any]) -> Any:
+    def _convert_function_call(self, call: relax.Call, args: list[Any]) -> Any:
         """Convert a Relax function call."""
         func_name = call.op.name_hint
         call_args = [self.convert_expr(arg, args) for arg in call.args]
@@ -476,7 +476,7 @@ def _convert_function_call(self, call: relax.Call, args: List[Any]) -> Any:
             # Regular function call - return first argument as fallback
             return call_args[0] if call_args else self._create_fallback_tensor()
 
-    def _convert_operator_call(self, call: relax.Call, args: List[Any]) -> Any:
+    def _convert_operator_call(self, call: relax.Call, args: list[Any]) -> Any:
         """Convert a Relax operator call to PyTorch equivalent."""
         op_name = call.op.name
         call_args = [self.convert_expr(arg, args) for arg in call.args]
@@ -556,7 +556,7 @@ def _convert_operator_call(self, call: relax.Call, args: List[Any]) -> Any:
             return f"<unknown_op: {op_name}({', '.join(map(str, call_args))})>"
 
     def _handle_functional_operation(
-        self, pytorch_op: str, call: relax.Call, call_args: List[Any]
+        self, pytorch_op: str, call: relax.Call, call_args: list[Any]
     ) -> Any:
         """Handle PyTorch functional operations with special parameter handling."""
         # Neural network function
@@ -571,7 +571,7 @@ def _handle_functional_operation(
                 axis = call.attrs.axis
                 if hasattr(axis, "value"):
                     axis = int(axis.value)
-                elif isinstance(axis, (int, float)):
+                elif isinstance(axis, int | float):
                     axis = int(axis)
 
             if axis is not None:
@@ -582,7 +582,7 @@ def _handle_functional_operation(
         else:
             return func(*call_args)
 
-    def _convert_extern_func_call(self, call: relax.Call, args: List[Any]) -> Any:
+    def _convert_extern_func_call(self, call: relax.Call, args: list[Any]) -> Any:
         """Convert an external function call."""
         func_name = call.op.global_symbol
         call_args = [self.convert_expr(arg, args) for arg in call.args]
@@ -594,7 +594,7 @@ def _convert_extern_func_call(self, call: relax.Call, args: List[Any]) -> Any:
         else:
             return call_args[0] if call_args else self._create_fallback_tensor()
 
-    def _convert_call_tir(self, call: relax.Call, args: List[Any]) -> Any:
+    def _convert_call_tir(self, call: relax.Call, args: list[Any]) -> Any:
         """Convert call_tir to Python equivalent with DLPack conversion."""
         # Extract TIR function name and arguments
         tir_func = call.args[0]
@@ -708,7 +708,7 @@ def _convert_call_tir(self, call: relax.Call, args: List[Any]) -> Any:
             else:
                 return converted_args[0] if converted_args else torch.tensor([])
 
-    def _convert_call_dps_packed(self, call: relax.Call, args: List[Any]) -> Any:
+    def _convert_call_dps_packed(self, call: relax.Call, args: list[Any]) -> Any:
         """Convert call_dps_packed to Python equivalent with DLPack conversion."""
         # Extract packed function name and arguments
         packed_func = call.args[0]
@@ -799,7 +799,7 @@ def _convert_constant(self, const: relax.Constant) -> Any:
                 return data
         return self._create_fallback_tensor()
 
-    def _convert_seq_expr(self, seq: relax.SeqExpr, args: List[Any]) -> Any:
+    def _convert_seq_expr(self, seq: relax.SeqExpr, args: list[Any]) -> Any:
         """Convert a Relax sequence expression."""
         # Convert blocks
         for block in seq.blocks:
@@ -813,12 +813,12 @@ def _convert_seq_expr(self, seq: relax.SeqExpr, args: List[Any]) -> Any:
         # Convert body
         return self.convert_expr(seq.body, args)
 
-    def _convert_tuple(self, tuple_expr: relax.Tuple, args: List[Any]) -> Any:
+    def _convert_tuple(self, tuple_expr: relax.Tuple, args: list[Any]) -> Any:
         """Convert a Relax tuple to Python tuple."""
         elements = [self.convert_expr(elem, args) for elem in tuple_expr.fields]
         return tuple(elements)
 
-    def _convert_tuple_get_item(self, get_item: relax.TupleGetItem, args: List[Any]) -> Any:
+    def _convert_tuple_get_item(self, get_item: relax.TupleGetItem, args: list[Any]) -> Any:
         """Convert a Relax tuple get item to Python equivalent."""
         tuple_expr = self.convert_expr(get_item.tuple_value, args)
         index = get_item.index
@@ -827,7 +827,7 @@ def _convert_tuple_get_item(self, get_item: relax.TupleGetItem, args: List[Any])
         else:
             return self._create_fallback_tensor()
 
-    def _convert_if(self, if_expr: relax.If, args: List[Any]) -> Any:
+    def _convert_if(self, if_expr: relax.If, args: list[Any]) -> Any:
         """Convert a Relax if expression to Python equivalent."""
         condition = self.convert_expr(if_expr.cond, args)
         true_branch = self.convert_expr(if_expr.true_branch, args)
@@ -845,7 +845,7 @@ def _convert_if(self, if_expr: relax.If, args: List[Any]) -> Any:
                 else self._create_fallback_tensor()
             )
 
-    def _convert_expand_dims(self, call: relax.Call, args: List[Any]) -> Any:
+    def _convert_expand_dims(self, call: relax.Call, args: list[Any]) -> Any:
         """Convert expand_dims to torch.unsqueeze with proper axis handling."""
         if len(call.args) < 1:
             return self._create_fallback_tensor()
@@ -866,7 +866,7 @@ def _convert_expand_dims(self, call: relax.Call, args: List[Any]) -> Any:
             if hasattr(axis, "value"):
                 # It's a TVM IntImm or similar, get the value
                 axis = int(axis.value)
-            elif isinstance(axis, (int, float)):
+            elif isinstance(axis, int | float):
                 axis = int(axis)
 
         if axis is None:
@@ -875,7 +875,7 @@ def _convert_expand_dims(self, call: relax.Call, args: List[Any]) -> Any:
         # Use torch.unsqueeze with the correct axis
         return torch.unsqueeze(tensor_arg, dim=axis)
 
-    def _convert_reduction_op(self, call: relax.Call, args: List[Any], op_name: str) -> Any:
+    def _convert_reduction_op(self, call: relax.Call, args: list[Any], op_name: str) -> Any:
         """Convert reduction operations with axis and keepdims parameters."""
         if len(call.args) < 1:
             return f"<{op_name}_error: insufficient arguments>"
@@ -899,7 +899,7 @@ def _convert_reduction_op(self, call: relax.Call, args: List[Any], op_name: str)
                 elif hasattr(axis, "value"):
                     # It's a TVM IntImm, get the value
                     axis = int(axis.value)
-                elif isinstance(axis, (int, float)):
+                elif isinstance(axis, int | float):
                     axis = int(axis)
 
             if hasattr(call.attrs, "keepdims"):
@@ -926,7 +926,7 @@ def _convert_reduction_op(self, call: relax.Call, args: List[Any], op_name: str)
         else:
             return func(tensor_arg)
 
-    def _convert_squeeze(self, call: relax.Call, args: List[Any]) -> Any:
+    def _convert_squeeze(self, call: relax.Call, args: list[Any]) -> Any:
         """Convert squeeze to torch.squeeze with proper axis handling."""
         if len(call.args) < 1:
             return "<squeeze_error: insufficient arguments>"
@@ -943,7 +943,7 @@ def _convert_squeeze(self, call: relax.Call, args: List[Any]) -> Any:
                 axis = [int(item.value) if hasattr(item, "value") else int(item) for item in axis]
             elif hasattr(axis, "value"):
                 axis = int(axis.value)
-            elif isinstance(axis, (int, float)):
+            elif isinstance(axis, int | float):
                 axis = int(axis)
 
         # Call torch.squeeze with appropriate parameters
@@ -952,7 +952,7 @@ def _convert_squeeze(self, call: relax.Call, args: List[Any]) -> Any:
         else:
             return torch.squeeze(tensor_arg)
 
-    def _convert_tensor_ops(self, call: relax.Call, args: List[Any], op_name: str) -> Any:
+    def _convert_tensor_ops(self, call: relax.Call, args: list[Any], op_name: str) -> Any:
         """Convert tensor operations like concat, split, stack."""
         if len(call.args) < 1:
             return f"<{op_name}_error: insufficient arguments>"
@@ -974,7 +974,7 @@ def _convert_tensor_ops(self, call: relax.Call, args: List[Any], op_name: str) -
                 axis = call.attrs.axis
                 if hasattr(axis, "value"):
                     axis = int(axis.value)
-                elif isinstance(axis, (int, float)):
+                elif isinstance(axis, int | float):
                     axis = int(axis)
             return torch.cat(tensors, dim=axis)
 
@@ -987,7 +987,7 @@ def _convert_tensor_ops(self, call: relax.Call, args: List[Any], op_name: str) -
                 axis = call.attrs.axis
                 if hasattr(axis, "value"):
                     axis = int(axis.value)
-                elif isinstance(axis, (int, float)):
+                elif isinstance(axis, int | float):
                     axis = int(axis)
 
             # Handle indices_or_sections parameter
@@ -995,7 +995,7 @@ def _convert_tensor_ops(self, call: relax.Call, args: List[Any], op_name: str) -
                 indices_or_sections = call.attrs.indices_or_sections
                 if hasattr(indices_or_sections, "value"):
                     indices_or_sections = int(indices_or_sections.value)
-                elif isinstance(indices_or_sections, (int, float)):
+                elif isinstance(indices_or_sections, int | float):
                     indices_or_sections = int(indices_or_sections)
 
                 # If indices_or_sections is an integer, it means split into N equal parts
@@ -1022,14 +1022,14 @@ def _convert_tensor_ops(self, call: relax.Call, args: List[Any], op_name: str) -
                 axis = call.attrs.axis
                 if hasattr(axis, "value"):
                     axis = int(axis.value)
-                elif isinstance(axis, (int, float)):
+                elif isinstance(axis, int | float):
                     axis = int(axis)
             return torch.stack(tensors, dim=axis)
 
         else:
             return f"<{op_name}_error: unsupported operation>"
 
-    def _convert_reshape(self, call: relax.Call, args: List[Any]) -> Any:
+    def _convert_reshape(self, call: relax.Call, args: list[Any]) -> Any:
         """Convert reshape operation."""
         if len(call.args) < 2:
             return "<reshape_error: insufficient arguments>"
@@ -1052,14 +1052,14 @@ def _convert_reshape(self, call: relax.Call, args: List[Any]) -> Any:
         else:
             # Try to convert as expression
             converted_shape = self.convert_expr(shape_arg, args)
-            if isinstance(converted_shape, (list, tuple)):
+            if isinstance(converted_shape, list | tuple):
                 shape = tuple(int(v) for v in converted_shape)
             else:
                 shape = (int(converted_shape),)
 
         return torch.reshape(tensor_arg, shape)
 
-    def _convert_permute_dims(self, call: relax.Call, args: List[Any]) -> Any:
+    def _convert_permute_dims(self, call: relax.Call, args: list[Any]) -> Any:
         """Convert permute_dims operation."""
         if len(call.args) < 1:
             return "<permute_dims_error: insufficient arguments>"
@@ -1073,7 +1073,7 @@ def _convert_permute_dims(self, call: relax.Call, args: List[Any]) -> Any:
             if hasattr(axes, "__iter__") and not isinstance(axes, str):
                 # Convert TVM Array or Python list/tuple to tuple of ints
                 axes = tuple(int(v.value) if hasattr(v, "value") else int(v) for v in axes)
-            elif isinstance(axes, (list, tuple)):
+            elif isinstance(axes, list | tuple):
                 axes = tuple(int(v) for v in axes)
             else:
                 axes = (int(axes),)
@@ -1082,7 +1082,7 @@ def _convert_permute_dims(self, call: relax.Call, args: List[Any]) -> Any:
 
         return torch.permute(tensor_arg, axes)
 
-    def _convert_take(self, call: relax.Call, args: List[Any]) -> Any:
+    def _convert_take(self, call: relax.Call, args: list[Any]) -> Any:
         """Convert take operation."""
         if len(call.args) < 2:
             return "<take_error: insufficient arguments>"
@@ -1096,7 +1096,7 @@ def _convert_take(self, call: relax.Call, args: List[Any]) -> Any:
             axis = call.attrs.axis
             if hasattr(axis, "value"):
                 axis = int(axis.value)
-            elif isinstance(axis, (int, float)):
+            elif isinstance(axis, int | float):
                 axis = int(axis)
 
         if axis is not None:
@@ -1110,7 +1110,7 @@ def _convert_take(self, call: relax.Call, args: List[Any]) -> Any:
             # No axis specified, use torch.take (flattens the tensor)
             return torch.take(tensor_arg, indices_arg)
 
-    def _convert_flip(self, call: relax.Call, args: List[Any]) -> Any:
+    def _convert_flip(self, call: relax.Call, args: list[Any]) -> Any:
         """Convert flip operation."""
         if len(call.args) < 1:
             return "<flip_error: insufficient arguments>"
@@ -1123,7 +1123,7 @@ def _convert_flip(self, call: relax.Call, args: List[Any]) -> Any:
             axis = call.attrs.axis
             if hasattr(axis, "value"):
                 axis = int(axis.value)
-            elif isinstance(axis, (int, float)):
+            elif isinstance(axis, int | float):
                 axis = int(axis)
 
         if axis is not None:
@@ -1135,7 +1135,7 @@ def _convert_flip(self, call: relax.Call, args: List[Any]) -> Any:
 
         return torch.flip(tensor_arg, dims=dims)
 
-    def _convert_tile(self, call: relax.Call, args: List[Any]) -> Any:
+    def _convert_tile(self, call: relax.Call, args: list[Any]) -> Any:
         """Convert tile operation."""
         if len(call.args) < 1:
             return "<tile_error: insufficient arguments>"
@@ -1148,7 +1148,7 @@ def _convert_tile(self, call: relax.Call, args: List[Any]) -> Any:
             # Handle TVM Array type
             if hasattr(repeats, "__iter__") and not isinstance(repeats, str):
                 repeats = tuple(int(v.value) if hasattr(v, "value") else int(v) for v in repeats)
-            elif isinstance(repeats, (list, tuple)):
+            elif isinstance(repeats, list | tuple):
                 repeats = tuple(int(v) for v in repeats)
             else:
                 repeats = (int(repeats),)
@@ -1157,7 +1157,7 @@ def _convert_tile(self, call: relax.Call, args: List[Any]) -> Any:
 
         return torch.tile(tensor_arg, dims=repeats)
 
-    def _convert_repeat(self, call: relax.Call, args: List[Any]) -> Any:
+    def _convert_repeat(self, call: relax.Call, args: list[Any]) -> Any:
         """Convert repeat operation."""
         if len(call.args) < 1:
             return "<repeat_error: insufficient arguments>"
@@ -1172,14 +1172,14 @@ def _convert_repeat(self, call: relax.Call, args: List[Any]) -> Any:
             repeats = call.attrs.repeats
             if hasattr(repeats, "value"):
                 repeats = int(repeats.value)
-            elif isinstance(repeats, (int, float)):
+            elif isinstance(repeats, int | float):
                 repeats = int(repeats)
 
         if call.attrs and hasattr(call.attrs, "axis"):
             axis = call.attrs.axis
             if hasattr(axis, "value"):
                 axis = int(axis.value)
-            elif isinstance(axis, (int, float)):
+            elif isinstance(axis, int | float):
                 axis = int(axis)
 
         if axis is not None:
@@ -1194,9 +1194,7 @@ def _convert_shape_expr(self, shape_expr: relax.ShapeExpr) -> Any:
         return f"<shape: {shape_expr}>"
 
 
-def convert_relax_to_pyfunc(
-    ir_module: IRModule, relax_function_names: Union[str, List[str]]
-) -> IRModule:
+def convert_relax_to_pyfunc(ir_module: IRModule, relax_function_names: str | list[str]) -> IRModule:
     """Convert Relax functions to Python functions.
 
     Args:
diff --git a/python/tvm/relax/struct_info.py b/python/tvm/relax/struct_info.py
index 2f8ccc62e482..d1bbd7897aeb 100644
--- a/python/tvm/relax/struct_info.py
+++ b/python/tvm/relax/struct_info.py
@@ -18,7 +18,7 @@
 # ruff: noqa: F401
 """The struct info nodes of the Relax language."""
 
-from typing import List, Optional, Union
+from typing import Optional, Union
 
 import tvm_ffi
 
@@ -51,13 +51,13 @@ class PrimStructInfo(StructInfo):
        value.
     """
 
-    value: Optional[PrimExpr]
+    value: PrimExpr | None
     dtype: str
 
     def __init__(
         self,
-        dtype: Optional[Union[str, DataType]] = None,
-        value: Optional[Union[int, float, PrimExpr]] = None,
+        dtype: str | DataType | None = None,
+        value: int | float | PrimExpr | None = None,
         span: Span = None,
     ) -> None:
         # Guard against incorrect usage.  For backwards compatibility,
@@ -69,7 +69,7 @@ def __init__(
         # inline variable definitions when used in a function
         # signature, and requires separate arguments to distinguish
         # the two cases.)
-        if isinstance(dtype, (PrimExpr, int, float)):
+        if isinstance(dtype, PrimExpr | int | float):
             raise TypeError(
                 f"The first positional argument of PrimStructInfo must be the datatype, "
                 f", but received {type(dtype)}.  "
@@ -92,7 +92,7 @@ def __init__(
                     "However, the value {value} has dtype {value.dtype}, "
                     "but the specified dtype was {dtype}."
                 )
-            elif isinstance(value, (int, float)):
+            elif isinstance(value, int | float):
                 value = tvm.tir.const(value, dtype)
 
         # Use relax's default integer type if not otherwise specified.
@@ -122,12 +122,12 @@ class ShapeStructInfo(StructInfo):
     Do not specify values and ndim at the same time.
     """
 
-    values: Optional[List[PrimExpr]]
+    values: list[PrimExpr] | None
     ndim: int
     span: Span
 
     def __init__(
-        self, values: Optional[List[PrimExpr]] = None, ndim: int = -1, span: Span = None
+        self, values: list[PrimExpr] | None = None, ndim: int = -1, span: Span = None
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.ShapeStructInfo,
@@ -160,21 +160,21 @@ class TensorStructInfo(StructInfo):
     Do not specify shape and ndim at the same time.
     """
 
-    shape: Optional[Expr]
+    shape: Expr | None
     dtype: str
-    vdevice: Optional[VDevice]
+    vdevice: VDevice | None
     ndim: int
     span: Span
 
     def __init__(
         self,
-        shape: Union[Optional[Expr], List[PrimExpr]] = None,
+        shape: Expr | None | list[PrimExpr] = None,
         dtype: str = "float32",
-        vdevice: Union[Optional[VDevice], str] = None,
+        vdevice: VDevice | None | str = None,
         ndim: int = -1,
         span: Span = None,
     ) -> None:
-        if isinstance(shape, (list, tuple, Array)):
+        if isinstance(shape, list | tuple | Array):
             shape = ShapeExpr(shape)
         self.__init_handle_by_constructor__(
             _ffi_api.TensorStructInfo,
@@ -196,10 +196,10 @@ class TupleStructInfo(StructInfo):
         The struct info of the fields.
     """
 
-    fields: List[StructInfo]
+    fields: list[StructInfo]
     span: Span
 
-    def __init__(self, fields: List[StructInfo], span: Span = None) -> None:
+    def __init__(self, fields: list[StructInfo], span: Span = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.TupleStructInfo, fields, span)  # type: ignore
 
 
@@ -222,14 +222,14 @@ class FuncStructInfo(StructInfo):
         we still consider it impure.
     """
 
-    params: Optional[List[StructInfo]]
+    params: list[StructInfo] | None
     ret: StructInfo
-    derive_func: Optional[EnvFunc]
+    derive_func: EnvFunc | None
     purity: bool
     span: Span
 
     def __init__(
-        self, params: List[StructInfo], ret: StructInfo, purity: bool = True, span: Span = None
+        self, params: list[StructInfo], ret: StructInfo, purity: bool = True, span: Span = None
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.FuncStructInfo,
@@ -242,8 +242,8 @@ def __init__(
     @staticmethod
     def opaque_func(
         *,
-        ret: Optional[StructInfo] = None,
-        derive_func: Optional[Union[str, EnvFunc]] = None,
+        ret: StructInfo | None = None,
+        derive_func: str | EnvFunc | None = None,
         purity: bool = False,
         span: Span = None,
     ) -> "FuncStructInfo":
diff --git a/python/tvm/relax/testing/ast_printer.py b/python/tvm/relax/testing/ast_printer.py
index 8383e176eded..636dea360d8e 100644
--- a/python/tvm/relax/testing/ast_printer.py
+++ b/python/tvm/relax/testing/ast_printer.py
@@ -22,7 +22,7 @@
 but it can be useful for tutorials and debugging.
 """
 
-from typing import Iterable
+from collections.abc import Iterable
 
 import tvm
 from tvm import relax
diff --git a/python/tvm/relax/testing/lib_comparator.py b/python/tvm/relax/testing/lib_comparator.py
index ffc6ead39e48..1ad5c9405f2c 100644
--- a/python/tvm/relax/testing/lib_comparator.py
+++ b/python/tvm/relax/testing/lib_comparator.py
@@ -17,7 +17,7 @@
 # pylint: disable=unused-argument
 """Tools to compare libraries."""
 
-from typing import Iterable, List, Tuple, Union
+from collections.abc import Iterable
 
 import tvm
 import tvm.testing
@@ -64,8 +64,8 @@ def __init__(self, mod, device, verbose=True, rtol=1e-5, atol=1e-5):
     def compare(
         self,
         name: str,
-        ref_args: Union[List[tvm.runtime.Tensor], Tuple[tvm.runtime.Tensor, ...]],
-        new_args: Union[List[tvm.runtime.Tensor], Tuple[tvm.runtime.Tensor, ...]],
+        ref_args: list[tvm.runtime.Tensor] | tuple[tvm.runtime.Tensor, ...],
+        new_args: list[tvm.runtime.Tensor] | tuple[tvm.runtime.Tensor, ...],
         ret_indices: Iterable[int],
     ):
         """Comparison function, can be overloaded.
diff --git a/python/tvm/relax/testing/nn.py b/python/tvm/relax/testing/nn.py
index 755ba6c2b48c..8e36680c95a4 100644
--- a/python/tvm/relax/testing/nn.py
+++ b/python/tvm/relax/testing/nn.py
@@ -18,8 +18,8 @@
 # ruff: noqa: RUF005
 """PyTorch-like nn.Module API for constructing workloads."""
 
-import typing
-from typing import Any, Callable, List, Union
+from collections.abc import Callable
+from typing import Any
 
 import numpy as np  # type: ignore
 
@@ -38,7 +38,7 @@ def emit_te(func: Callable, *args: Any, **kwargs: Any) -> relax.Var:
 
 def checkpoint(
     func: Callable, *args: Any, **kwargs: Any
-) -> Union[relax.Var, List[relax.Var], List[Any]]:
+) -> relax.Var | list[relax.Var] | list[Any]:
     """Mark function(*args, **kwargs) should be computed in a checkpointed manner during backward.
 
     To be specific, args and kwargs will be checkpointed, and func(*args, **kwargs) will be
@@ -47,7 +47,7 @@ def checkpoint(
     args = [start_checkpoint(v) if isinstance(v, relax.Expr) else v for v in args]
     kwargs = {k: start_checkpoint(v) if isinstance(v, relax.Expr) else v for k, v in kwargs.items()}
     result = func(*args, **kwargs)
-    if isinstance(result, (list, tuple)):
+    if isinstance(result, list | tuple):
         result = [end_checkpoint(v) if isinstance(v, relax.Expr) else v for v in result]
     else:
         assert isinstance(result, relax.Expr)
@@ -57,7 +57,7 @@ def checkpoint(
 
 def emit_checkpoint(
     func: Callable, *args: Any, **kwargs: Any
-) -> Union[relax.Var, List[relax.Var], List[Any]]:
+) -> relax.Var | list[relax.Var] | list[Any]:
     """Mark function(*args, **kwargs) should be computed in a checkpointed manner during backward.
 
     To be specific, args and kwargs will be checkpointed, and func(*args, **kwargs) will be
@@ -77,7 +77,7 @@ def emit_checkpoint(
         for k, v in kwargs.items()
     }
     result = func(*args, **kwargs)
-    if isinstance(result, (list, tuple)):
+    if isinstance(result, list | tuple):
         result = list(result)
         for i, v in enumerate(result):
             if isinstance(v, relax.Expr):
@@ -93,11 +93,11 @@ def emit_checkpoint(
 
 
 def emit_checkpoint_sequential(
-    functions: List[Callable],
-    segments: Union[int, List[int]],
+    functions: list[Callable],
+    segments: int | list[int],
     input: relax.Var,
     checkpoint_last: bool = False,
-) -> Union[relax.Var, List[relax.Var], List[Any]]:
+) -> relax.Var | list[relax.Var] | list[Any]:
     """A helper function for checkpointing sequential models. This interface has similar purpose
     as torch.utils.checkpoint.checkpoint_sequential.
 
@@ -193,10 +193,8 @@ def _try_unique_name(name: str):
 class Placeholder(relax.Var):
     """A placeholder variable that can represent model input."""
 
-    def __init__(
-        self, shape: Union[List[Any], typing.Tuple[Any, ...]], dtype="float32", name="data"
-    ):
-        if not isinstance(shape, (list, tuple)):
+    def __init__(self, shape: list[Any] | tuple[Any, ...], dtype="float32", name="data"):
+        if not isinstance(shape, list | tuple):
             raise TypeError("the shape of Placeholder is expected to be a list or a tuple")
         super().__init__(_try_unique_name(name), relax.TensorStructInfo(shape, dtype))
 
@@ -204,10 +202,8 @@ def __init__(
 class Parameter(relax.Var):
     """A special kind of relax Var that represents model parameter(weight)."""
 
-    def __init__(
-        self, shape: Union[List[Any], typing.Tuple[Any, ...]], dtype="float32", name="param"
-    ):
-        if not isinstance(shape, (list, tuple)):
+    def __init__(self, shape: list[Any] | tuple[Any, ...], dtype="float32", name="param"):
+        if not isinstance(shape, list | tuple):
             raise TypeError("the shape of Parameter is expected to be a list or a tuple")
 
         super().__init__(_try_unique_name(name), relax.TensorStructInfo(shape, dtype))
@@ -251,7 +247,7 @@ def forward(self, input: relax.Expr) -> relax.Var:
 
     define_subroutine: bool = False
 
-    def parameters(self) -> List[Parameter]:
+    def parameters(self) -> list[Parameter]:
         """Return the list of parameters in the module."""
         return _unpack_params(self.__dict__)
 
@@ -263,7 +259,7 @@ def __call__(self, *args, **kwargs):
         return self.forward(*args, **kwargs)
 
 
-def _unpack_params(value: object) -> List[relax.Var]:
+def _unpack_params(value: object) -> list[relax.Var]:
     if isinstance(value, Parameter):
         return [value]
     if isinstance(value, Module):
@@ -273,7 +269,7 @@ def _unpack_params(value: object) -> List[relax.Var]:
         for v in value.values():
             params += _unpack_params(v)
         return params
-    if isinstance(value, (list, tuple)):
+    if isinstance(value, list | tuple):
         params = []
         for v in value:
             params += _unpack_params(v)
@@ -281,7 +277,7 @@ def _unpack_params(value: object) -> List[relax.Var]:
     return []
 
 
-def init_params(mod: tvm.IRModule) -> List[tvm.runtime.Tensor]:
+def init_params(mod: tvm.IRModule) -> list[tvm.runtime.Tensor]:
     """Utility function to initialize model's parameters."""
     shape_dict = {v.name_hint: v.struct_info.shape for v in mod["main"].params}
     params = []
diff --git a/python/tvm/relax/testing/transform.py b/python/tvm/relax/testing/transform.py
index 843d6ef6310c..14da04ebd7d3 100644
--- a/python/tvm/relax/testing/transform.py
+++ b/python/tvm/relax/testing/transform.py
@@ -19,7 +19,6 @@
 
 import logging
 import os
-from typing import Dict, List, Set, Tuple
 
 import tvm_ffi
 
@@ -35,7 +34,7 @@ def ApplyEmptyCppMutator() -> tvm.ir.transform.Pass:
     return packed_func()
 
 
-def dataflow_liveness_analysis(block: DataflowBlock) -> Dict[Var, Tuple[int, int]]:
+def dataflow_liveness_analysis(block: DataflowBlock) -> dict[Var, tuple[int, int]]:
     """
     Inner function for the dataflow inplace transformation exposed for testing.
     """
@@ -50,8 +49,8 @@ def dataflow_liveness_analysis(block: DataflowBlock) -> Dict[Var, Tuple[int, int
 
 
 def dataflow_alias_analysis(
-    block: DataflowBlock, inputs: List[Var]
-) -> Tuple[Dict[Var, Set[int]], Dict[int, List[Set[int]]]]:
+    block: DataflowBlock, inputs: list[Var]
+) -> tuple[dict[Var, set[int]], dict[int, list[set[int]]]]:
     """
     Inner function for the dataflow inplace transformation exposed for testing.
     """
@@ -91,8 +90,8 @@ def __init__(self, _binding_idx, _arg_idxs):
 
 
 def dataflow_inplace_analysis(
-    block: DataflowBlock, inputs: List[Var], mod: IRModule
-) -> Tuple[List[Tuple[int, Set[int]]], List[Tuple[int, Set[int]]]]:
+    block: DataflowBlock, inputs: list[Var], mod: IRModule
+) -> tuple[list[tuple[int, set[int]]], list[tuple[int, set[int]]]]:
     """
     Inner function for the dataflow inplace transformation exposed for testing.
     """
@@ -109,8 +108,8 @@ def convert(opp_list):
 
 
 def dataflow_single_inplace_call(
-    mod: IRModule, call: Call, inplace_indices: List[int]
-) -> Tuple[Call, IRModule]:
+    mod: IRModule, call: Call, inplace_indices: list[int]
+) -> tuple[Call, IRModule]:
     """
     Inner function for the dataflow inplace transformation exposed for testing.
     """
diff --git a/python/tvm/relax/testing/vm.py b/python/tvm/relax/testing/vm.py
index fb08f95e2fd6..2537d8e97da1 100644
--- a/python/tvm/relax/testing/vm.py
+++ b/python/tvm/relax/testing/vm.py
@@ -17,7 +17,7 @@
 # pylint: disable=invalid-name
 """Testing utilities for relax VM"""
 
-from typing import Any, List
+from typing import Any
 
 import numpy as np  # type: ignore
 
@@ -75,7 +75,7 @@ def get_device_id(device):
     return device.index
 
 
-def check_saved_func(vm: relax.VirtualMachine, func_name: str, *inputs: List[Any]) -> Object:
+def check_saved_func(vm: relax.VirtualMachine, func_name: str, *inputs: list[Any]) -> Object:
     # uses save_function to create a closure with the given inputs
     # and ensure the result is the same
     # (assumes the functions return tensors and that they're idempotent)
diff --git a/python/tvm/relax/training/loss.py b/python/tvm/relax/training/loss.py
index 1c315161cf55..ba9f7b5ed0b8 100644
--- a/python/tvm/relax/training/loss.py
+++ b/python/tvm/relax/training/loss.py
@@ -18,10 +18,8 @@
 # ruff: noqa: RUF012
 """Loss functions library for relax."""
 
-from typing import Optional, Union
-
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 # isort: on
 
@@ -31,7 +29,7 @@
 from ..op.nn import log_softmax, nll_loss
 
 
-def _create_param_var(param: Union[Var, StructInfo], param_name: str) -> Var:
+def _create_param_var(param: Var | StructInfo, param_name: str) -> Var:
     """If param is a StructInfo, create a Var with the given StructInfo and name.
 
     If param is a Var, create a Var with the same StructInfo and name as the given param Var."""
@@ -140,8 +138,8 @@ def __init__(self, reduction: Literal["mean", "sum", "none"] = "mean") -> None:
 
     def __call__(
         self,
-        predictions: Union[Var, StructInfo],
-        targets: Union[Var, StructInfo],
+        predictions: Var | StructInfo,
+        targets: Var | StructInfo,
     ) -> Function:
         """Get the relax function of L1Loss. If the parameters are
         struct info, it will create corresponding variables.
@@ -189,8 +187,8 @@ def __init__(self, reduction: Literal["mean", "sum", "none"] = "mean") -> None:
 
     def __call__(
         self,
-        predictions: Union[Var, StructInfo],
-        targets: Union[Var, StructInfo],
+        predictions: Var | StructInfo,
+        targets: Var | StructInfo,
     ) -> Function:
         """Get the relax function of MSELoss. If the parameters are
         struct info, it will create corresponding variables.
@@ -249,9 +247,9 @@ def __init__(
 
     def __call__(
         self,
-        predictions: Union[Var, StructInfo],
-        targets: Union[Var, StructInfo],
-        weights: Optional[Union[Var, StructInfo]] = None,
+        predictions: Var | StructInfo,
+        targets: Var | StructInfo,
+        weights: Var | StructInfo | None = None,
     ) -> Function:
         """Get the relax function of CrossEntropyLoss. If the parameters are
         struct info, it will create corresponding variables.
@@ -322,9 +320,9 @@ def __init__(
 
     def __call__(
         self,
-        predictions: Union[Var, StructInfo],
-        targets: Union[Var, StructInfo],
-        weights: Optional[Union[Var, StructInfo]] = None,
+        predictions: Var | StructInfo,
+        targets: Var | StructInfo,
+        weights: Var | StructInfo | None = None,
     ) -> Function:
         """Get the relax function of CategoricalCrossEntropyLoss. If the parameters are
         struct info, it will create corresponding variables.
diff --git a/python/tvm/relax/training/optimizer.py b/python/tvm/relax/training/optimizer.py
index cea95e175610..505057c736fe 100644
--- a/python/tvm/relax/training/optimizer.py
+++ b/python/tvm/relax/training/optimizer.py
@@ -18,7 +18,7 @@
 """Provide abstraction for defining optimizers and a set of common optimizers."""
 
 from decimal import Decimal
-from typing import List, Optional, Tuple, Union
+from typing import Optional, Union
 
 import numpy as np  # type: ignore
 
@@ -101,7 +101,7 @@ class Optimizer:
 
     dtype: str
     name: str
-    param_list: List[Var]
+    param_list: list[Var]
     state: tvm.ir.Array
 
     def __init__(self, name: str) -> None:
@@ -110,7 +110,7 @@ def __init__(self, name: str) -> None:
         self.state = None
         self.dtype = None
 
-    def init(self, params: Union[Var, List[Var]]) -> "Optimizer":
+    def init(self, params: Var | list[Var]) -> "Optimizer":
         """Set the parameters, determine the dtype, and construct the initial state for the
         optimizer.
 
@@ -135,7 +135,7 @@ def init(self, params: Union[Var, List[Var]]) -> "Optimizer":
         self.state = None
         return self
 
-    def _set_params_and_dtype(self, params: List[Var]) -> None:
+    def _set_params_and_dtype(self, params: list[Var]) -> None:
         """Check params is legal and set the param_list and dtype of the optimizer."""
         params_set = set()
         dtype = None
@@ -227,7 +227,7 @@ def SGD(
 
 
 # TODO(chaofan, yixin): Support symbolic shapes
-def _get_shape_as_int_list(var: Var) -> List[int]:
+def _get_shape_as_int_list(var: Var) -> list[int]:
     return [int(val) for val in var.struct_info.shape]
 
 
@@ -268,7 +268,7 @@ def __init__(self, lr: float, weight_decay: float = 0) -> None:
         self.lr = float(lr)
         self.weight_decay = float(weight_decay)
 
-    def init(self, params: Union[Var, List[Var]]) -> "SGD":
+    def init(self, params: Var | list[Var]) -> "SGD":
         """Set the parameters, determine the dtype, and construct the initial state for the
         optimizer.
 
@@ -409,7 +409,7 @@ def __init__(
         self.dampening = float(dampening)
         self.nesterov = nesterov
 
-    def init(self, params: Union[Var, List[Var]]) -> "MomentumSGD":
+    def init(self, params: Var | list[Var]) -> "MomentumSGD":
         """Set the parameters, determine the dtype, and construct the initial state for the
         optimizer.
 
@@ -561,7 +561,7 @@ def Adam(param_tuple, grad_tuple, state_tuple):
     def __init__(
         self,
         lr: float,
-        betas: Tuple[float, float] = (0.9, 0.999),
+        betas: tuple[float, float] = (0.9, 0.999),
         eps: float = 1e-08,
         weight_decay: float = 0,
     ) -> None:
@@ -572,7 +572,7 @@ def __init__(
         self.eps = float(eps)
         self.weight_decay = float(weight_decay)
 
-    def init(self, params: Union[Var, List[Var]]) -> "Adam":
+    def init(self, params: Var | list[Var]) -> "Adam":
         """Set the parameters, determine the dtype, and construct the initial state for the
         optimizer.
 
diff --git a/python/tvm/relax/training/setup_trainer.py b/python/tvm/relax/training/setup_trainer.py
index 1b1dff25aba2..a32279c181ec 100644
--- a/python/tvm/relax/training/setup_trainer.py
+++ b/python/tvm/relax/training/setup_trainer.py
@@ -17,8 +17,6 @@
 # pylint: disable=not-callable, unused-argument
 """Setup Trainer Pass."""
 
-from typing import List
-
 import tvm
 from tvm import TVMError
 from tvm.ir.module import IRModule
@@ -120,7 +118,7 @@ def optimizer(params, gradients, optim_states):
     STATE_NUM_ATTR_KEY: str = "state_num"
 
     def __init__(
-        self, loss: Loss, optimizer: Optimizer, loss_args: List[TensorStructInfo], legalize=True
+        self, loss: Loss, optimizer: Optimizer, loss_args: list[TensorStructInfo], legalize=True
     ):
         self._loss = loss
         self._optimizer = optimizer
@@ -140,14 +138,14 @@ def _check_well_formed(self, mod: IRModule):
 
         # Check function attrs
         if self.PARAM_NUM_ATTR_KEY not in mod.attrs or not isinstance(
-            mod.attrs[self.PARAM_NUM_ATTR_KEY], (IntImm, int)
+            mod.attrs[self.PARAM_NUM_ATTR_KEY], IntImm | int
         ):
             raise ValueError(
                 f"SetupTrainer: The backbone module should has an integer attribute named "
                 f"{self.PARAM_NUM_ATTR_KEY}"
             )
         if self.STATE_NUM_ATTR_KEY not in mod.attrs or not isinstance(
-            mod.attrs[self.STATE_NUM_ATTR_KEY], (IntImm, int)
+            mod.attrs[self.STATE_NUM_ATTR_KEY], IntImm | int
         ):
             raise ValueError(
                 f"SetupTrainer: The backbone module should has an integer attribute named "
diff --git a/python/tvm/relax/training/trainer.py b/python/tvm/relax/training/trainer.py
index 612491a57478..ce9d2368b7c3 100644
--- a/python/tvm/relax/training/trainer.py
+++ b/python/tvm/relax/training/trainer.py
@@ -17,8 +17,6 @@
 # pylint: disable=invalid-name
 """Unified Trainer API for relax training."""
 
-from typing import Dict, List, Optional, Union
-
 import numpy as np  # type: ignore
 
 import tvm
@@ -102,13 +100,13 @@ def __init__(
             )
         ]
 
-        self._params: List[Optional[Tensor]] = [None] * self._param_num
-        self._param_name_to_pos: Dict[str, int] = {
+        self._params: list[Tensor | None] = [None] * self._param_num
+        self._param_name_to_pos: dict[str, int] = {
             p.name_hint: i for i, p in enumerate(self._param_vars)
         }
 
-        self._states: List[Optional[Tensor]] = [None] * self._state_num
-        self._state_name_to_pos: Dict[str, int] = {
+        self._states: list[Tensor | None] = [None] * self._state_num
+        self._state_name_to_pos: dict[str, int] = {
             s.name_hint: i for i, s in enumerate(self._state_vars)
         }
 
@@ -155,7 +153,7 @@ def zero_init_states(self):
 
     def load_params(
         self,
-        params: Union[List[Union[np.ndarray, Tensor]], Dict[str, Union[np.ndarray, Tensor]]],
+        params: list[np.ndarray | Tensor] | dict[str, np.ndarray | Tensor],
     ):
         """Load parameters from a dict or a list. Will convert parameters into tvm.runtime.Tensor
         in self.device.
@@ -189,7 +187,7 @@ def load_params(
 
     def load_states(
         self,
-        states: Union[List[Union[np.ndarray, Tensor]], Dict[str, Union[np.ndarray, Tensor]]],
+        states: list[np.ndarray | Tensor] | dict[str, np.ndarray | Tensor],
     ):
         """Load model states from a dict or a list. Will convert states into tvm.runtime.Tensor
         in self.device.
@@ -221,7 +219,7 @@ def load_states(
         else:
             raise ValueError("The type of extern_states should be either list or dict")
 
-    def export_params(self) -> Dict[str, Tensor]:
+    def export_params(self) -> dict[str, Tensor]:
         """Export parameters to a dict (parameter name -> Tensor).
 
         Returns
@@ -231,7 +229,7 @@ def export_params(self) -> Dict[str, Tensor]:
         """
         return {key: self._params[pos] for key, pos in self._param_name_to_pos.items()}
 
-    def export_states(self) -> Dict[str, Tensor]:
+    def export_states(self) -> dict[str, Tensor]:
         """Export model states to a dict (parameter name -> Tensor).
 
         Returns
@@ -257,7 +255,7 @@ def _check_inited(self):
                 "inference."
             )
 
-    def predict(self, *input_instances: Union[np.ndarray, Tensor]) -> Tensor:
+    def predict(self, *input_instances: np.ndarray | Tensor) -> Tensor:
         """Call the `backbone` function and return the prediction result of the backbone.
 
         Parameters
@@ -275,7 +273,7 @@ def predict(self, *input_instances: Union[np.ndarray, Tensor]) -> Tensor:
         self._check_inited()
         if len(input_instances) != self._input_num:
             raise ValueError("The length of the input does not match the backbone")
-        all_inputs: List[Tensor] = (
+        all_inputs: list[Tensor] = (
             [tvm.runtime.tensor(i, self.device) for i in input_instances]
             + self._params
             + self._states
@@ -291,8 +289,8 @@ def predict(self, *input_instances: Union[np.ndarray, Tensor]) -> Tensor:
 
     def update(
         self,
-        input_instances: Union[np.ndarray, Tensor, List[Union[np.ndarray, Tensor]]],
-        targets: Union[np.ndarray, Tensor, List[Union[np.ndarray, Tensor]]],
+        input_instances: np.ndarray | Tensor | list[np.ndarray | Tensor],
+        targets: np.ndarray | Tensor | list[np.ndarray | Tensor],
     ) -> Tensor:
         """Update parameters and model states. It will calculate the gradients of parameters
         and update them using the `optimizer` function.
@@ -329,7 +327,7 @@ def update(
         if len(input_instances) != self._input_num:
             raise ValueError("The length of the input does not match the backbone")
 
-        all_inputs: List[Tensor] = (
+        all_inputs: list[Tensor] = (
             [tvm.runtime.tensor(i, self.device) for i in input_instances]
             + self._params
             + self._states
@@ -352,8 +350,8 @@ def update(
 
     def profile_adjoint(
         self,
-        input_instances: List[Union[np.ndarray, Tensor]],
-        targets: List[Union[np.ndarray, Tensor]],
+        input_instances: list[np.ndarray | Tensor],
+        targets: list[np.ndarray | Tensor],
     ) -> tvm.runtime.profiling.Report:
         """Profile the adjoint function. It requires the VM to be constructed with `profile=True`,
         and runs `tvm.relax.VirtualMachine.profile()` internally.
@@ -387,7 +385,7 @@ def profile_adjoint(
         if len(input_instances) != self._input_num:
             raise ValueError("The length of the input does not match the backbone")
 
-        all_inputs: List[Tensor] = (
+        all_inputs: list[Tensor] = (
             [tvm.runtime.tensor(i) for i in input_instances]
             + self._params
             + self._states
diff --git a/python/tvm/relax/training/utils.py b/python/tvm/relax/training/utils.py
index 4ab620cf731c..395f8c7fe23a 100644
--- a/python/tvm/relax/training/utils.py
+++ b/python/tvm/relax/training/utils.py
@@ -17,7 +17,7 @@
 # pylint: disable=invalid-name, unused-argument
 """Utility functions for relax training."""
 
-from typing import Callable, Optional
+from collections.abc import Callable
 
 from tvm_ffi import register_global_func
 
@@ -33,7 +33,7 @@ def AppendLoss(
     func_name: str,
     loss_function: Function,
     num_backbone_outputs: int = 1,
-    new_func_name: Optional[str] = None,
+    new_func_name: str | None = None,
 ) -> tvm.ir.transform.Pass:
     """Append the loss function to the backbone function specified by `func_name`. Generally, the
     loss function is generated by instances of `relax.training.Loss`.
@@ -160,7 +160,7 @@ def predict_loss(x: R.Tensor((2, 4), "float32"), y: R.Tensor((2, 4), "float32"),
     )
 
 
-def register_te_gradient(te_grad_name: str, te_grad_func: Optional[Callable] = None):
+def register_te_gradient(te_grad_name: str, te_grad_func: Callable | None = None):
     """Register a te gradient function bind with name te_grad_name. te_grad_name can be referenced
     later in call_tir_with_grad nodes.
 
diff --git a/python/tvm/relax/transform/attach_external_modules.py b/python/tvm/relax/transform/attach_external_modules.py
index ab184abf343b..7becf3fd68a4 100644
--- a/python/tvm/relax/transform/attach_external_modules.py
+++ b/python/tvm/relax/transform/attach_external_modules.py
@@ -19,7 +19,7 @@
 Note: "external modules" here refers to `relax.frontend.nn.ExternModule`.
 """
 
-from typing import TYPE_CHECKING, List
+from typing import TYPE_CHECKING
 
 from tvm.ir import IRModule
 from tvm.ir.transform import PassContext, module_pass
@@ -32,7 +32,7 @@
 class AttachExternModules:  # pylint: disable=too-few-public-methods
     """Attach variable bounds to each Relax function, which primarily helps with memory planning."""
 
-    def __init__(self, extern_modules: List["ExternModule"]):
+    def __init__(self, extern_modules: list["ExternModule"]):
         self.extern_modules = extern_modules
 
     def transform_module(self, mod: IRModule, _ctx: PassContext) -> IRModule:
diff --git a/python/tvm/relax/transform/ipc_allreduce_rewrite.py b/python/tvm/relax/transform/ipc_allreduce_rewrite.py
index de5c22863403..0dee1440047f 100644
--- a/python/tvm/relax/transform/ipc_allreduce_rewrite.py
+++ b/python/tvm/relax/transform/ipc_allreduce_rewrite.py
@@ -18,8 +18,6 @@
 The pass is written in Python for experiment, fast development.
 """
 
-from typing import Dict
-
 import tvm
 from tvm import relax
 from tvm.ir.module import IRModule
@@ -59,12 +57,12 @@ def transform_module(self, mod: IRModule, _ctx: tvm.transform.PassContext) -> IR
 class _Visitor(PyExprVisitor):  # pylint: disable=abstract-method
     def __init__(self, allreduce_strategy: int) -> None:
         self.allreduce_strategy = allreduce_strategy
-        self.alloc_map: Dict[Var, relax.Call] = {}
-        self.binding_replacement_map: Dict[relax.Expr, relax.Expr] = {}
+        self.alloc_map: dict[Var, relax.Call] = {}
+        self.binding_replacement_map: dict[relax.Expr, relax.Expr] = {}
         self.builtin_alloc_tensor_op = tvm.ir.Op.get("relax.builtin.alloc_tensor")
         self.reshape_op = tvm.ir.Op.get("relax.reshape")
 
-    def visit(self, mod: IRModule) -> Dict[relax.Expr, relax.Expr]:
+    def visit(self, mod: IRModule) -> dict[relax.Expr, relax.Expr]:
         """Entry point"""
         for _, func in mod.functions_items():
             if isinstance(func, relax.Function):
@@ -128,7 +126,7 @@ class _Rewriter(PyExprMutator):
     """Rewrite the IRModule according to the binding replacement provided by the visitor."""
 
     def __init__(
-        self, mod: IRModule, binding_replacement_map: Dict[relax.Expr, relax.Expr]
+        self, mod: IRModule, binding_replacement_map: dict[relax.Expr, relax.Expr]
     ) -> None:
         super().__init__(mod)
         self.mod = mod
diff --git a/python/tvm/relax/transform/lazy_transform_params.py b/python/tvm/relax/transform/lazy_transform_params.py
index b7540d339874..b254519e110e 100644
--- a/python/tvm/relax/transform/lazy_transform_params.py
+++ b/python/tvm/relax/transform/lazy_transform_params.py
@@ -18,8 +18,6 @@
 # ruff: noqa: RUF005
 """Relax LazyTransformParams pass."""
 
-from typing import Optional
-
 import tvm
 from tvm import IRModule, relax
 from tvm.relax.expr_functor import PyExprMutator, PyExprVisitor, mutator, visitor
@@ -218,7 +216,7 @@ def unpack_sinfo(sinfo):
             # direct iterate over the struct info annotation
             for param in func.params[num_input:]:
                 for sinfo in unpack_sinfo(param.struct_info):
-                    if isinstance(sinfo, (relax.PrimStructInfo, relax.ShapeStructInfo)):
+                    if isinstance(sinfo, relax.PrimStructInfo | relax.ShapeStructInfo):
                         params.append(relax.Var("symbolic_var_holder", sinfo))
 
         return relax.Function(
@@ -232,7 +230,7 @@ def unpack_sinfo(sinfo):
 
 @mutator
 class LazyInputMutator(PyExprMutator):
-    def __init__(self, func_creator, mod: Optional[IRModule] = None) -> None:
+    def __init__(self, func_creator, mod: IRModule | None = None) -> None:
         self.func_creator = func_creator
         super().__init__(mod)
 
@@ -301,7 +299,7 @@ def visit_tuple_getitem_(self, node: relax.TupleGetItem) -> relax.Expr:
 
 @mutator
 class LazyOutputMutator(PyExprMutator):
-    def __init__(self, func_creator, mod: Optional[IRModule] = None) -> None:
+    def __init__(self, func_creator, mod: IRModule | None = None) -> None:
         self.func_creator = func_creator
         self.killed_vars = set()
         super().__init__(mod)
diff --git a/python/tvm/relax/transform/legalize_ops/common.py b/python/tvm/relax/transform/legalize_ops/common.py
index 98d063ff6a3f..1d76d8614b41 100644
--- a/python/tvm/relax/transform/legalize_ops/common.py
+++ b/python/tvm/relax/transform/legalize_ops/common.py
@@ -16,7 +16,7 @@
 # under the License.
 """Common functionality for legalization."""
 
-from typing import Callable, Optional, Union
+from collections.abc import Callable
 
 import tvm
 from tvm import te
@@ -43,7 +43,7 @@
 
 def _try_convert_to_scalar_const(
     expr: Expr, python_native: bool = False
-) -> Union[Expr, FloatImm, IntImm, bool, float, int]:
+) -> Expr | FloatImm | IntImm | bool | float | int:
     """Check if the input Expr is a scalar constant.
     If it is, return its plain value with the same data type or in native python type.
     If it is not, return the input expr.
@@ -79,7 +79,7 @@ def _try_convert_to_scalar_const(
     return expr
 
 
-def _call_topi_without_attr(te_func: TEFunc, primfunc_name: Optional[str] = None) -> LegalizeFunc:
+def _call_topi_without_attr(te_func: TEFunc, primfunc_name: str | None = None) -> LegalizeFunc:
     """A common wrapper util for the ops who has no attributes and whose
     legalization is simply passing its arguments to some TE function.
 
diff --git a/python/tvm/relax/transform/legalize_ops/create.py b/python/tvm/relax/transform/legalize_ops/create.py
index 354b09b677fe..3ab432056f2d 100644
--- a/python/tvm/relax/transform/legalize_ops/create.py
+++ b/python/tvm/relax/transform/legalize_ops/create.py
@@ -18,8 +18,6 @@
 # ruff: noqa: E731
 """Default legalization function for creation operators."""
 
-from typing import Optional
-
 import numpy as np
 
 from tvm import tir, topi
@@ -29,7 +27,7 @@
 from .common import LegalizeFunc, _try_convert_to_scalar_const, register_legalize
 
 
-def _full(is_like: bool, fill_value: Optional[float], primfunc_name: str) -> LegalizeFunc:
+def _full(is_like: bool, fill_value: float | None, primfunc_name: str) -> LegalizeFunc:
     def full_call_te(bb: BlockBuilder, call: Call) -> Expr:
         _fill_value = (
             _try_convert_to_scalar_const(call.args[1], python_native=True)
@@ -110,7 +108,7 @@ def _arange(bb: BlockBuilder, call: Call) -> Expr:
     dtype = call.attrs.dtype
 
     def is_const_scalar(x: PrimValue):
-        return isinstance(x.value, (tir.IntImm, tir.FloatImm))
+        return isinstance(x.value, tir.IntImm | tir.FloatImm)
 
     if all([is_const_scalar(x) for x in call.args]):
         return const(np.arange(start.value, end.value, step.value, dtype=dtype), dtype=dtype)
diff --git a/python/tvm/relax/transform/legalize_ops/linear_algebra.py b/python/tvm/relax/transform/legalize_ops/linear_algebra.py
index 1392d14f5ddd..f8dcab0a752a 100644
--- a/python/tvm/relax/transform/legalize_ops/linear_algebra.py
+++ b/python/tvm/relax/transform/legalize_ops/linear_algebra.py
@@ -108,11 +108,11 @@ def _einsum(bb: BlockBuilder, call: Call) -> Expr:
     n_field = len(t.struct_info.fields)
     while isinstance(t, Var):
         binding = bb.lookup_binding(t)
-        if not isinstance(binding, (Tuple, Var)):
+        if not isinstance(binding, Tuple | Var):
             break
         t = binding
 
-    assert isinstance(t, (Tuple, Var))
+    assert isinstance(t, Tuple | Var)
     fields = (
         t.fields if isinstance(t, Tuple) else [bb.emit(TupleGetItem(t, i)) for i in range(n_field)]
     )
diff --git a/python/tvm/relax/transform/legalize_ops/manipulate.py b/python/tvm/relax/transform/legalize_ops/manipulate.py
index d087fcf8c025..41b191a3bf8b 100644
--- a/python/tvm/relax/transform/legalize_ops/manipulate.py
+++ b/python/tvm/relax/transform/legalize_ops/manipulate.py
@@ -18,8 +18,6 @@
 # ruff: noqa: RUF005
 """Default legalization function for manipulate operators."""
 
-from typing import Optional
-
 import tvm
 from tvm import relax, s_tir, te, tir, topi
 from tvm.relax.op.base import call_tir
@@ -62,11 +60,11 @@ def _concat(bb: BlockBuilder, call: Call) -> Expr:
     n_field = len(t.struct_info.fields)
     while isinstance(t, Var):
         binding = bb.lookup_binding(t)
-        if not isinstance(binding, (Tuple, Var)):
+        if not isinstance(binding, Tuple | Var):
             break
         t = binding
 
-    assert isinstance(t, (Tuple, Var))
+    assert isinstance(t, Tuple | Var)
     fields = (
         t.fields if isinstance(t, Tuple) else [bb.emit(TupleGetItem(t, i)) for i in range(n_field)]
     )
@@ -130,11 +128,11 @@ def _stack(bb: BlockBuilder, call: Call) -> Expr:
     # Follow bindings to find the actual tuple
     while isinstance(t, Var):
         binding = bb.lookup_binding(t)
-        if not isinstance(binding, (Tuple, Var)):
+        if not isinstance(binding, Tuple | Var):
             break
         t = binding
 
-    assert isinstance(t, (Tuple, Var))
+    assert isinstance(t, Tuple | Var)
 
     # Extract fields from either Tuple or bound Var
     fields = (
@@ -146,7 +144,7 @@ def _stack(bb: BlockBuilder, call: Call) -> Expr:
 
 @register_legalize("relax.repeat")
 def _repeat(bb: BlockBuilder, call: Call) -> Expr:
-    def te_repeat(data: te.Tensor, repeats: IntImm, axis: Optional[IntImm]):
+    def te_repeat(data: te.Tensor, repeats: IntImm, axis: IntImm | None):
         if axis is None:
             # flatten data
             out_shape = data.shape[0]
@@ -224,11 +222,11 @@ def _meshgrid(bb: BlockBuilder, call: Call) -> Expr:
     n_field = len(t.struct_info.fields)
     while isinstance(t, Var):
         binding = bb.lookup_binding(t)
-        if not isinstance(binding, (Tuple, Var)):
+        if not isinstance(binding, Tuple | Var):
             break
         t = binding
 
-    assert isinstance(t, (Tuple, Var))
+    assert isinstance(t, Tuple | Var)
     fields = (
         t.fields if isinstance(t, Tuple) else [bb.emit(TupleGetItem(t, i)) for i in range(n_field)]
     )
diff --git a/python/tvm/relax/transform/legalize_ops/nn.py b/python/tvm/relax/transform/legalize_ops/nn.py
index 8c6a78933b26..fe01ec62410d 100644
--- a/python/tvm/relax/transform/legalize_ops/nn.py
+++ b/python/tvm/relax/transform/legalize_ops/nn.py
@@ -19,7 +19,6 @@
 
 import logging
 import math
-from typing import Optional
 
 from tvm import s_tir, te, tir, topi
 
@@ -671,7 +670,7 @@ def _te_attention(
     v: te.Tensor,
     bias: te.Tensor,
     scale: tir.FloatImm,
-    causal_mask: Optional[str],
+    causal_mask: str | None,
 ) -> te.Tensor:
     batch_size, seq_len, num_head, head_dim = q.shape
     _, seq_len_kv, _, head_dim_v = v.shape
diff --git a/python/tvm/relax/transform/legalize_ops/qdq.py b/python/tvm/relax/transform/legalize_ops/qdq.py
index 6788b42b86f3..206417c6d0a1 100644
--- a/python/tvm/relax/transform/legalize_ops/qdq.py
+++ b/python/tvm/relax/transform/legalize_ops/qdq.py
@@ -17,8 +17,6 @@
 # pylint: disable=invalid-name
 """Default legalization function for quantize/dequantize operators."""
 
-from typing import Union
-
 import tvm
 from tvm import te, tir
 
@@ -34,7 +32,7 @@ def clip_cast(val, dtype):
 
 
 def is_const_scalar(x):
-    return isinstance(x, (tvm.tir.IntImm, tvm.tir.FloatImm))
+    return isinstance(x, tvm.tir.IntImm | tvm.tir.FloatImm)
 
 
 @register_legalize("relax.quantize")
@@ -48,8 +46,8 @@ def _quantize(bb: BlockBuilder, call: Call) -> Expr:
 
     def te_quantize(
         data: te.Tensor,
-        scale: Union[te.Tensor, tir.IntImm, tir.FloatImm],
-        zp: Union[te.Tensor, tir.IntImm, tir.FloatImm],
+        scale: te.Tensor | tir.IntImm | tir.FloatImm,
+        zp: te.Tensor | tir.IntImm | tir.FloatImm,
     ):
         def quantize_compute(*indices):
             scale_value = scale if is_const_scalar(scale) else scale[indices[axis]]
@@ -96,8 +94,8 @@ def _dequantize(bb: BlockBuilder, call: Call) -> Expr:
 
     def te_dequantize(
         data: te.Tensor,
-        scale: Union[te.Tensor, tir.IntImm, tir.FloatImm],
-        zp: Union[te.Tensor, tir.IntImm, tir.FloatImm],
+        scale: te.Tensor | tir.IntImm | tir.FloatImm,
+        zp: te.Tensor | tir.IntImm | tir.FloatImm,
     ):
         def dequantize_compute(*indices):
             scale_value = scale if is_const_scalar(scale) else scale[indices[axis]]
diff --git a/python/tvm/relax/transform/legalize_ops/statistical.py b/python/tvm/relax/transform/legalize_ops/statistical.py
index c4f544635650..f9967e5f7bdd 100644
--- a/python/tvm/relax/transform/legalize_ops/statistical.py
+++ b/python/tvm/relax/transform/legalize_ops/statistical.py
@@ -17,8 +17,6 @@
 # pylint: disable=invalid-name
 """Default legalization function for statistical operators."""
 
-from typing import List, Tuple, Union
-
 from tvm import te, tir, topi
 
 from ...block_builder import BlockBuilder
@@ -33,7 +31,7 @@ def statistical_call_te(bb: BlockBuilder, call: Call) -> Expr:
     return statistical_call_te
 
 
-def _compute_shape_prod(x: te.Tensor, axis: List[tir.IntImm]) -> tir.PrimExpr:
+def _compute_shape_prod(x: te.Tensor, axis: list[tir.IntImm]) -> tir.PrimExpr:
     shape_prod = tir.const(1, "int32")
     axes = [_axis.value for _axis in axis] if axis is not None else range(0, len(x.shape))
     for dim in axes:
@@ -41,13 +39,13 @@ def _compute_shape_prod(x: te.Tensor, axis: List[tir.IntImm]) -> tir.PrimExpr:
     return shape_prod
 
 
-def _te_mean(x: te.Tensor, axis: List[tir.IntImm], keepdims: bool) -> te.Tensor:
+def _te_mean(x: te.Tensor, axis: list[tir.IntImm], keepdims: bool) -> te.Tensor:
     shape_prod = _compute_shape_prod(x, axis)
     res_sum = topi.sum(x, axis, keepdims)
     return topi.divide(res_sum, shape_prod)
 
 
-def _te_variance(x: te.Tensor, axis: List[tir.IntImm], keepdims: bool) -> te.Tensor:
+def _te_variance(x: te.Tensor, axis: list[tir.IntImm], keepdims: bool) -> te.Tensor:
     dev = x - _te_mean(x, axis, True)
     return _te_mean(dev * dev, axis, keepdims)
     # This version has better memory locality and performance
@@ -57,8 +55,8 @@ def _te_variance(x: te.Tensor, axis: List[tir.IntImm], keepdims: bool) -> te.Ten
 
 
 def _te_median(
-    x: te.Tensor, axis: List[tir.IntImm], keepdims: bool
-) -> Union[te.Tensor, Tuple[te.Tensor, te.Tensor]]:
+    x: te.Tensor, axis: list[tir.IntImm], keepdims: bool
+) -> te.Tensor | tuple[te.Tensor, te.Tensor]:
     # currently only supports one axis or no axis ~ same pytorch
     # todo: support multiple axis ~ same numpy
     shape_prod = _compute_shape_prod(x, axis)
@@ -99,7 +97,7 @@ def _mean(bb: BlockBuilder, call: Call) -> Expr:
 
 @register_legalize("relax.std")
 def _std(bb: BlockBuilder, call: Call) -> Expr:
-    def te_std(x: te.Tensor, axis: List[tir.IntImm], keepdims: bool) -> te.Tensor:
+    def te_std(x: te.Tensor, axis: list[tir.IntImm], keepdims: bool) -> te.Tensor:
         return topi.sqrt(_te_variance(x, axis, keepdims))
 
     return bb.call_te(
diff --git a/python/tvm/relax/transform/transform.py b/python/tvm/relax/transform/transform.py
index 2b88c41a0df6..8911423c65f4 100644
--- a/python/tvm/relax/transform/transform.py
+++ b/python/tvm/relax/transform/transform.py
@@ -22,7 +22,8 @@
 import inspect
 import types
 import warnings
-from typing import Callable, Dict, List, Mapping, Optional, Sequence, Tuple, Union
+from collections.abc import Callable, Mapping, Sequence
+from typing import Optional, Union
 
 import numpy as np  # type: ignore
 import tvm_ffi
@@ -52,7 +53,7 @@ class DataflowBlockPass(tvm.ir.transform.Pass):
 
 
 def Gradient(
-    func_name: str, require_grads: Optional[Union[Var, List[Var]]] = None, target_index: int = 0
+    func_name: str, require_grads: Var | list[Var] | None = None, target_index: int = 0
 ) -> tvm.ir.transform.Pass:
     """Reverse-mode automatic differentiation.
 
@@ -222,7 +223,7 @@ def main_adjoint(
                 # return value: (orig_return_values, tuple(adjoints))
                 return ((lv1, lv2), (x_adjoint, y_adjoint))
     """
-    if require_grads is not None and not isinstance(require_grads, (list, tvm_ffi.Array)):
+    if require_grads is not None and not isinstance(require_grads, list | tvm_ffi.Array):
         require_grads = [require_grads]
 
     return _ffi_api.Gradient(func_name, require_grads, target_index)  # type: ignore
@@ -641,7 +642,7 @@ def AttachGlobalSymbol() -> tvm.ir.transform.Pass:
 
 def BindParams(
     func_name: str,
-    params: Dict[Union[str, Var], Union[tvm.runtime.Tensor, np.ndarray]],
+    params: dict[str | Var, tvm.runtime.Tensor | np.ndarray],
 ) -> tvm.ir.transform.Pass:
     """Bind params of function of the module to constant tensors.
 
@@ -661,7 +662,7 @@ def BindParams(
     for k, v in params.items():
         if isinstance(v, np.ndarray):
             v = tvm.runtime.tensor(v)
-        assert isinstance(v, (tvm.runtime.Tensor, tvm.relax.Constant)), (
+        assert isinstance(v, tvm.runtime.Tensor | tvm.relax.Constant), (
             f"param values are expected to be TVM.Tensor,"
             f"numpy.ndarray or tvm.relax.Constant, but got {type(v)}"
         )
@@ -671,8 +672,8 @@ def BindParams(
 
 
 def BindSymbolicVars(
-    binding_map: Mapping[Union[str, tvm.tir.Var], tvm.tir.PrimExpr],
-    func_name: Optional[str] = None,
+    binding_map: Mapping[str | tvm.tir.Var, tvm.tir.PrimExpr],
+    func_name: str | None = None,
 ) -> tvm.ir.transform.Pass:
     """Bind params of function of the module to constant tensors.
 
@@ -699,8 +700,8 @@ def BindSymbolicVars(
 
 
 def RunCodegen(
-    target_options: Optional[dict] = None,
-    entry_functions: Optional[List[str]] = None,
+    target_options: dict | None = None,
+    entry_functions: list[str] | None = None,
 ) -> tvm.ir.transform.Pass:
     """Produce the runtime::Module with an annotated codegen and global symbol.
 
@@ -884,15 +885,15 @@ class FusionPattern(Object):
     pattern: DFPattern
     annotation_patterns: Mapping[str, DFPattern]
     check: Callable[[PatternCheckContext], bool]
-    attrs_getter: Callable[[Dict[str, Expr]], Dict[str, str]]
+    attrs_getter: Callable[[dict[str, Expr]], dict[str, str]]
 
     def __init__(
         self,
         name: str,
         pattern: DFPattern,
-        annotation_patterns: Optional[Mapping[str, DFPattern]] = None,
-        check: Optional[Callable[[PatternCheckContext], bool]] = None,
-        attrs_getter: Optional[Callable[[Dict[str, Expr]], Dict[str, str]]] = None,
+        annotation_patterns: Mapping[str, DFPattern] | None = None,
+        check: Callable[[PatternCheckContext], bool] | None = None,
+        attrs_getter: Callable[[dict[str, Expr]], dict[str, str]] | None = None,
     ):
         if annotation_patterns is None:
             annotation_patterns = {}
@@ -902,10 +903,10 @@ def __init__(
 
 
 def FuseOpsByPattern(
-    patterns: List[Union[FusionPattern, Tuple]],
+    patterns: list[FusionPattern | tuple],
     bind_constants: bool = True,
     annotate_codegen: bool = False,
-    entry_functions: Optional[List[str]] = None,
+    entry_functions: list[str] | None = None,
 ) -> tvm.ir.transform.Pass:
     """Apply pattern matching to each function in the given module, and group matched expressions
     into a new function.
@@ -1003,7 +1004,7 @@ def SplitLayoutRewritePreproc() -> tvm.ir.transform.Pass:
     return _ffi_api.SplitLayoutRewritePreproc()  # type: ignore
 
 
-def LiftTransformParams(shared_transform: Union[bool, List[str]] = False) -> tvm.ir.transform.Pass:
+def LiftTransformParams(shared_transform: bool | list[str] = False) -> tvm.ir.transform.Pass:
     """Lift transformation of the parameters of a function.
 
     When some inputs of the function is marked as 'parameters' (the model weights), this pass
@@ -1041,7 +1042,7 @@ def LiftTransformParams(shared_transform: Union[bool, List[str]] = False) -> tvm
     return _ffi_api.LiftTransformParams(shared_transform)  # type: ignore
 
 
-def BundleModelParams(param_tuple_name: Optional[str] = None) -> tvm.ir.transform.Pass:
+def BundleModelParams(param_tuple_name: str | None = None) -> tvm.ir.transform.Pass:
     """Bundle several model parameters into a single tuple paramters
 
     For each function, if the function has the attribute "num_input",
@@ -1065,8 +1066,8 @@ def BundleModelParams(param_tuple_name: Optional[str] = None) -> tvm.ir.transfor
 
 
 def LegalizeOps(
-    customize_legalize_map: Optional[Dict[str, LegalizeFunc]] = None,
-    skip_ops: Optional[List[str]] = None,
+    customize_legalize_map: dict[str, LegalizeFunc] | None = None,
+    skip_ops: list[str] | None = None,
     enable_warning: bool = False,
 ):
     """Legalize high-level operator calls in Relax functions to call_tir
@@ -1191,7 +1192,7 @@ def RealizeVDevice() -> tvm.ir.transform.Pass:
 
 
 def MetaScheduleApplyDatabase(
-    work_dir: Optional[str] = None, enable_warning: bool = False
+    work_dir: str | None = None, enable_warning: bool = False
 ) -> tvm.ir.transform.Pass:
     """Apply the best schedule from tuning database.
 
@@ -1232,11 +1233,11 @@ def MetaScheduleTuneTIR(
 
 
 def MetaScheduleTuneIRMod(
-    params: Dict[str, Tensor],
+    params: dict[str, Tensor],
     work_dir: str,
     max_trials_global: int,
-    max_trials_per_task: Optional[int] = None,
-    op_names: Optional[List[str]] = None,
+    max_trials_per_task: int | None = None,
+    op_names: list[str] | None = None,
 ) -> tvm.ir.transform.Pass:
     """Tune Relax IRModule with MetaSchedule.
 
@@ -1287,7 +1288,7 @@ def FewShotTuning(
     return _ffi_api.FewShotTuning(valid_count, benchmark)  # type: ignore
 
 
-def DecomposeOpsForInference(func_name: Optional[str] = None) -> tvm.ir.transform.Pass:
+def DecomposeOpsForInference(func_name: str | None = None) -> tvm.ir.transform.Pass:
     """Decompose composite operators that are composed by other operators during inference.
     For example, the result of batch norm (a triple) will be simplified. Attention, tensor_to_shape,
     etc. can be also decomposed into a number of simplified operators as well.
@@ -1306,7 +1307,7 @@ def DecomposeOpsForInference(func_name: Optional[str] = None) -> tvm.ir.transfor
     return _ffi_api.DecomposeOpsForInference(func_name)  # type: ignore
 
 
-def DecomposeOpsForTraining(func_name: Optional[str] = None) -> tvm.ir.transform.Pass:
+def DecomposeOpsForTraining(func_name: str | None = None) -> tvm.ir.transform.Pass:
     """Decompose composite operators that are composed by other operators during training.
     For example, the result of batch norm (a triple) will be simplified. Attention, tensor_to_shape,
     etc. can be also decomposed into a number of simplified operators as well.
@@ -1326,10 +1327,10 @@ def DecomposeOpsForTraining(func_name: Optional[str] = None) -> tvm.ir.transform
 
 
 def AlterOpImpl(
-    op_impl_map: Dict[str, PrimFunc],
-    op_buffer_transforms: Dict[str, List[Union[IndexMap, Callable]]],
-    op_buffer_axis_separators: Dict[str, List[Union[IndexMap.AXIS_SEPARATOR, Callable]]],
-    op_buffer_input_axis_separators: Dict[str, List[Union[IndexMap.AXIS_SEPARATOR, Callable]]],
+    op_impl_map: dict[str, PrimFunc],
+    op_buffer_transforms: dict[str, list[IndexMap | Callable]],
+    op_buffer_axis_separators: dict[str, list[str | Callable]],  # str=IndexMap.AXIS_SEPARATOR
+    op_buffer_input_axis_separators: dict[str, list[str | Callable]],  # str=IndexMap.AXIS_SEPARATOR
 ):
     """Replace all PrimFunc's which have matching 'operator_name' attribute, with replacement
     PrimFunc that could possibly have different layouts on i/o buffers. The layout
@@ -1358,7 +1359,7 @@ def AlterOpImpl(
             # Extract the index_map
             if isinstance(transform, Callable):
                 transform = IndexMap.from_func_with_separators(transform)[0]
-            elif isinstance(transform, (Array, tuple)) and isinstance(transform[0], IndexMap):
+            elif isinstance(transform, Array | tuple) and isinstance(transform[0], IndexMap):
                 transform = transform[0]
             l.append(transform)
         op_buffer_transforms[operator_name] = l
@@ -1372,8 +1373,8 @@ def AlterOpImpl(
 
 
 def ConvertLayout(
-    desired_layouts: Dict[str, List[str]],
-    layout_cb: Optional[Callable] = None,
+    desired_layouts: dict[str, list[str]],
+    layout_cb: Callable | None = None,
 ) -> tvm.ir.transform.Pass:
     """Automatic layout conversion pass.
 
@@ -1396,7 +1397,7 @@ def ConvertLayout(
     return _ffi_api.ConvertLayout(desired_layouts, layout_cb)  # type: ignore
 
 
-def DeadCodeElimination(entry_functions: Optional[List[str]] = None) -> tvm.ir.transform.Pass:
+def DeadCodeElimination(entry_functions: list[str] | None = None) -> tvm.ir.transform.Pass:
     """Remove dead code in the IRModule.
     Currently it removes:
 
@@ -1427,7 +1428,7 @@ def DeadCodeElimination(entry_functions: Optional[List[str]] = None) -> tvm.ir.t
 
 
 def ToMixedPrecision(
-    out_dtype="float32", fp16_input_names: Optional[List[str]] = None
+    out_dtype="float32", fp16_input_names: list[str] | None = None
 ) -> tvm.ir.transform.Pass:
     """Automatic mixed precision pass. Currently the pass assumes the input module to be fp32
     only, and will automatically cast fp32 to fp16 for certain ops.
@@ -1450,7 +1451,7 @@ def ToMixedPrecision(
     return _ffi_api.ToMixedPrecision(out_dtype, fp16_input_names)  # type: ignore
 
 
-def SplitCallTIRByPattern(patterns: List[PrimFunc], fcodegen: Callable) -> tvm.ir.transform.Pass:
+def SplitCallTIRByPattern(patterns: list[PrimFunc], fcodegen: Callable) -> tvm.ir.transform.Pass:
     """Split a PrimFunc into 2 parts: the first part is a TIR PrimFunc which is
        matched with some pattern, and the second part is the rest of the original
        PrimFunc. It will call fcodegen to generate the code for the matched pattern
@@ -1472,7 +1473,7 @@ def SplitCallTIRByPattern(patterns: List[PrimFunc], fcodegen: Callable) -> tvm.i
     return _ffi_api.SplitCallTIRByPattern(patterns, fcodegen)  # type: ignore
 
 
-def UpdateParamStructInfo(sinfo_func: Callable[[Var], Optional[StructInfo]]):
+def UpdateParamStructInfo(sinfo_func: Callable[[Var], StructInfo | None]):
     """Update struct info of parameters
 
     Update struct info of parameters.  Internal bindings and function
@@ -1671,7 +1672,7 @@ def function_pass(
     name=None,
     required=None,
     traceable=False,
-) -> Union[Callable, FunctionPass]:
+) -> Callable | FunctionPass:
     """Decorate a function pass.
 
     This function returns a callback when pass_func
@@ -1763,7 +1764,7 @@ def transform(func, mod, ctx):
         raise ValueError("Please provide opt_level for the function pass.")
 
     required = required if required else []
-    if not isinstance(required, (list, tuple)):
+    if not isinstance(required, list | tuple):
         raise TypeError("Required is expected to be the type of " + "list/tuple.")
 
     def create_function_pass(pass_arg):
@@ -1772,7 +1773,7 @@ def create_function_pass(pass_arg):
         info = tvm.transform.PassInfo(opt_level, fname, required, traceable)
         if inspect.isclass(pass_arg):
             return _wrap_class_function_pass(pass_arg, info)
-        if not isinstance(pass_arg, (types.FunctionType, types.LambdaType)):
+        if not isinstance(pass_arg, types.FunctionType | types.LambdaType):
             raise TypeError("pass_func must be a callable for Function pass")
         return _ffi_api.MakeFunctionPass(pass_arg, info)  # type: ignore
 
@@ -1815,7 +1816,7 @@ def __getattr__(self, name):
 
 def dataflowblock_pass(
     pass_func=None, opt_level=None, name=None, required=None, traceable=False
-) -> Union[Callable, DataflowBlockPass]:
+) -> Callable | DataflowBlockPass:
     """Decorate a dataflowblock pass.
 
     This function returns a callback when pass_func
@@ -1915,7 +1916,7 @@ def transform(block, mod, ctx):
         raise ValueError("Please provide opt_level for the dataflowblock pass.")
 
     required = required if required else []
-    if not isinstance(required, (list, tuple)):
+    if not isinstance(required, list | tuple):
         raise TypeError("Required is expected to be the type of " + "list/tuple.")
 
     def create_dataflowblock_pass(pass_arg):
@@ -1924,7 +1925,7 @@ def create_dataflowblock_pass(pass_arg):
         info = tvm.transform.PassInfo(opt_level, fname, required, traceable)
         if inspect.isclass(pass_arg):
             return _wrap_class_dataflowblock_pass(pass_arg, info)
-        if not isinstance(pass_arg, (types.FunctionType, types.LambdaType)):
+        if not isinstance(pass_arg, types.FunctionType | types.LambdaType):
             raise TypeError("pass_func must be a callable for DataflowBlock pass")
         return _ffi_api.MakeDataflowBlockPass(pass_arg, info)  # type: ignore
 
diff --git a/python/tvm/relax/type_converter.py b/python/tvm/relax/type_converter.py
index e84b457a0140..50014cba5a9f 100644
--- a/python/tvm/relax/type_converter.py
+++ b/python/tvm/relax/type_converter.py
@@ -29,7 +29,8 @@
 
 import functools
 import inspect
-from typing import Any, Callable, List, Optional, TypeVar
+from collections.abc import Callable
+from typing import Any, TypeVar
 
 import tvm
 
@@ -40,7 +41,7 @@ class _ArgsConverter:
     """A helper class to convert the arguments to Expr."""
 
     @staticmethod
-    def convert(args_to_expr: List[str], args_to_list_expr: List[str]):
+    def convert(args_to_expr: list[str], args_to_list_expr: list[str]):
         """Convert the arguments to Expr.
 
         Parameters
@@ -154,7 +155,7 @@ def auto(func: FType) -> FType:
         """Decorator for automatically convert the arguments to Expr according to type annotation.
         Only two patterns are supported:
 
-        1. The argument is Expr or Optional[Expr].
+        1. The argument is Expr or Expr | None.
 
         2. The argument is List[Expr] or Optional[List[Expr]].
 
@@ -167,9 +168,9 @@ def auto(func: FType) -> FType:
 
         for param in sig.parameters.values():
             anno = param.annotation
-            if anno in (Expr, Optional[Expr]):
+            if anno in (Expr, Expr | None):
                 args_to_expr.append(param.name)
-            if anno in (List[Expr], Optional[List[Expr]]):
+            if anno in (list[Expr], list[Expr] | None):
                 args_to_list_expr.append(param.name)
 
         return _ArgsConverter.convert(args_to_expr, args_to_list_expr)(func)
diff --git a/python/tvm/relax/utils.py b/python/tvm/relax/utils.py
index 7126947091f0..f76459294aba 100644
--- a/python/tvm/relax/utils.py
+++ b/python/tvm/relax/utils.py
@@ -22,8 +22,8 @@
 
 import itertools
 import string
-from typing import Any, Callable, Dict, List, Optional
-from typing import Tuple as typing_Tuple
+from collections.abc import Callable
+from typing import Any, Optional
 
 import tvm_ffi
 
@@ -43,7 +43,7 @@
 from .type_converter import args_converter  # pylint: disable=unused-import
 
 
-def metadata_partitioner(rx_txt: str) -> List[str]:
+def metadata_partitioner(rx_txt: str) -> list[str]:
     """Extract Relax program and metadata section.
 
     Parameters
@@ -117,10 +117,10 @@ def convert_to_expr(value: Any) -> Expr:
     if isinstance(tvm_value, PrimExpr):
         return PrimValue(value)
     # Case 3
-    if isinstance(tvm_value, (str,)):
+    if isinstance(tvm_value, str):
         return StringImm(value)
     # Case 4
-    if isinstance(value, (tuple, list)):
+    if isinstance(value, tuple | list):
         # `convert_to_expr` ensures that all elements are `Expr` if no exception raises
         return rx_Tuple([convert_to_expr(v) for v in value])
     raise TypeError(f"Cannot convert {value} with type {type(value)} to `relax.Expr`")
@@ -148,7 +148,7 @@ def copy_with_new_vars(func: Function) -> Function:
 
 def gen_call_tir_inputs(
     func: Callable, *args: Any, **kwargs: Any
-) -> typing_Tuple[tir.PrimFunc, Expr, List[TensorStructInfo], Optional[ShapeExpr]]:
+) -> tuple[tir.PrimFunc, Expr, list[TensorStructInfo], ShapeExpr | None]:
     """Generate the inputs for call_tir according to the te function.
     This function converts arguments from relax expression to te tensor,
     The callback func should return a te tensor or a list of te tensors.
@@ -173,7 +173,7 @@ def gen_call_tir_inputs(
         out_sinfo, and tir_vars.
     """
 
-    tir_var_map: Dict[tir.Var, tir.PrimExpr] = {}
+    tir_var_map: dict[tir.Var, tir.PrimExpr] = {}
 
     call_tir_args = []
     create_primfunc_args = []
@@ -270,11 +270,11 @@ def _convert_te_arg_helper(arg):
                     else:
                         return _convert_te_arg_helper(arg.struct_info.value)
 
-            elif isinstance(arg, (list, Array)):
+            elif isinstance(arg, list | Array):
                 return [_convert_te_arg_helper(x) for x in arg]
             elif isinstance(arg, tuple):
                 return tuple(_convert_te_arg_helper(x) for x in arg)
-            elif isinstance(arg, (dict, Map)):
+            elif isinstance(arg, dict | Map):
                 for key in arg:
                     assert isinstance(key, str), (
                         "emit_te only supports dict with string as the key currently"
@@ -285,7 +285,7 @@ def _convert_te_arg_helper(arg):
                 new_arg = tir.stmt_functor.substitute(arg, tir_var_map)
                 extra_tir_args_list.append(new_arg)
                 return new_arg
-            elif isinstance(arg, (int, float, str, Type, Attrs)) or arg is None:
+            elif isinstance(arg, int | float | str | Type | Attrs) or arg is None:
                 return arg
             raise TypeError(f"not supported type in emit_te: {type(arg)}")
 
@@ -293,8 +293,8 @@ def _convert_te_arg_helper(arg):
         return new_arg
 
     def _get_unbound_tir_vars(
-        args: List[te_Tensor], extra_tir_args: List[PrimExpr]
-    ) -> List[tir.Var]:
+        args: list[te_Tensor], extra_tir_args: list[PrimExpr]
+    ) -> list[tir.Var]:
         """get unbound TIR vars (i.e TIR vars used in the shape but is not
         itself a dimension of a shape)"""
 
@@ -324,18 +324,18 @@ def _populate_used_vars(expr):
         diff = used_vars - bound_vars
         return list(diff)
 
-    def _get_vdevice(arg: Any) -> Optional[VDevice]:
+    def _get_vdevice(arg: Any) -> VDevice | None:
         """get the virtual device from arguments."""
         vdevice = None
         if isinstance(arg, Expr):  # type: ignore
             if isinstance(arg.struct_info, TensorStructInfo):
                 vdevice = arg.struct_info.vdevice
-        elif isinstance(arg, (list, Array, tuple)):
+        elif isinstance(arg, list | Array | tuple):
             for x in arg:
                 vdevice = _get_vdevice(x)
                 if vdevice is not None:
                     return vdevice
-        elif isinstance(arg, (dict, Map)):
+        elif isinstance(arg, dict | Map):
             for k in arg:
                 vdevice = _get_vdevice(arg[k])
                 if vdevice is not None:
@@ -343,7 +343,7 @@ def _get_vdevice(arg: Any) -> Optional[VDevice]:
         return vdevice
 
     def _shape_with_old_tir_var(
-        shape_values: List[tir.PrimExpr], tir_var_inverse_map: Dict[tir.Var, tir.PrimExpr]
+        shape_values: list[tir.PrimExpr], tir_var_inverse_map: dict[tir.Var, tir.PrimExpr]
     ):
         return ShapeExpr(
             [tir.stmt_functor.substitute(value, tir_var_inverse_map) for value in shape_values]
@@ -357,7 +357,7 @@ def _shape_with_old_tir_var(
 
     te_out = func(*te_args, **te_kwargs)
     assert isinstance(te_out, te_Tensor) or (
-        isinstance(te_out, (tuple, list, Array)) and all(isinstance(t, te_Tensor) for t in te_out)
+        isinstance(te_out, tuple | list | Array) and all(isinstance(t, te_Tensor) for t in te_out)
     ), "only support te.tensor or tuple/list/Array of te.tensor as function output"
 
     outs = [te_out] if isinstance(te_out, te_Tensor) else list(te_out)
diff --git a/python/tvm/relax/vm_build.py b/python/tvm/relax/vm_build.py
index 873b801c8d70..ffe92a1daf8c 100644
--- a/python/tvm/relax/vm_build.py
+++ b/python/tvm/relax/vm_build.py
@@ -17,8 +17,6 @@
 # pylint: disable=invalid-name, no-member
 """VM build logics"""
 
-from typing import Dict, List, Optional, Union
-
 import tvm
 from tvm import relax
 from tvm.ir.module import IRModule
@@ -83,8 +81,8 @@ def _vmcodegen(
 
 def _auto_attach_system_lib_prefix(
     tir_mod: tvm.IRModule,
-    target: Optional[tvm.target.Target] = None,
-    system_lib: Optional[bool] = None,
+    target: tvm.target.Target | None = None,
+    system_lib: bool | None = None,
 ):
     """Automatically detect system lib req and attach prefix attr"""
     if target is not None:
@@ -106,13 +104,13 @@ def _is_device_module(mod: tvm.runtime.Module) -> bool:
 
 def _vmlink(
     builder: "relax.ExecBuilder",
-    target: Optional[Union[str, tvm.target.Target]],
-    tir_mod: Optional[tvm.IRModule] = None,
-    tir_pipeline: Optional[Union[str, tvm.transform.Pass]] = "default",
-    ext_libs: Optional[List[tvm.runtime.Module]] = None,
-    params: Optional[Dict[str, list]] = None,
+    target: str | tvm.target.Target | None,
+    tir_mod: tvm.IRModule | None = None,
+    tir_pipeline: str | tvm.transform.Pass | None = "default",
+    ext_libs: list[tvm.runtime.Module] | None = None,
+    params: dict[str, list] | None = None,
     *,
-    system_lib: Optional[bool] = None,
+    system_lib: bool | None = None,
 ):
     """
     Internal codegen function to make executable.
@@ -173,13 +171,13 @@ def _vmlink(
 
 def build(
     mod: tvm.IRModule,
-    target: Optional[Union[str, tvm.target.Target]] = None,
-    params: Optional[Dict[str, list]] = None,
-    relax_pipeline: Union[None, str, tvm.transform.Pass] = "default",
-    tir_pipeline: Union[None, str, tvm.transform.Pass] = "default",
+    target: str | tvm.target.Target | None = None,
+    params: dict[str, list] | None = None,
+    relax_pipeline: None | str | tvm.transform.Pass = "default",
+    tir_pipeline: None | str | tvm.transform.Pass = "default",
     exec_mode: str = "bytecode",
     *,
-    system_lib: Optional[bool] = None,
+    system_lib: bool | None = None,
 ) -> Executable:
     """
     Build an IRModule to VM executable.
@@ -272,7 +270,7 @@ def _extract_attrs(mod: tvm.IRModule):
     )
 
 
-def _filter_tir(mod: tvm.IRModule) -> Optional[tvm.IRModule]:
+def _filter_tir(mod: tvm.IRModule) -> tvm.IRModule | None:
     tir_mod = {gvar: func for gvar, func in mod.functions.items() if isinstance(func, PrimFunc)}
 
     if tir_mod:
diff --git a/python/tvm/rpc/client.py b/python/tvm/rpc/client.py
index 15967db59f4f..dce5861959b1 100644
--- a/python/tvm/rpc/client.py
+++ b/python/tvm/rpc/client.py
@@ -270,7 +270,7 @@ def __init__(self):
 def _popen_session(binary):
     temp = utils.tempdir()
 
-    if isinstance(binary, (bytes, bytearray)):
+    if isinstance(binary, bytes | bytearray):
         path_exec = temp.relpath("server.minrpc")
         with open(path_exec, "wb") as outfile:
             outfile.write(binary)
@@ -542,7 +542,7 @@ def connect(
         if session_timeout:
             key += f" -timeout={session_timeout}"
         session_constructor_args = session_constructor_args if session_constructor_args else []
-        if not isinstance(session_constructor_args, (list, tuple)):
+        if not isinstance(session_constructor_args, list | tuple):
             raise TypeError("Expect the session constructor to be a list or tuple")
         sess = _ffi_api.Connect(url, port, key, enable_logging, *session_constructor_args)
     except NameError:
diff --git a/python/tvm/rpc/server_ios_launcher.py b/python/tvm/rpc/server_ios_launcher.py
index e3cb957ebc6c..04c12409c110 100644
--- a/python/tvm/rpc/server_ios_launcher.py
+++ b/python/tvm/rpc/server_ios_launcher.py
@@ -27,7 +27,7 @@
 import threading
 import time
 from enum import Enum
-from typing import AnyStr, Dict, List
+from typing import AnyStr
 
 
 class OSName(Enum):
@@ -54,7 +54,7 @@ class RPCServerMode(Enum):
     tracker = "tracker"
 
 
-def get_list_of_available_simulators() -> Dict[AnyStr, List]:
+def get_list_of_available_simulators() -> dict[AnyStr, list]:
     """
     List of simulators available on the system. Simulators are presented as a dictionary.
     The dictionary key is the name of the operating system of the simulator.
@@ -75,10 +75,10 @@ def get_list_of_available_simulators() -> Dict[AnyStr, List]:
     return available_simulators
 
 
-def grep_by_system(available_devices: Dict[AnyStr, List], system_name: OSName) -> List[Dict]:
+def grep_by_system(available_devices: dict[AnyStr, list], system_name: OSName) -> list[dict]:
     """Search for simulators that use the target operating system."""
 
-    def find_index_of_substr(search_field: List[AnyStr], target: AnyStr) -> int:
+    def find_index_of_substr(search_field: list[AnyStr], target: AnyStr) -> int:
         for i, item in enumerate(search_field):
             if target in item:
                 return i
@@ -89,13 +89,13 @@ def find_index_of_substr(search_field: List[AnyStr], target: AnyStr) -> int:
     return available_devices[keys[find_index_of_substr(keys, system_name.value)]]
 
 
-def grep_by_device(available_devices: List[Dict], device_name: IOSDevice) -> List[Dict]:
+def grep_by_device(available_devices: list[dict], device_name: IOSDevice) -> list[dict]:
     """Search for simulators that emulate a given device."""
 
     return [item for item in available_devices if device_name.value in item["name"]]
 
 
-def get_device_uid(target_device: Dict) -> AnyStr:
+def get_device_uid(target_device: dict) -> AnyStr:
     """Get a unique device ID."""
 
     return target_device["udid"]
@@ -217,7 +217,7 @@ def is_turned_off(udid: AnyStr) -> bool:
     return device["state"] == "Shutdown"
 
 
-def check_booted_device(devices: List[Dict]) -> Dict:
+def check_booted_device(devices: list[dict]) -> dict:
     """Check if there is already a booted device. If so, return this device."""
 
     for device in devices:
@@ -226,7 +226,7 @@ def check_booted_device(devices: List[Dict]) -> Dict:
     return {}
 
 
-def find_device(udid: AnyStr) -> Dict:
+def find_device(udid: AnyStr) -> dict:
     """Find device by its unique ID."""
 
     return_value = {}
diff --git a/python/tvm/runtime/_tensor.py b/python/tvm/runtime/_tensor.py
index 13557f9bd97a..7564ea8667d6 100644
--- a/python/tvm/runtime/_tensor.py
+++ b/python/tvm/runtime/_tensor.py
@@ -86,7 +86,7 @@ def __setitem__(self, in_slice, value):
         if isinstance(value, Tensor):
             if not value.same_as(self):
                 value.copyto(self)
-        elif isinstance(value, (np.ndarray, np.generic)):
+        elif isinstance(value, np.ndarray | np.generic):
             self.copyfrom(value)
         else:
             raise TypeError(f"type {type(value)} not supported")
@@ -249,7 +249,7 @@ def _copyto(self, target_nd):
         _ffi_api.TVMTensorCopyFromTo(self, target_nd)
         return target_nd
 
-    def _create_view(self, shape, dtype: Optional[str] = None, relative_byte_offset: int = 0):
+    def _create_view(self, shape, dtype: str | None = None, relative_byte_offset: int = 0):
         """Create a view into an existing array.
 
         The view shares the same allocation and datatype as the
@@ -348,7 +348,7 @@ def tensor(arr, device=None, mem_scope=None):
     """
     device = device or cpu()
 
-    if not isinstance(arr, (np.ndarray, Tensor)):
+    if not isinstance(arr, np.ndarray | Tensor):
         arr = np.array(arr)
     return empty(arr.shape, arr.dtype, device, mem_scope).copyfrom(arr)
 
diff --git a/python/tvm/runtime/disco/session.py b/python/tvm/runtime/disco/session.py
index 6200608c638b..b7fdec4e9895 100644
--- a/python/tvm/runtime/disco/session.py
+++ b/python/tvm/runtime/disco/session.py
@@ -22,7 +22,8 @@
 import logging
 import os
 import pickle
-from typing import Any, Callable, Optional, Sequence, Union
+from collections.abc import Callable, Sequence
+from typing import Any, Optional, Union
 
 import numpy as np
 from tvm_ffi import get_global_func, register_global_func, register_object
@@ -59,7 +60,7 @@ def debug_get_from_remote(self, worker_id: int) -> Any:
     def debug_copy_from(
         self,
         worker_id: int,
-        value: Union[np.ndarray, Tensor],
+        value: np.ndarray | Tensor,
     ) -> None:
         """Copy an Tensor value to remote for debugging purposes.
 
@@ -119,7 +120,7 @@ def empty(
         self,
         shape: Sequence[int],
         dtype: str,
-        device: Optional[Device] = None,
+        device: Device | None = None,
         worker0_only: bool = False,
         in_group: bool = True,
     ) -> DRef:
@@ -260,7 +261,7 @@ def copy_from_worker_0(self, host_array: Tensor, remote_array: DRef) -> None:
         """
         return _ffi_api.SessionCopyFromWorker0(self, host_array, remote_array)  # type: ignore # pylint: disable=no-member
 
-    def copy_to_worker_0(self, host_array: Tensor, remote_array: Optional[DRef] = None) -> DRef:
+    def copy_to_worker_0(self, host_array: Tensor, remote_array: DRef | None = None) -> DRef:
         """Copy the controller-side Tensor to worker-0.
 
         Parameters
@@ -290,7 +291,7 @@ def copy_to_worker_0(self, host_array: Tensor, remote_array: Optional[DRef] = No
     def load_vm_module(
         self,
         path: str,
-        device: Optional[Device] = None,
+        device: Device | None = None,
     ) -> DModule:
         """Load a VM module from a file.
 
@@ -330,8 +331,8 @@ def init_ccl(self, ccl: str, *device_ids):
 
     def broadcast(
         self,
-        src: Union[np.ndarray, Tensor],
-        dst: Optional[DRef] = None,
+        src: np.ndarray | Tensor,
+        dst: DRef | None = None,
         in_group: bool = True,
     ) -> DRef:
         """Broadcast an array to all workers
@@ -388,8 +389,8 @@ def broadcast_from_worker0(self, src: DRef, dst: DRef, in_group: bool = True) ->
 
     def scatter(
         self,
-        src: Union[np.ndarray, Tensor],
-        dst: Optional[DRef] = None,
+        src: np.ndarray | Tensor,
+        dst: DRef | None = None,
         in_group: bool = True,
     ) -> DRef:
         """Scatter an array across all workers
diff --git a/python/tvm/runtime/executable.py b/python/tvm/runtime/executable.py
index 19be4d529866..212896ccb2b7 100644
--- a/python/tvm/runtime/executable.py
+++ b/python/tvm/runtime/executable.py
@@ -18,7 +18,8 @@
 
 """Executable object for TVM Runtime"""
 
-from typing import Any, Callable, Dict, List, Optional
+from collections.abc import Callable
+from typing import Any
 
 import tvm
 from tvm.contrib import utils as _utils
@@ -32,7 +33,7 @@ class Executable:
     def __init__(self, mod: Module):
         """Initialize the Executable object."""
         self.mod: Module = mod
-        self._jitted_mod: Optional[Module] = None
+        self._jitted_mod: Module | None = None
 
     def __getitem__(self, name: str) -> PackedFunc:
         """Get the PackedFunc from the jitted module."""
@@ -45,8 +46,8 @@ def __call__(self, *args, **kwargs) -> Any:
     def jit(
         self,
         *,
-        fcompile: Optional[Callable[[str, List[str], Dict[str, Any]], None]] = None,
-        addons: Optional[List[str]] = None,
+        fcompile: Callable[[str, list[str], dict[str, Any]], None] | None = None,
+        addons: list[str] | None = None,
         force_recompile: bool = False,
         **kwargs,
     ) -> Module:
diff --git a/python/tvm/runtime/module.py b/python/tvm/runtime/module.py
index 09f48ad749c4..7586e0df1576 100644
--- a/python/tvm/runtime/module.py
+++ b/python/tvm/runtime/module.py
@@ -22,7 +22,7 @@
 import json
 import os
 import struct
-from typing import Sequence
+from collections.abc import Sequence
 
 import numpy as np
 from tvm_ffi import (
@@ -310,7 +310,7 @@ def get_source_format_from_module(module):
             options = []
             if "options" in kwargs:
                 opts = kwargs["options"]
-                options = opts if isinstance(opts, (list, tuple)) else [opts]
+                options = opts if isinstance(opts, list | tuple) else [opts]
             opts = options + ["-I" + path for path in find_include_path()]
             kwargs.update({"options": opts})
 
diff --git a/python/tvm/runtime/params.py b/python/tvm/runtime/params.py
index 11a6e85a43bc..1ea1c64262be 100644
--- a/python/tvm/runtime/params.py
+++ b/python/tvm/runtime/params.py
@@ -90,7 +90,7 @@ def load_param_dict(param_bytes):
     params : dict of str to Tensor
         The parameter dictionary.
     """
-    if isinstance(param_bytes, (bytes, str)):
+    if isinstance(param_bytes, bytes | str):
         param_bytes = bytearray(param_bytes)
     return _ffi_api.LoadParams(param_bytes)
 
diff --git a/python/tvm/runtime/profiling/__init__.py b/python/tvm/runtime/profiling/__init__.py
index fad96d992aa0..7ebc5e5c408d 100644
--- a/python/tvm/runtime/profiling/__init__.py
+++ b/python/tvm/runtime/profiling/__init__.py
@@ -17,7 +17,8 @@
 # under the License.
 """Registration of profiling objects in python."""
 
-from typing import Dict, Sequence, Optional
+from typing import Optional
+from collections.abc import Sequence
 from ... import ffi as _ffi
 from . import _ffi_api
 from .. import Object, Device
@@ -38,9 +39,9 @@ class Report(Object):
 
     def __init__(
         self,
-        calls: Sequence[Dict[str, Object]],
-        device_metrics: Dict[str, Dict[str, Object]],
-        configuration: Dict[str, Object],
+        calls: Sequence[dict[str, Object]],
+        device_metrics: dict[str, dict[str, Object]],
+        configuration: dict[str, Object],
     ):
         """Construct a profiling report from a list of metrics and per-device metrics.
 
diff --git a/python/tvm/runtime/script_printer.py b/python/tvm/runtime/script_printer.py
index 3efe6ac8da9e..431fdbee533c 100644
--- a/python/tvm/runtime/script_printer.py
+++ b/python/tvm/runtime/script_printer.py
@@ -17,7 +17,7 @@
 """Configuration of TVMScript printer"""
 
 import os
-from typing import Dict, List, Optional, Sequence
+from collections.abc import Sequence
 
 from tvm_ffi import get_global_func, register_object
 from tvm_ffi.access_path import AccessPath
@@ -47,15 +47,15 @@ class PrinterConfig(Object):
     syntax_sugar: bool
     show_object_address: bool
     show_all_struct_info: bool
-    path_to_underline: Optional[List[AccessPath]]
-    path_to_annotate: Optional[Dict[AccessPath, str]]
-    obj_to_underline: Optional[List[AccessPath]]
-    obj_to_annotate: Optional[Dict[AccessPath, str]]
+    path_to_underline: list[AccessPath] | None
+    path_to_annotate: dict[AccessPath, str] | None
+    obj_to_underline: list[AccessPath] | None
+    obj_to_annotate: dict[AccessPath, str] | None
 
     def __init__(
         self,
         *,
-        name: Optional[str] = None,
+        name: str | None = None,
         show_meta: bool = False,
         ir_prefix: str = "I",
         tir_prefix: str = "T",
@@ -67,14 +67,14 @@ def __init__(
         verbose_expr: bool = False,
         indent_spaces: int = 4,
         print_line_numbers: bool = False,
-        num_context_lines: Optional[int] = None,
+        num_context_lines: int | None = None,
         syntax_sugar: bool = True,
         show_object_address: bool = False,
         show_all_struct_info: bool = True,
-        path_to_underline: Optional[List[AccessPath]] = None,
-        path_to_annotate: Optional[Dict[AccessPath, str]] = None,
-        obj_to_underline: Optional[List[Object]] = None,
-        obj_to_annotate: Optional[Dict[Object, str]] = None,
+        path_to_underline: list[AccessPath] | None = None,
+        path_to_annotate: dict[AccessPath, str] | None = None,
+        obj_to_underline: list[Object] | None = None,
+        obj_to_annotate: dict[Object, str] | None = None,
     ) -> None:
         if num_context_lines is None:
             num_context_lines = -1
@@ -123,7 +123,7 @@ class Scriptable:
     def script(
         self,
         *,
-        name: Optional[str] = None,
+        name: str | None = None,
         show_meta: bool = False,
         ir_prefix: str = "I",
         tir_prefix: str = "T",
@@ -139,10 +139,10 @@ def script(
         syntax_sugar: bool = True,
         show_object_address: bool = False,
         show_all_struct_info: bool = True,
-        path_to_underline: Optional[List[AccessPath]] = None,
-        path_to_annotate: Optional[Dict[AccessPath, str]] = None,
-        obj_to_underline: Optional[List[Object]] = None,
-        obj_to_annotate: Optional[Dict[Object, str]] = None,
+        path_to_underline: list[AccessPath] | None = None,
+        path_to_annotate: dict[AccessPath, str] | None = None,
+        obj_to_underline: list[Object] | None = None,
+        obj_to_annotate: dict[Object, str] | None = None,
     ) -> str:
         """Print TVM IR into TVMScript text format
 
@@ -227,7 +227,7 @@ def script(
     def _relax_script(
         self,
         *,
-        name: Optional[str] = None,
+        name: str | None = None,
         show_meta: bool = False,
         ir_prefix: str = "I",
         tir_prefix: str = "T",
@@ -242,10 +242,10 @@ def _relax_script(
         num_context_lines: int = -1,
         syntax_sugar: bool = True,
         show_object_address: bool = False,
-        path_to_underline: Optional[List[AccessPath]] = None,
-        path_to_annotate: Optional[Dict[AccessPath, str]] = None,
-        obj_to_underline: Optional[List[Object]] = None,
-        obj_to_annotate: Optional[Dict[Object, str]] = None,
+        path_to_underline: list[AccessPath] | None = None,
+        path_to_annotate: dict[AccessPath, str] | None = None,
+        obj_to_underline: list[Object] | None = None,
+        obj_to_annotate: dict[Object, str] | None = None,
     ) -> str:
         return _relax_script(
             self,
@@ -274,10 +274,10 @@ def _relax_script(
 
     def show(
         self,
-        style: Optional[str] = None,
-        black_format: Optional[bool] = None,
+        style: str | None = None,
+        black_format: bool | None = None,
         *,
-        name: Optional[str] = None,
+        name: str | None = None,
         show_meta: bool = False,
         ir_prefix: str = "I",
         tir_prefix: str = "T",
@@ -293,10 +293,10 @@ def show(
         syntax_sugar: bool = True,
         show_object_address: bool = False,
         show_all_struct_info: bool = True,
-        path_to_underline: Optional[List[AccessPath]] = None,
-        path_to_annotate: Optional[Dict[AccessPath, str]] = None,
-        obj_to_underline: Optional[List[Object]] = None,
-        obj_to_annotate: Optional[Dict[Object, str]] = None,
+        path_to_underline: list[AccessPath] | None = None,
+        path_to_annotate: dict[AccessPath, str] | None = None,
+        obj_to_underline: list[Object] | None = None,
+        obj_to_annotate: dict[Object, str] | None = None,
     ) -> None:
         """A sugar for print highlighted TVM script.
 
diff --git a/python/tvm/runtime/support.py b/python/tvm/runtime/support.py
index de3d215dd1cf..d40f5c263618 100644
--- a/python/tvm/runtime/support.py
+++ b/python/tvm/runtime/support.py
@@ -199,6 +199,6 @@ def __setattr__(self, name, value):
     TVMDerivedObject.__doc__ = cls.__doc__
     TVMDerivedObject.__module__ = cls.__module__
     for key, value in cls.__dict__.items():
-        if isinstance(value, (classmethod, staticmethod)):
+        if isinstance(value, classmethod | staticmethod):
             setattr(TVMDerivedObject, key, value)
     return TVMDerivedObject
diff --git a/python/tvm/runtime/vm.py b/python/tvm/runtime/vm.py
index eb22870989c2..921739203e16 100644
--- a/python/tvm/runtime/vm.py
+++ b/python/tvm/runtime/vm.py
@@ -18,9 +18,10 @@
 # ruff: noqa: RUF005
 """The Relax virtual machine."""
 
+from collections.abc import Callable
 from enum import IntEnum
 from numbers import Integral, Number
-from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+from typing import Any
 
 import numpy as np  # type: ignore
 from tvm_ffi import register_global_func
@@ -46,9 +47,9 @@ class VirtualMachine:
 
     def __init__(
         self,
-        rt_mod: Union[tvm.runtime.Module, tvm.runtime.Executable],
-        device: Union[Device, List[Device]],
-        memory_cfg: Optional[Union[str, Dict[Device, str]]] = None,
+        rt_mod: tvm.runtime.Module | tvm.runtime.Executable,
+        device: Device | list[Device],
+        memory_cfg: str | dict[Device, str] | None = None,
         profile: bool = False,
     ) -> None:
         """
@@ -92,10 +93,10 @@ def __init__(
         self._set_instrument = self.module["set_instrument"]
         self._setup_device(device, memory_cfg)
 
-    def _setup_device(self, dev: Device, memory_cfg: Union[str, Dict[Device, str]]) -> None:
+    def _setup_device(self, dev: Device, memory_cfg: str | dict[Device, str]) -> None:
         """init devices and allocators."""
         devs = dev
-        if not isinstance(dev, (list, tuple)):
+        if not isinstance(dev, list | tuple):
             if not isinstance(dev, tvm.runtime.Device):
                 raise TypeError("dev is expected to be Device or List[Device]")
             devs = [dev]
@@ -150,9 +151,9 @@ def save_function(
         self,
         func_name: str,
         saved_name: str,
-        *args: List[Any],
+        *args: list[Any],
         include_return: bool = True,
-        **kwargs: Dict[str, Any],
+        **kwargs: dict[str, Any],
     ) -> None:
         """
         Convenience function. Takes a function from the module and saves
@@ -187,20 +188,20 @@ def save_function(
         kwargs : Dict[str, Any]
             Any named arguments to package up with the function
         """
-        cargs: List[Any] = []
+        cargs: list[Any] = []
         if kwargs:
             args = self._convert_func_named_args(func_name, args, **kwargs)
         for arg in args:
             self._convert(arg, cargs)
         self._save_function(func_name, saved_name, int(include_return), *cargs)
 
-    def _convert(self, arg: Any, cargs: List) -> None:
+    def _convert(self, arg: Any, cargs: list) -> None:
         """helper function to convert arguments to vm function."""
 
         def _gettype(arg):
             if isinstance(arg, np.float16):
                 return "float16"
-            elif isinstance(arg, (Integral, bool)):
+            elif isinstance(arg, Integral | bool):
                 return "int32"
             else:
                 return "float32"
@@ -212,12 +213,12 @@ def _gettype(arg):
             cargs.append(nd_arr)
         elif isinstance(arg, tvm.runtime.Tensor):
             cargs.append(arg)
-        elif isinstance(arg, (tuple, list)):
-            field_args: List[Any] = []
+        elif isinstance(arg, tuple | list):
+            field_args: list[Any] = []
             for field in arg:
                 self._convert(field, field_args)
             cargs.append(tuple(field_args))
-        elif isinstance(arg, (Number, bool)):
+        elif isinstance(arg, Number | bool):
             dtype = _gettype(arg)
             value = tvm.runtime.tensor(np.array(arg, dtype=dtype), device=tvm.cpu(0))
             cargs.append(value)
@@ -270,7 +271,7 @@ def set_input(self, func_name: str, *args: Any, **kwargs: Any) -> None:
         kwargs: dict of str to tvm.runtime.Tensor or np.ndarray
             Named arguments to the function.
         """
-        cargs: List[Any] = []
+        cargs: list[Any] = []
 
         if kwargs:
             args = self._convert_func_named_args(func_name, args, **kwargs)
@@ -296,7 +297,7 @@ def invoke_stateful(self, func_name: str) -> None:
         """
         self._invoke_stateful(func_name)
 
-    def get_outputs(self, func_name: str) -> Union[tvm.Object, Tuple[Any]]:
+    def get_outputs(self, func_name: str) -> tvm.Object | tuple[Any]:
         """
         Get the value output by the function by the given name
         after a call of `invoke_stateful`.
@@ -492,7 +493,7 @@ def profile(self, func_name: str, *args):
         report: tvm.runtime.profiling.Report
             The formatted profiling result, showing per-op timing measurements.
         """
-        cargs: List[Any] = []
+        cargs: list[Any] = []
 
         for arg in args:
             self._convert(arg, cargs)
diff --git a/python/tvm/s_tir/analysis/__init__.py b/python/tvm/s_tir/analysis/__init__.py
index 821d3ba04f4a..6956b515e8e6 100644
--- a/python/tvm/s_tir/analysis/__init__.py
+++ b/python/tvm/s_tir/analysis/__init__.py
@@ -18,7 +18,7 @@
 """Analysis utilities for Schedulable TensorIR (S-TIR)."""
 
 # pylint: disable=invalid-name
-from typing import Dict, List, Optional, Union
+from typing import Optional, Union
 
 import tvm
 from tvm.ir import IRModule
@@ -31,8 +31,8 @@
 
 
 def get_sblock_access_region(
-    block: SBlock, buffer_var_map: Dict[Var, Buffer]
-) -> List[List[BufferRegion]]:
+    block: SBlock, buffer_var_map: dict[Var, Buffer]
+) -> list[list[BufferRegion]]:
     """Detect which regions of tensors in this block are read or written to.
        Regions are sorted by order of appearance in the AST.
 
@@ -56,8 +56,8 @@ def get_sblock_access_region(
 
 
 def get_sblock_read_write_region(
-    block: SBlock, buffer_var_map: Dict[Var, Buffer]
-) -> List[List[BufferRegion]]:
+    block: SBlock, buffer_var_map: dict[Var, Buffer]
+) -> list[list[BufferRegion]]:
     """Auto detect the block read/write region according to its body stmt.
        An opaque access will be counted as both a read and a write access
 
@@ -77,7 +77,7 @@ def get_sblock_read_write_region(
     return _ffi_api.GetSBlockReadWriteRegion(block, buffer_var_map)  # type: ignore
 
 
-def detect_buffer_access_lca(func: PrimFunc) -> Dict[Buffer, Stmt]:
+def detect_buffer_access_lca(func: PrimFunc) -> dict[Buffer, Stmt]:
     """Detect the lowest common ancestor(LCA) of buffer access, including both high-level
     access (BufferLoad, BufferStore) and low-level access (BufferLoad, BufferStore and opaque
     access).
@@ -96,7 +96,7 @@ def detect_buffer_access_lca(func: PrimFunc) -> Dict[Buffer, Stmt]:
     return _ffi_api.detect_buffer_access_lca(func)  # type: ignore # pylint: disable=no-member
 
 
-def find_anchor_sblock(mod: IRModule) -> Optional[SBlock]:
+def find_anchor_sblock(mod: IRModule) -> SBlock | None:
     """Find the "anchor block" of the given module.
 
     We define the anchor block to be the block with (1) an init statement and (2) having
@@ -123,7 +123,7 @@ def find_anchor_sblock(mod: IRModule) -> Optional[SBlock]:
     return _ffi_api.find_anchor_sblock(mod)  # type: ignore # pylint: disable=no-member
 
 
-def verify_gpu_code(func: PrimFunc, constraints: Dict[str, int]) -> bool:
+def verify_gpu_code(func: PrimFunc, constraints: dict[str, int]) -> bool:
     """Verify if module contains illegal host side direct memory access.
 
     Parameters
@@ -143,8 +143,8 @@ def verify_gpu_code(func: PrimFunc, constraints: Dict[str, int]) -> bool:
 
 
 def calculate_allocated_bytes(
-    func_or_mod: Union[PrimFunc, IRModule],
-) -> Dict[str, Dict[str, int]]:
+    func_or_mod: PrimFunc | IRModule,
+) -> dict[str, dict[str, int]]:
     """Calculate allocated memory per memory scope required by TIR PrimFuncs.
 
     Parameters
@@ -160,14 +160,14 @@ def calculate_allocated_bytes(
         dict with function names as keys and a dict of allocated sizes as values. If a single
         PrimFunc is passed, the function name is returned as "main"
     """
-    if not isinstance(func_or_mod, (PrimFunc, IRModule)):
+    if not isinstance(func_or_mod, PrimFunc | IRModule):
         raise TypeError(
             f"Expected argument to be PrimFunc or IRModule, but received {type(func_or_mod)}"
         )
     return _ffi_api.calculate_allocated_bytes(func_or_mod)  # type: ignore
 
 
-def estimate_tir_flops(stmt_or_mod: Union[Stmt, IRModule]) -> float:
+def estimate_tir_flops(stmt_or_mod: Stmt | IRModule) -> float:
     """Estimate the FLOPs of a TIR fragment.
 
     Parameters
@@ -194,7 +194,7 @@ def OOBChecker():
     return _ffi_api.OOBChecker()  # type: ignore
 
 
-def get_vtcm_compaction_passes() -> List[tvm.transform.Pass]:
+def get_vtcm_compaction_passes() -> list[tvm.transform.Pass]:
     """Utility function to get the list of lowering passes to be applied to calculate the compacted
     VTCM allocation size
 
diff --git a/python/tvm/s_tir/data_layout.py b/python/tvm/s_tir/data_layout.py
index 9004a1a85ffd..01dd88e3b018 100644
--- a/python/tvm/s_tir/data_layout.py
+++ b/python/tvm/s_tir/data_layout.py
@@ -16,8 +16,6 @@
 # under the License.
 """Data layout."""
 
-from typing import Union
-
 import tvm_ffi
 
 from tvm.runtime import Object
@@ -191,9 +189,7 @@ def layout(layout_str: str, dtype: str = "int32") -> Layout:
     return _ffi_api.Layout(layout_str, dtype)  # type: ignore
 
 
-def bijective_layout(
-    src_layout: Union[str, Layout], dst_layout: Union[str, Layout]
-) -> BijectiveLayout:
+def bijective_layout(src_layout: str | Layout, dst_layout: str | Layout) -> BijectiveLayout:
     """Create a bijective layout mapping.
 
     Parameters
diff --git a/python/tvm/s_tir/dlight/adreno/convolution.py b/python/tvm/s_tir/dlight/adreno/convolution.py
index 5c99f3706a80..7a59c62b1a65 100644
--- a/python/tvm/s_tir/dlight/adreno/convolution.py
+++ b/python/tvm/s_tir/dlight/adreno/convolution.py
@@ -17,8 +17,6 @@
 # pylint: disable=missing-docstring, invalid-name
 """A Conv2d schedule rule for Adreno GPU operators."""
 
-from typing import Optional, Union
-
 from tvm import s_tir, tir
 from tvm.target import Target
 
@@ -64,11 +62,11 @@ def schedule_conv2d(sch: s_tir.Schedule, blk: s_tir.schedule.SBlockRV):
 
     def apply(  # pylint: disable=too-many-locals,missing-docstring
         self,
-        func: Union[tir.PrimFunc, s_tir.Schedule],
+        func: tir.PrimFunc | s_tir.Schedule,
         target: Target,
         _: bool,
-    ) -> Optional[s_tir.Schedule]:
-        if not (isinstance(func, (tir.PrimFunc, s_tir.Schedule))) or not self.is_target_available(
+    ) -> s_tir.Schedule | None:
+        if not (isinstance(func, tir.PrimFunc | s_tir.Schedule)) or not self.is_target_available(
             target
         ):
             return None
diff --git a/python/tvm/s_tir/dlight/adreno/fallback.py b/python/tvm/s_tir/dlight/adreno/fallback.py
index a485fd5a2257..f628583194c8 100644
--- a/python/tvm/s_tir/dlight/adreno/fallback.py
+++ b/python/tvm/s_tir/dlight/adreno/fallback.py
@@ -33,8 +33,6 @@
 # under the License.
 """Dlight Adreno Fallback Schedules"""
 
-from typing import List, Union
-
 from tvm import s_tir, tir
 from tvm.target import Target
 
@@ -64,8 +62,8 @@ class Fallback(AdrenoScheduleRule):
 
     @staticmethod
     def schedule_inline_blocks(
-        sch: s_tir.Schedule, blocks: List[s_tir.schedule.SBlockRV]
-    ) -> List[s_tir.schedule.SBlockRV]:
+        sch: s_tir.Schedule, blocks: list[s_tir.schedule.SBlockRV]
+    ) -> list[s_tir.schedule.SBlockRV]:
         """
         Auto Inlines Injective and Element-wise Operations while trying to omit data pad blocks...
         """
@@ -173,7 +171,7 @@ def apply(  # pylint: disable=too-many-locals
         func: tir.PrimFunc,
         target: Target,
         _: bool,
-    ) -> Union[None, s_tir.Schedule, List[s_tir.Schedule]]:
+    ) -> None | s_tir.Schedule | list[s_tir.Schedule]:
         # pylint: disable=invalid-name
 
         if not isinstance(func, tir.PrimFunc) or not self.is_target_available(target):
diff --git a/python/tvm/s_tir/dlight/adreno/layout_transform.py b/python/tvm/s_tir/dlight/adreno/layout_transform.py
index daf749eb793c..aebd21aa38e6 100644
--- a/python/tvm/s_tir/dlight/adreno/layout_transform.py
+++ b/python/tvm/s_tir/dlight/adreno/layout_transform.py
@@ -37,8 +37,6 @@
 
 "Schedules for Texture Based Layout Transforms"
 
-from typing import List, Union
-
 from tvm import s_tir, tir
 from tvm.target import Target
 
@@ -55,12 +53,12 @@ def __init__(self, use_op_name=True):
     # TODO: Try using Coalesced Writes...
     def apply(  # pylint: disable=too-many-locals
         self,
-        func: Union[tir.PrimFunc, s_tir.Schedule],
+        func: tir.PrimFunc | s_tir.Schedule,
         target: Target,
         _: bool,
-    ) -> Union[None, s_tir.Schedule, List[s_tir.Schedule]]:
+    ) -> None | s_tir.Schedule | list[s_tir.Schedule]:
         # pylint: disable=invalid-name
-        if not (isinstance(func, (tir.PrimFunc, s_tir.Schedule))) or not self.is_target_available(
+        if not (isinstance(func, tir.PrimFunc | s_tir.Schedule)) or not self.is_target_available(
             target
         ):
             return None
diff --git a/python/tvm/s_tir/dlight/adreno/utils.py b/python/tvm/s_tir/dlight/adreno/utils.py
index 0e6bca34d79d..a820cf2b3151 100644
--- a/python/tvm/s_tir/dlight/adreno/utils.py
+++ b/python/tvm/s_tir/dlight/adreno/utils.py
@@ -17,7 +17,6 @@
 """Utilis for Adreno operators."""
 
 # pylint: disable=import-outside-toplevel, unused-argument, invalid-name, missing-function-docstring
-from typing import List, Optional
 
 from tvm import s_tir
 from tvm.target import Target
@@ -68,14 +67,14 @@ def get_texture_storage(block_info: SBlockInfo):
 
 
 def schedule_inline_blocks(
-    sch: s_tir.Schedule, blocks: Optional[List[s_tir.schedule.SBlockRV]] = None
+    sch: s_tir.Schedule, blocks: list[s_tir.schedule.SBlockRV] | None = None
 ):
     from .fallback import Fallback
 
     return Fallback.schedule_inline_blocks(sch, blocks)
 
 
-def schedule_default(sch, blocks: Optional[List[s_tir.schedule.SBlockRV]] = None):
+def schedule_default(sch, blocks: list[s_tir.schedule.SBlockRV] | None = None):
     from .fallback import Fallback
 
     ret = []
diff --git a/python/tvm/s_tir/dlight/analysis/common_analysis.py b/python/tvm/s_tir/dlight/analysis/common_analysis.py
index 3a148361092e..9a8f6a20dccf 100644
--- a/python/tvm/s_tir/dlight/analysis/common_analysis.py
+++ b/python/tvm/s_tir/dlight/analysis/common_analysis.py
@@ -20,10 +20,9 @@
 """Analysis on TIR blocks, loops and functions."""
 
 from collections import namedtuple
-from typing import List, Optional, Set, Tuple, Union
+from typing import Literal
 
 from tvm_ffi import get_global_func
-from typing_extensions import Literal
 
 from tvm import ir, s_tir, tir
 from tvm.runtime import DataType
@@ -54,7 +53,7 @@ def __init__(
         self.loop_rv = loop_rv
 
     @property
-    def dom(self) -> Union[int, tir.PrimExpr]:
+    def dom(self) -> int | tir.PrimExpr:
         """The iteration domain of the loop."""
         return int(self._dom) if isinstance(self._dom, tir.IntImm) else self._dom
 
@@ -71,23 +70,23 @@ def __repr__(self) -> str:
 RemIndex = namedtuple("RemIndex", ["sub", "div"])  # c%len
 DivIndex = namedtuple("DivIndex", ["sub", "div"])  # c//len
 MergeIndex = namedtuple("MulIndex", ["dom", "mul", "sub"])  # co*len + cb
-BufIndex = List[Union[Index, RemIndex, DivIndex, MergeIndex, None]]
+BufIndex = list[Index | RemIndex | DivIndex | MergeIndex | None]
 
 
 class BufferInfo:
     "Information about Buffer. Provides useful analysis"
 
     buf_region: tir.BufferRegion
-    shape: Tuple[int]
-    assoc_lps: List[Union[s_tir.schedule.LoopRV, None]]
-    assoc_lps_info: List[Union[tir.For, None]]
+    shape: tuple[int]
+    assoc_lps: list[s_tir.schedule.LoopRV | None]
+    assoc_lps_info: list[tir.For | None]
 
     def __init__(
         self,
         sch: s_tir.Schedule,
         block_rv: s_tir.schedule.SBlockRV,
         buf_region: tir.BufferRegion,
-        lps: Union[List[s_tir.schedule.LoopRV], None],
+        lps: list[s_tir.schedule.LoopRV] | None,
     ):
         block = sch.get(block_rv)
         if lps is None:
@@ -170,14 +169,14 @@ class SBlockInfo:
     """Information about a TIR block."""
 
     name: str
-    iters: List[IterInfo]
+    iters: list[IterInfo]
     block_rv: s_tir.schedule.SBlockRV
     _reduction_block: bool
 
     def __init__(
         self,
         name: str,
-        iters: List[IterInfo],
+        iters: list[IterInfo],
         block_rv: s_tir.schedule.SBlockRV,
         reduction_block: bool = False,
     ):
@@ -187,16 +186,16 @@ def __init__(
         self.iters = iters
         self._reduction_block = reduction_block
 
-    def dom(self) -> List[Union[int, tir.PrimExpr]]:
+    def dom(self) -> list[int | tir.PrimExpr]:
         """The iteration domain of the block."""
         return [i.dom for i in self.iters]
 
-    def read_bufs(self, sch: s_tir.Schedule) -> List[BufferInfo]:
+    def read_bufs(self, sch: s_tir.Schedule) -> list[BufferInfo]:
         block_stmt = sch.get(self.block_rv)
         lps = sch.get_loops(self.block_rv)
         return [BufferInfo(sch, self.block_rv, buf, lps) for buf in block_stmt.reads]
 
-    def write_bufs(self, sch: s_tir.Schedule) -> List[BufferInfo]:
+    def write_bufs(self, sch: s_tir.Schedule) -> list[BufferInfo]:
         block_stmt = sch.get(self.block_rv)
         lps = sch.get_loops(self.block_rv)
         return [BufferInfo(sch, self.block_rv, buf, lps) for buf in block_stmt.writes]
@@ -229,7 +228,7 @@ def _check_unit_var_range(dom: ir.Range, var: tir.Var) -> bool:
                 return False
         return True
 
-    def get_loops(self) -> List[s_tir.schedule.LoopRV]:
+    def get_loops(self) -> list[s_tir.schedule.LoopRV]:
         return [iter_info.loop_rv for iter_info in self.iters]
 
     def is_reduction(self) -> bool:
@@ -285,7 +284,7 @@ def __repr__(self) -> str:
 _normalize_prim_func = get_global_func("s_tir.schedule.NormalizePrimFunc")
 
 
-def normalize_prim_func(sch: s_tir.Schedule) -> Optional[List[SBlockInfo]]:
+def normalize_prim_func(sch: s_tir.Schedule) -> list[SBlockInfo] | None:
     """Normalize the primfunc to normal form"""
     try:
         result = _normalize_prim_func(sch)
@@ -300,7 +299,7 @@ def _iter_kind(i: tir.IterVar) -> str:
             tir.IterVar.CommReduce: "R",
         }.get(i.iter_type, "O")
 
-    blocks: List[SBlockInfo] = []
+    blocks: list[SBlockInfo] = []
     for block, loops, iters, is_reduction in zip(*result):
         blocks.append(
             SBlockInfo(
@@ -385,8 +384,8 @@ def get_root_block(sch: Schedule, func_name: str = "main") -> SBlockRV:
 
 
 def collect_block_iter_vars_used_in_access_region(
-    block: tir.SBlock, region: List[ir.Range]
-) -> Set[tir.Var]:
+    block: tir.SBlock, region: list[ir.Range]
+) -> set[tir.Var]:
     """Collect the block iter variables used in the access region of a buffer region."""
     tir_vars = set()
     for expr in region:
@@ -396,7 +395,7 @@ def collect_block_iter_vars_used_in_access_region(
     return tir_vars
 
 
-def collect_vars_used_in_prim_expr(expr: tir.PrimExpr) -> Set[tir.Var]:
+def collect_vars_used_in_prim_expr(expr: tir.PrimExpr) -> set[tir.Var]:
     """Collect the variables used in the PrimExpr."""
     tir_vars = set()
 
diff --git a/python/tvm/s_tir/dlight/analysis/gemv.py b/python/tvm/s_tir/dlight/analysis/gemv.py
index c0ef5ab3b2a0..79b1b1b1a5ec 100644
--- a/python/tvm/s_tir/dlight/analysis/gemv.py
+++ b/python/tvm/s_tir/dlight/analysis/gemv.py
@@ -16,8 +16,6 @@
 # under the License.
 """Analysis for GEMV."""
 
-from typing import List, Optional
-
 from tvm import arith, ir, s_tir, tir
 
 from .common_analysis import (
@@ -28,7 +26,7 @@
 )
 
 
-def get_reduction_expr(block: tir.SBlock) -> Optional[tir.PrimExpr]:
+def get_reduction_expr(block: tir.SBlock) -> tir.PrimExpr | None:
     """Extracts the reduction expression from a TIR block.
 
     This function checks whether the given TIR block follows a reduction pattern
@@ -59,7 +57,7 @@ def get_reduction_expr(block: tir.SBlock) -> Optional[tir.PrimExpr]:
     return buffer_store.value.b
 
 
-def is_gemv(sch: s_tir.Schedule, block_info: SBlockInfo) -> Optional[List[tir.Buffer]]:
+def is_gemv(sch: s_tir.Schedule, block_info: SBlockInfo) -> list[tir.Buffer] | None:
     """Check if the block is a GEMV.
 
     Parameters
@@ -104,7 +102,7 @@ def is_gemv(sch: s_tir.Schedule, block_info: SBlockInfo) -> Optional[List[tir.Bu
 def normalize(
     sch: s_tir.Schedule,
     block_info: SBlockInfo,
-) -> Optional[bool]:
+) -> bool | None:
     """Normalize the main block."""
     block_stmt: tir.SBlock = sch.get(block_info.block_rv)
     access = arith.normalize_to_iter_sum(
diff --git a/python/tvm/s_tir/dlight/base/common_schedules.py b/python/tvm/s_tir/dlight/base/common_schedules.py
index 43a856c3bac5..348e8b98712a 100644
--- a/python/tvm/s_tir/dlight/base/common_schedules.py
+++ b/python/tvm/s_tir/dlight/base/common_schedules.py
@@ -17,7 +17,7 @@
 # ruff: noqa: E722
 """Common schedule strategies for TIR."""
 
-from typing import Callable, List
+from collections.abc import Callable
 
 from tvm import s_tir
 
@@ -26,8 +26,8 @@
 
 def try_inline(
     sch: s_tir.Schedule,
-    blocks: List[SBlockInfo],
-) -> List[SBlockInfo]:
+    blocks: list[SBlockInfo],
+) -> list[SBlockInfo]:
     """Try to inline as many blocks as possible, and return the remaining blocks.
 
     Parameters
@@ -64,8 +64,8 @@ def _trial(func: Callable):
 
 def try_inline_contiguous_spatial(
     sch: s_tir.Schedule,
-    block_infos: List[SBlockInfo],
-) -> List[SBlockInfo]:
+    block_infos: list[SBlockInfo],
+) -> list[SBlockInfo]:
     """Try to inline contiguous spatial blocks in a schedule
 
     Parameters
diff --git a/python/tvm/s_tir/dlight/base/schedule_rule.py b/python/tvm/s_tir/dlight/base/schedule_rule.py
index 794398df44a1..8764ceb3eb18 100644
--- a/python/tvm/s_tir/dlight/base/schedule_rule.py
+++ b/python/tvm/s_tir/dlight/base/schedule_rule.py
@@ -16,7 +16,7 @@
 # under the License.
 """A lightweight wrapper on an arbitrary function that can be used to schedule a TIR PrimFunc."""
 
-from typing import Callable, List, Union
+from collections.abc import Callable
 
 from tvm import s_tir, tir
 from tvm.target import Target
@@ -38,7 +38,7 @@ def apply(
         func: tir.PrimFunc,
         target: Target,
         tunable: bool,
-    ) -> Union[None, s_tir.Schedule, List[s_tir.Schedule]]:
+    ) -> None | s_tir.Schedule | list[s_tir.Schedule]:
         """Apply the ScheduleRule to the given PrimFunc.
 
         Parameters
@@ -65,7 +65,7 @@ def from_callable(
         [
             Callable[
                 [tir.PrimFunc, Target, bool],
-                Union[None, s_tir.Schedule, List[s_tir.Schedule]],
+                None | s_tir.Schedule | list[s_tir.Schedule],
             ],
         ],
         "ScheduleRule",
@@ -97,7 +97,7 @@ def apply(
                     func: tir.PrimFunc,
                     target: Target,
                     tunable: bool,
-                ) -> Union[None, s_tir.Schedule, List[s_tir.Schedule]]:
+                ) -> None | s_tir.Schedule | list[s_tir.Schedule]:
                     return f(func, target, tunable)
 
             _Rule.__name__ = name
diff --git a/python/tvm/s_tir/dlight/base/transform.py b/python/tvm/s_tir/dlight/base/transform.py
index 90ef6bd9ca59..049a920fefac 100644
--- a/python/tvm/s_tir/dlight/base/transform.py
+++ b/python/tvm/s_tir/dlight/base/transform.py
@@ -19,8 +19,6 @@
 or a space for MetaSchedule tuning
 """
 
-from typing import List, Optional
-
 from tvm import s_tir, tir
 from tvm.ir import IRModule
 from tvm.ir.transform import PassContext, module_pass
@@ -83,9 +81,9 @@ def transform_module(  # pylint: disable=missing-function-docstring
 def _apply_rules(
     func: tir.PrimFunc,
     target: Target,
-    rules: List[ScheduleRule],
+    rules: list[ScheduleRule],
     tunable: bool,
-) -> Optional[List[s_tir.Schedule]]:
+) -> list[s_tir.Schedule] | None:
     for rule in rules:
         space = rule.apply(func, target, tunable)
         if space is None:
diff --git a/python/tvm/s_tir/dlight/base/utils.py b/python/tvm/s_tir/dlight/base/utils.py
index ce21878ab900..41abae75aff5 100644
--- a/python/tvm/s_tir/dlight/base/utils.py
+++ b/python/tvm/s_tir/dlight/base/utils.py
@@ -17,13 +17,11 @@
 # pylint: disable=missing-docstring
 """Utility methods for generic GPU."""
 
-from typing import List, Optional, Union
-
 from tvm import DataType, s_tir, tir
 from tvm.target import Target
 
 
-def get_bytes(dtype: Union[DataType, str]) -> int:
+def get_bytes(dtype: DataType | str) -> int:
     if isinstance(dtype, str):
         dtype = DataType(dtype)
     return dtype.itemsize
@@ -67,9 +65,9 @@ def max_threads_per_block(target: Target) -> int:
 
 def suggest_threads_per_block(
     target: Target,
-    loops: List[tir.For],
+    loops: list[tir.For],
     max_threads_for_dynamic_loop: int = 32,
-) -> List[int]:
+) -> list[int]:
     if target.kind.name == "cuda":
         threads = 1024
     elif target.kind.name == "rocm":
@@ -80,8 +78,8 @@ def suggest_threads_per_block(
         threads = 256
     else:
         threads = 64
-    results: List[Optional[int]] = []
-    dynamic: List[int] = []
+    results: list[int | None] = []
+    dynamic: list[int] = []
     for i, loop in enumerate(loops):
         loop_extent = loop.extent
         if isinstance(loop_extent, tir.IntImm):
diff --git a/python/tvm/s_tir/dlight/benchmark/bench.py b/python/tvm/s_tir/dlight/benchmark/bench.py
index daa4e09e40df..34d361fa752a 100644
--- a/python/tvm/s_tir/dlight/benchmark/bench.py
+++ b/python/tvm/s_tir/dlight/benchmark/bench.py
@@ -16,7 +16,8 @@
 # under the License.
 """Extract self-contained benchmarking scripts for dynamic shape workloads"""
 
-from typing import TYPE_CHECKING, Callable, Dict, List, Optional, Tuple, Union
+from collections.abc import Callable
+from typing import TYPE_CHECKING, Optional
 
 import tvm
 from tvm import relax
@@ -39,15 +40,15 @@
 
 
 def benchmark(
-    mod_or_func: Union[PrimFunc, IRModule],
+    mod_or_func: PrimFunc | IRModule,
     *,
-    dym_var_sample: Dict[str, int],
-    args: Optional[List[Union[relax.TensorStructInfo, Tuple[Tuple[Union[int, str], ...], str]]]],
-    target: Optional[Union[str, tvm.target.Target]] = None,
-    func_name: Optional[str] = None,
+    dym_var_sample: dict[str, int],
+    args: list[relax.TensorStructInfo | tuple[tuple[int | str, ...], str]] | None,
+    target: str | tvm.target.Target | None = None,
+    func_name: str | None = None,
     evaluator_config: Optional["EvaluatorConfig"] = None,
     rpc_config: Optional["RPCConfig"] = None,
-) -> Tuple[List[Tuple[Tuple[int, ...], str]], float, float]:
+) -> tuple[list[tuple[tuple[int, ...], str]], float, float]:
     """Benchmark a PrimFunc or IRModule with dynamic input shapes.
 
     Parameters
@@ -105,8 +106,8 @@ def benchmark(
     input_infos = populuate_input_shape(args, dym_var_sample)
     # generate input tensors, including scalars
     # scalars are appended to the end of the list due to parsing order
-    input_tensors: List[Union[tvm.runtime.Tensor, int]] = []
-    scalar_input_tensors: List[int] = []
+    input_tensors: list[tvm.runtime.Tensor | int] = []
+    scalar_input_tensors: list[int] = []
     for input_shape, input_dtype in input_infos:
         if input_dtype == "scalar":
             # special case like [n], generate int value
@@ -152,22 +153,20 @@ def benchmark(
 
 
 def benchmark_prim_func(
-    mod_or_func: Union[PrimFunc, IRModule],
+    mod_or_func: PrimFunc | IRModule,
     *,
-    dym_var_sample_func: Callable[[Dict[str, str]], Dict[str, int]] = default_dym_var_sample_func,
-    args: Optional[
-        List[Union[relax.TensorStructInfo, Tuple[Tuple[Union[int, str], ...], str]]]
-    ] = None,
-    dym_var_dict: Optional[Dict[str, str]] = None,
+    dym_var_sample_func: Callable[[dict[str, str]], dict[str, int]] = default_dym_var_sample_func,
+    args: list[relax.TensorStructInfo | tuple[tuple[int | str, ...], str]] | None = None,
+    dym_var_dict: dict[str, str] | None = None,
     sample_number: int = 5,
-    target: Optional[Union[str, tvm.target.Target]] = None,
-    weight: Optional[int] = 1,
-    relax_func_name: Optional[str] = None,
-    prim_func_name: Optional[str] = None,
+    target: str | tvm.target.Target | None = None,
+    weight: int | None = 1,
+    relax_func_name: str | None = None,
+    prim_func_name: str | None = None,
     evaluator_config: Optional["EvaluatorConfig"] = None,
     rpc_config: Optional["RPCConfig"] = None,
-    sort_by: Optional[str] = None,
-    desc: Optional[bool] = True,
+    sort_by: str | None = None,
+    desc: bool | None = True,
 ):
     """Benchmark a PrimFunc or IRModule with dynamic input shapes and show results.
 
@@ -235,13 +234,13 @@ def benchmark_prim_func(
 
 def benchmark_relax_func(
     mod: tvm.ir.IRModule,
-    relax_func: Union[tvm.ir.GlobalVar, str],
+    relax_func: tvm.ir.GlobalVar | str,
     sample_number: int = 2,
     dym_var_sample_func: Callable[
-        [Dict[str, str]],
-        Dict[str, int],
+        [dict[str, str]],
+        dict[str, int],
     ] = default_dym_var_sample_func,
-    target: Union[str, dict, tvm.target.Target] = None,
+    target: str | dict | tvm.target.Target = None,
     evaluator_config: Optional["EvaluatorConfig"] = None,
     rpc_config: Optional["RPCConfig"] = None,
 ) -> None:
diff --git a/python/tvm/s_tir/dlight/benchmark/extract.py b/python/tvm/s_tir/dlight/benchmark/extract.py
index 3f14bbb0ee56..f25d018ed2f2 100644
--- a/python/tvm/s_tir/dlight/benchmark/extract.py
+++ b/python/tvm/s_tir/dlight/benchmark/extract.py
@@ -17,7 +17,6 @@
 """Performance debug tool for dynamic shape workloads"""
 
 from pathlib import Path
-from typing import Dict, List, Optional, Tuple, Union
 
 import cloudpickle
 
@@ -66,8 +65,8 @@
 
 
 def extract_shape(
-    arg: Union[Tuple, List, relax.Tuple, relax.ShapeStructInfo],
-) -> List[relax.ShapeStructInfo]:
+    arg: tuple | list | relax.Tuple | relax.ShapeStructInfo,
+) -> list[relax.ShapeStructInfo]:
     """Extract shape information from a relax argument.
 
     Parameters
@@ -80,7 +79,7 @@ def extract_shape(
     result : List[relax.ShapeStructInfo]
         The extracted shape information.
     """
-    if isinstance(arg, (tuple, list, tvm.relax.Tuple)):
+    if isinstance(arg, tuple | list | tvm.relax.Tuple):
         results = []
         for sub_arg in arg:
             results.extend(extract_shape(sub_arg))
@@ -89,14 +88,14 @@ def extract_shape(
 
 
 def extract_dynamic_var(
-    func_dict: Dict[
+    func_dict: dict[
         tvm.ir.GlobalVar,
-        Dict[
+        dict[
             tvm.ir.GlobalVar,
-            List[Tuple[List, int]],
+            list[tuple[list, int]],
         ],
     ],
-) -> Dict[tvm.ir.GlobalVar, Dict[str, str]]:
+) -> dict[tvm.ir.GlobalVar, dict[str, str]]:
     """Extract dynamic shape variables from a relax function dictionary.
 
     Parameters
@@ -115,7 +114,7 @@ def extract_dynamic_var(
     result : Dict[tvm.ir.GlobalVar, Dict[str, str]]
         The dictionary of dynamic shape variables. Given in format {"n": "int32", "m": "int32"}.
     """
-    dym_var_dict: Dict[tvm.ir.GlobalVar, Dict[str, str]] = {}
+    dym_var_dict: dict[tvm.ir.GlobalVar, dict[str, str]] = {}
     for gv in func_dict:  # pylint: disable=invalid-name,too-many-nested-blocks
         dym_var_dict[gv] = {}
         for functor in func_dict[gv]:
@@ -141,7 +140,7 @@ def extract_dynamic_var(
 
 
 def update_records(
-    records: Dict[List[relax.ShapeStructInfo], int], new_args: List[relax.ShapeStructInfo]
+    records: dict[list[relax.ShapeStructInfo], int], new_args: list[relax.ShapeStructInfo]
 ) -> None:
     """Update the count of a function input argument config.
 
@@ -161,7 +160,7 @@ def update_records(
 
 def extract_func_info_from_prim_func(
     func: tvm.tir.PrimFunc,
-) -> Tuple[List[Tuple[Tuple[Union[tvm.tir.Var, int], ...], str]], Dict[str, str]]:
+) -> tuple[list[tuple[tuple[tvm.tir.Var | int, ...], str]], dict[str, str]]:
     """Extract function input information from a PrimFunc.
 
     Parameters
@@ -196,9 +195,9 @@ def extract_func_info_from_prim_func(
 
 def extract_all_func_info_from_relax(
     mod: tvm.ir.IRModule,
-) -> Tuple[
-    Dict[tvm.ir.GlobalVar, Dict[tvm.ir.GlobalVar, List[Tuple[List, int]]]],
-    Dict[tvm.ir.GlobalVar, Dict[str, str]],
+) -> tuple[
+    dict[tvm.ir.GlobalVar, dict[tvm.ir.GlobalVar, list[tuple[list, int]]]],
+    dict[tvm.ir.GlobalVar, dict[str, str]],
 ]:
     """Extract function input information from a relax module.
 
@@ -215,7 +214,7 @@ def extract_all_func_info_from_relax(
     ]
         The function input information and dynamic shape variable dictionary.
     """
-    relax_func_dict: Dict[tvm.ir.GlobalVar, Dict[tvm.ir.GlobalVar, List[Tuple[List, int]]]] = {}
+    relax_func_dict: dict[tvm.ir.GlobalVar, dict[tvm.ir.GlobalVar, list[tuple[list, int]]]] = {}
     for gv, func in mod.functions_items():  # pylint: disable=invalid-name,too-many-nested-blocks
         if isinstance(func, tvm.relax.Function):
             for block in func.body.blocks:
@@ -243,11 +242,11 @@ def extract_prim_func(  # pylint: disable=too-many-arguments
     prim_func_name: str,
     func: tvm.tir.PrimFunc,
     *,
-    func_args: Optional[List[Tuple[Tuple[Union[tvm.relax.expr.Call, int], ...], str]]] = None,
-    dym_var_dict: Optional[Dict[str, str]] = None,
+    func_args: list[tuple[tuple[tvm.relax.expr.Call | int, ...], str]] | None = None,
+    dym_var_dict: dict[str, str] | None = None,
     weight: int = 1,
     sample_number: int = 5,
-    target: Optional[Union[str, dict, tvm.target.Target]] = None,
+    target: str | dict | tvm.target.Target | None = None,
 ) -> str:
     """Extract a self-contained PrimFunc test file from a Relax module.
 
@@ -284,7 +283,7 @@ def extract_prim_func(  # pylint: disable=too-many-arguments
         target = tvm.target.Target.current()
         if target is None:
             raise ValueError("Target is not specified.")
-    elif isinstance(target, (str, dict)):
+    elif isinstance(target, str | dict):
         target = tvm.target.Target(target)
     elif not isinstance(target, tvm.target.Target):
         raise TypeError("Unsupported target type: " + str(type(target)))
@@ -315,7 +314,7 @@ def extract_from_relax(
     mod: tvm.ir.IRModule,
     model_name: str,
     file_path: str,
-    target: Optional[Union[str, dict, tvm.target.Target]] = None,
+    target: str | dict | tvm.target.Target | None = None,
 ) -> None:
     """Extract self-contained PrimFunc test files from a Relax module.
 
diff --git a/python/tvm/s_tir/dlight/benchmark/utils.py b/python/tvm/s_tir/dlight/benchmark/utils.py
index d9067e3e4bdb..b91dab290467 100644
--- a/python/tvm/s_tir/dlight/benchmark/utils.py
+++ b/python/tvm/s_tir/dlight/benchmark/utils.py
@@ -16,12 +16,12 @@
 # under the License.
 """Util functions for benchmarking dynamic shape workloads"""
 
-from typing import Any, Dict, List, Tuple, Union
+from typing import Any
 
 import tvm
 from tvm import relax
 
-INPUT_SHAPE_TYPE = List[Tuple[Tuple[int, ...], str]]  # pylint: disable=invalid-name
+INPUT_SHAPE_TYPE = list[tuple[tuple[int, ...], str]]  # pylint: disable=invalid-name
 
 
 def get_func_name_from_gv(gv: tvm.ir.GlobalVar) -> str:  # pylint: disable=invalid-name
@@ -40,7 +40,7 @@ def get_func_name_from_gv(gv: tvm.ir.GlobalVar) -> str:  # pylint: disable=inval
     return gv.name_hint
 
 
-def dym_var_sample_str(sample: Dict[Union[str, tvm.relax.expr.Call], int]) -> str:
+def dym_var_sample_str(sample: dict[str | tvm.relax.expr.Call, int]) -> str:
     """Convert a variable value sample to a string.
 
     Parameters
@@ -57,8 +57,8 @@ def dym_var_sample_str(sample: Dict[Union[str, tvm.relax.expr.Call], int]) -> st
 
 
 def populuate_input_shape(
-    input_infos: List[Union[relax.TensorStructInfo, Tuple[Tuple[Union[int, str], ...], str]]],
-    dym_var_sample: Dict[str, int],
+    input_infos: list[relax.TensorStructInfo | tuple[tuple[int | str, ...], str]],
+    dym_var_sample: dict[str, int],
 ) -> INPUT_SHAPE_TYPE:
     """
     Populate input shapes with dynamic shape variable samples.
@@ -101,7 +101,7 @@ def populuate_input_shape(
     return results
 
 
-def default_dym_var_sample_func(dym_var_dict: Dict[str, str]) -> Dict[str, int]:
+def default_dym_var_sample_func(dym_var_dict: dict[str, str]) -> dict[str, int]:
     """
     Default dynamic shape variable sample function.
     Sample a random value for each dynamic shape variable.
@@ -128,7 +128,7 @@ def default_dym_var_sample_func(dym_var_dict: Dict[str, str]) -> Dict[str, int]:
 
 
 def print_results(
-    bench_results: List[Dict[str, Any]], sort_by: str = "WxTime(ms)", desc: bool = True
+    bench_results: list[dict[str, Any]], sort_by: str = "WxTime(ms)", desc: bool = True
 ):
     """Print benchmark results.
 
diff --git a/python/tvm/s_tir/dlight/cpu/gemv.py b/python/tvm/s_tir/dlight/cpu/gemv.py
index 1f2d27c2007d..3d97864bccbd 100644
--- a/python/tvm/s_tir/dlight/cpu/gemv.py
+++ b/python/tvm/s_tir/dlight/cpu/gemv.py
@@ -16,8 +16,6 @@
 # under the License.
 """A rule for GEMV and DecodeGEMV."""
 
-from typing import List, Optional, Union
-
 from tvm import s_tir, tir
 from tvm.target import Target
 
@@ -35,7 +33,7 @@ def apply(  # pylint: disable=too-many-locals,too-many-branches,too-many-return-
         func: tir.PrimFunc,
         target: Target,
         _: bool,
-    ) -> Union[None, s_tir.Schedule, List[s_tir.Schedule]]:
+    ) -> None | s_tir.Schedule | list[s_tir.Schedule]:
         if not isinstance(func, tir.PrimFunc) or not self.is_target_available(target):
             return None
         sch = s_tir.Schedule(func)
@@ -79,8 +77,8 @@ def sch_inner_reduction(  # pylint: disable=too-many-arguments, too-many-positio
         sch: s_tir.Schedule,
         target: Target,
         block: s_tir.schedule.SBlockRV,
-        vector_input_buffers: List[tir.Buffer],
-        epilogue_info: Optional[SBlockInfo],
+        vector_input_buffers: list[tir.Buffer],
+        epilogue_info: SBlockInfo | None,
     ):
         """Schedule the inner reduction block."""
 
diff --git a/python/tvm/s_tir/dlight/gpu/fallback.py b/python/tvm/s_tir/dlight/gpu/fallback.py
index bb990571ad62..73c525f4a5d9 100644
--- a/python/tvm/s_tir/dlight/gpu/fallback.py
+++ b/python/tvm/s_tir/dlight/gpu/fallback.py
@@ -17,8 +17,6 @@
 # pylint: disable=missing-docstring
 """A fallback schedule rule for GPU operators."""
 
-from typing import List, Tuple
-
 from tvm import s_tir, tir
 from tvm.target import Target
 
@@ -51,11 +49,11 @@ def apply(  # pylint: disable=too-many-locals,missing-docstring
             return None
 
         block_infos = try_inline(sch, block_infos)
-        reduction_blocks: List[Tuple[s_tir.schedule.SBlockRV, s_tir.schedule.LoopRV]] = []
+        reduction_blocks: list[tuple[s_tir.schedule.SBlockRV, s_tir.schedule.LoopRV]] = []
         for block in block_infos:
-            s_loops: List[s_tir.schedule.LoopRV] = []
-            r_loops: List[s_tir.schedule.LoopRV] = []
-            o_loops: List[s_tir.schedule.LoopRV] = []
+            s_loops: list[s_tir.schedule.LoopRV] = []
+            r_loops: list[s_tir.schedule.LoopRV] = []
+            o_loops: list[s_tir.schedule.LoopRV] = []
             dom_kind = block.dom_kind()
             block = block.block_rv
 
diff --git a/python/tvm/s_tir/dlight/gpu/gemv.py b/python/tvm/s_tir/dlight/gpu/gemv.py
index 80df2435c86a..5c4ba19e91d1 100644
--- a/python/tvm/s_tir/dlight/gpu/gemv.py
+++ b/python/tvm/s_tir/dlight/gpu/gemv.py
@@ -18,7 +18,6 @@
 """A rule for GEMV and DecodeGEMV."""
 
 from functools import reduce
-from typing import List, Optional, Union
 
 from tvm import s_tir, tir
 from tvm.target import Target
@@ -42,7 +41,7 @@ def apply(  # pylint: disable=too-many-locals,too-many-branches,too-many-return-
         func: tir.PrimFunc,
         target: Target,
         _: bool,
-    ) -> Union[None, s_tir.Schedule, List[s_tir.Schedule]]:
+    ) -> None | s_tir.Schedule | list[s_tir.Schedule]:
         if not isinstance(func, tir.PrimFunc) or not self.is_target_available(target):
             return None
         sch = s_tir.Schedule(func)
@@ -90,8 +89,8 @@ def sch_inner_reduction(  # pylint: disable=too-many-arguments, invalid-name, un
         sch: s_tir.Schedule,
         target: Target,
         block: s_tir.schedule.SBlockRV,
-        vector_input_buffers: List[tir.Buffer],
-        epilogue_info: Optional[SBlockInfo],
+        vector_input_buffers: list[tir.Buffer],
+        epilogue_info: SBlockInfo | None,
     ):
         """Schedule the inner reduction block."""
 
@@ -430,8 +429,8 @@ def sch_outer_reduction(  # pylint: disable=too-many-arguments, invalid-name, un
         sch: s_tir.Schedule,
         target: Target,
         block: s_tir.schedule.SBlockRV,
-        vector_input_buffers: List[tir.Buffer],
-        epilogue_info: Optional[SBlockInfo],
+        vector_input_buffers: list[tir.Buffer],
+        epilogue_info: SBlockInfo | None,
     ):
         """Schedule the outer reduction block."""
 
@@ -635,8 +634,8 @@ def sch_outer_reduction_fallback(  # pylint: disable=too-many-arguments, invalid
         sch: s_tir.Schedule,
         target: Target,
         block: s_tir.schedule.SBlockRV,
-        vector_input_buffers: List[tir.Buffer],
-        epilogue_info: Optional[SBlockInfo],
+        vector_input_buffers: list[tir.Buffer],
+        epilogue_info: SBlockInfo | None,
     ):
         """Schedule the outer reduction block."""
         # NOTE: Only Android is supported so far
diff --git a/python/tvm/s_tir/dlight/gpu/general_reduction.py b/python/tvm/s_tir/dlight/gpu/general_reduction.py
index bcfc8cce93d0..0005153b3ea7 100644
--- a/python/tvm/s_tir/dlight/gpu/general_reduction.py
+++ b/python/tvm/s_tir/dlight/gpu/general_reduction.py
@@ -17,8 +17,6 @@
 # pylint: disable=invalid-name
 """Reduction rule for operators including softmax, layer norm, RMS norm, etc"""
 
-from typing import List, Union
-
 from tvm import arith, s_tir, tir
 from tvm.target import Target
 
@@ -35,7 +33,7 @@ def apply(  # pylint: disable=too-many-locals
         func: tir.PrimFunc,
         target: Target,
         _: bool,
-    ) -> Union[None, s_tir.Schedule, List[s_tir.Schedule]]:
+    ) -> None | s_tir.Schedule | list[s_tir.Schedule]:
         if not isinstance(func, tir.PrimFunc) or not self.is_target_available(target):
             return None
 
diff --git a/python/tvm/s_tir/dlight/gpu/low_batch_gemv.py b/python/tvm/s_tir/dlight/gpu/low_batch_gemv.py
index 23b2e3266b43..ab08ec23e21d 100644
--- a/python/tvm/s_tir/dlight/gpu/low_batch_gemv.py
+++ b/python/tvm/s_tir/dlight/gpu/low_batch_gemv.py
@@ -18,7 +18,7 @@
 """A rule for low-batch GEMM / decode-GEMM using GEMV schedule."""
 
 from functools import reduce
-from typing import List, Literal, Optional, Set, Union
+from typing import Literal
 
 from tvm import arith, ir, s_tir, tir
 from tvm.target import Target
@@ -34,7 +34,7 @@
 from .base import GPUScheduleRule
 
 
-def _get_reduction_expr(block: tir.SBlock) -> Optional[tir.PrimExpr]:
+def _get_reduction_expr(block: tir.SBlock) -> tir.PrimExpr | None:
     # Detect and return `Y` in `X[...] = X[...] + Y`
     buffer_store = block.body
     if not isinstance(buffer_store, tir.BufferStore):
@@ -50,7 +50,7 @@ def _get_reduction_expr(block: tir.SBlock) -> Optional[tir.PrimExpr]:
     return buffer_store.value.b
 
 
-def is_gemv(sch: s_tir.Schedule, block_info: SBlockInfo) -> Optional[List[tir.Buffer]]:
+def is_gemv(sch: s_tir.Schedule, block_info: SBlockInfo) -> list[tir.Buffer] | None:
     """Check if the block is a low batch GEMM.
 
     Parameters
@@ -111,7 +111,7 @@ def is_gemv(sch: s_tir.Schedule, block_info: SBlockInfo) -> Optional[List[tir.Bu
     return ret if 0 < len(ret) < len(block_stmt.reads) else None
 
 
-def detect_dominant_read(block: tir.SBlock, const_iter_vars: Set[tir.Var]) -> tir.PrimExpr:
+def detect_dominant_read(block: tir.SBlock, const_iter_vars: set[tir.Var]) -> tir.PrimExpr:
     """Detect the dominant read indices in the block."""
     dominant_read = None
     num_read_iters = -1
@@ -131,7 +131,7 @@ def detect_dominant_read(block: tir.SBlock, const_iter_vars: Set[tir.Var]) -> ti
 def normalize(
     sch: s_tir.Schedule,
     block_info: SBlockInfo,
-) -> Optional[bool]:
+) -> bool | None:
     """Normalize the main block."""
     block_stmt: tir.SBlock = sch.get(block_info.block_rv)
     const_iter_vars = set(
@@ -199,7 +199,7 @@ def apply(  # pylint: disable=too-many-locals,too-many-branches,too-many-return-
         func: tir.PrimFunc,
         target: Target,
         _: bool,
-    ) -> Union[None, s_tir.Schedule, List[s_tir.Schedule]]:
+    ) -> None | s_tir.Schedule | list[s_tir.Schedule]:
         if not isinstance(func, tir.PrimFunc) or not self.is_target_available(target):
             return None
         sch = s_tir.Schedule(func)
@@ -291,10 +291,10 @@ def sch_inner_reduction(  # pylint: disable=too-many-arguments, invalid-name, un
         sch: s_tir.Schedule,
         target: Target,
         block: s_tir.schedule.SBlockRV,
-        dequantize_block: Optional[s_tir.schedule.SBlockRV],
-        pad_input_block: Optional[s_tir.schedule.SBlockRV],
-        vector_input_buffers: List[tir.Buffer],
-        epilogue_info: Optional[SBlockInfo],
+        dequantize_block: s_tir.schedule.SBlockRV | None,
+        pad_input_block: s_tir.schedule.SBlockRV | None,
+        vector_input_buffers: list[tir.Buffer],
+        epilogue_info: SBlockInfo | None,
         batch_pad: int,
     ):
         """Schedule the inner reduction block."""
@@ -603,10 +603,10 @@ def sch_outer_reduction(  # pylint: disable=too-many-arguments, invalid-name, un
         sch: s_tir.Schedule,
         target: Target,
         block: s_tir.schedule.SBlockRV,
-        dequantize_block: Optional[s_tir.schedule.SBlockRV],
-        pad_input_block: Optional[s_tir.schedule.SBlockRV],
-        vector_input_buffers: List[tir.Buffer],
-        epilogue_info: Optional[SBlockInfo],
+        dequantize_block: s_tir.schedule.SBlockRV | None,
+        pad_input_block: s_tir.schedule.SBlockRV | None,
+        vector_input_buffers: list[tir.Buffer],
+        epilogue_info: SBlockInfo | None,
         batch_pad: int,
     ):
         """Schedule the outer reduction block."""
diff --git a/python/tvm/s_tir/dlight/gpu/matmul.py b/python/tvm/s_tir/dlight/gpu/matmul.py
index c42072039e1e..0bacb3e00237 100644
--- a/python/tvm/s_tir/dlight/gpu/matmul.py
+++ b/python/tvm/s_tir/dlight/gpu/matmul.py
@@ -20,7 +20,6 @@
 
 from dataclasses import dataclass
 from enum import Enum
-from typing import Dict, List, Optional, Set, Tuple
 
 from tvm import s_tir, tir
 from tvm.ir import Range
@@ -143,11 +142,11 @@ def _is_one(x: PrimExpr) -> bool:
 
 
 def make_iter_fusion_index_map(
-    traits: List[IterTrait],
-    kind_order: List[IterKind],
+    traits: list[IterTrait],
+    kind_order: list[IterKind],
 ) -> tir.IndexMap:
-    fused_iters: Dict[IterKind, PrimExpr] = {}
-    input_iters: List[tir.Var] = []
+    fused_iters: dict[IterKind, PrimExpr] = {}
+    input_iters: list[tir.Var] = []
     for i, trait in enumerate(traits):
         v_i = tir.Var(f"i{i}", trait.extent.dtype)
         input_iters.append(v_i)
@@ -160,14 +159,14 @@ def make_iter_fusion_index_map(
         else:
             fused_iters[trait.kind] = v_i
 
-    final_indices: List[tir.PrimExpr] = [
+    final_indices: list[tir.PrimExpr] = [
         fused_iters.get(kind, tir.IntImm(traits[0].extent.dtype, 0)) for kind in kind_order
     ]
 
     return tir.IndexMap(input_iters, final_indices, None)
 
 
-def detect_iter_traits(block: tir.SBlock) -> Optional[Tuple[List[IterTrait]]]:
+def detect_iter_traits(block: tir.SBlock) -> tuple[list[IterTrait]] | None:
     """Detect iter traits based on the pattern C[S, I, J] += A[S, I, K] * B[S, J, K]
 
     Parameters
@@ -186,8 +185,8 @@ def detect_iter_traits(block: tir.SBlock) -> Optional[Tuple[List[IterTrait]]]:
     if len(block.reads) != 2 or len(block.writes) != 1:
         return None
 
-    def get_access_axes(region: List[Range]) -> Set[Var]:
-        axes: Set[Var] = set()
+    def get_access_axes(region: list[Range]) -> set[Var]:
+        axes: set[Var] = set()
         for r in region:
             if not _is_one(r.extent):
                 raise ValueError("Expect elemwise block access")
@@ -201,7 +200,7 @@ def get_access_axes(region: List[Range]) -> Set[Var]:
     except ValueError:
         return None
 
-    traits: Dict[Var, IterTrait] = {}
+    traits: dict[Var, IterTrait] = {}
     for iter_var in block.iter_vars:
         var = iter_var.var
         kind: IterKind
@@ -237,7 +236,7 @@ def get_access_axes(region: List[Range]) -> Set[Var]:
     return A_traits, B_traits, C_traits, block_traits
 
 
-def get_index_map(block: tir.SBlock) -> Optional[Tuple[tir.IndexMap, ...]]:
+def get_index_map(block: tir.SBlock) -> tuple[tir.IndexMap, ...] | None:
     """Get index maps for the block
 
     Parameters
@@ -327,7 +326,7 @@ def is_spatial(block: SBlockRV) -> bool:
     return reduction_blocks
 
 
-def get_in_out_dtypes(block: tir.SBlock) -> Tuple[str]:
+def get_in_out_dtypes(block: tir.SBlock) -> tuple[str]:
     """
     Detect In/Out data types for the given block based on the analysis if read/write buffers.
     """
@@ -352,7 +351,7 @@ def apply(  # pylint: disable=too-many-locals,missing-docstring
         func: tir.PrimFunc,
         target: Target,
         _: bool,
-    ) -> Optional[s_tir.Schedule]:
+    ) -> s_tir.Schedule | None:
         from tvm.s_tir.tensor_intrin.metal import (  # pylint: disable=import-outside-toplevel
             get_simdgroup_intrin_group,
         )
@@ -493,7 +492,7 @@ def apply(  # pylint: disable=too-many-locals,missing-docstring
         func: tir.PrimFunc,
         target: Target,
         _: bool,
-    ) -> Optional[s_tir.Schedule]:
+    ) -> s_tir.Schedule | None:
         from tvm.s_tir.tensor_intrin.cuda import (  # pylint: disable=import-outside-toplevel
             get_wmma_intrin_group,
         )
@@ -714,7 +713,7 @@ def apply(  # pylint: disable=too-many-locals,missing-docstring
         func: tir.PrimFunc,
         target: Target,
         _: bool,
-    ) -> Optional[s_tir.Schedule]:
+    ) -> s_tir.Schedule | None:
         from tvm.s_tir.tensor_intrin.cuda import (  # pylint: disable=import-outside-toplevel
             get_wmma_intrin_group,
         )
@@ -968,7 +967,7 @@ def apply(  # pylint: disable=too-many-locals,missing-docstring
         func: tir.PrimFunc,
         target: Target,
         _: bool,
-    ) -> Optional[s_tir.Schedule]:
+    ) -> s_tir.Schedule | None:
         if not isinstance(func, tir.PrimFunc) or not self.is_target_available(target):
             return None
         sch = s_tir.Schedule(func)
@@ -1130,8 +1129,8 @@ def sch_outer_reduction(
         sch: s_tir.Schedule,
         config: Config,
         reduction_block: s_tir.schedule.SBlockRV,
-        blocks: List[s_tir.schedule.SBlockRV],
-    ) -> Optional[s_tir.Schedule]:
+        blocks: list[s_tir.schedule.SBlockRV],
+    ) -> s_tir.Schedule | None:
         """Get vectorization factor"""
 
         def get_max_factor(n, factors):
diff --git a/python/tvm/s_tir/dlight/gpu/reduction.py b/python/tvm/s_tir/dlight/gpu/reduction.py
index 3b4afbc7d244..f0c58ff3c423 100644
--- a/python/tvm/s_tir/dlight/gpu/reduction.py
+++ b/python/tvm/s_tir/dlight/gpu/reduction.py
@@ -17,7 +17,7 @@
 """A rule for reduction."""
 
 # TODO: combine reduction rule and general reduction rule into one file.
-from typing import List, Mapping, Optional, Tuple, Union
+from collections.abc import Mapping
 
 from tvm import arith, ir, s_tir, tir
 from tvm.target import Target
@@ -32,7 +32,7 @@
 from .base import GPUScheduleRule
 
 
-def _get_reduction_expr(block: tir.SBlock) -> Optional[tir.PrimExpr]:
+def _get_reduction_expr(block: tir.SBlock) -> tir.PrimExpr | None:
     # Detect and return `Y` in `X[...] = X[...] + Y`
     buffer_store = block.body
     if not isinstance(buffer_store, tir.BufferStore):
@@ -60,7 +60,7 @@ def apply(  # pylint: disable=too-many-locals,too-many-branches,too-many-return-
         func: tir.PrimFunc,
         target: Target,
         _: bool,
-    ) -> Union[None, s_tir.Schedule, List[s_tir.Schedule]]:
+    ) -> None | s_tir.Schedule | list[s_tir.Schedule]:
         if not isinstance(func, tir.PrimFunc) or not self.is_target_available(target):
             return None
         sch = s_tir.Schedule(func)
@@ -116,7 +116,7 @@ def _normalize(  # pylint: disable=too-many-branches
         sch: s_tir.Schedule,
         block_info: SBlockInfo,
         access: arith.IterSumExpr,
-    ) -> Tuple[Optional[bool], Optional[int], Optional[Mapping[int, int]], Optional[int]]:
+    ) -> tuple[bool | None, int | None, Mapping[int, int] | None, int | None]:
         if access.base != 0:
             return None, None, None, None
         iter_to_info = {i.var: i for i in block_info.iters}
@@ -179,8 +179,8 @@ def _sch_inner_reduction(  # pylint: disable=too-many-arguments
         sch: s_tir.Schedule,
         target: Target,
         block: s_tir.schedule.SBlockRV,
-        unroll_spatial_factor: Optional[int],
-        epilogue_info: Optional[SBlockInfo],
+        unroll_spatial_factor: int | None,
+        epilogue_info: SBlockInfo | None,
         loop_order,
         s_split_index,
     ):
@@ -238,8 +238,8 @@ def _sch_inner_spatial(
         _: Target,
         block: s_tir.schedule.SBlockRV,
         block_info: SBlockInfo,
-        unroll_spatial_factor: Optional[int],
-        epilogue_info: Optional[SBlockInfo],
+        unroll_spatial_factor: int | None,
+        epilogue_info: SBlockInfo | None,
         loop_order,
         s_split_index,
     ):
diff --git a/python/tvm/s_tir/dlight/gpu/rmsnorm.py b/python/tvm/s_tir/dlight/gpu/rmsnorm.py
index c4aa377f9d82..9497a1589713 100644
--- a/python/tvm/s_tir/dlight/gpu/rmsnorm.py
+++ b/python/tvm/s_tir/dlight/gpu/rmsnorm.py
@@ -17,8 +17,6 @@
 # pylint: disable=missing-docstring
 """A RMS norm schedule rule for GPU operators."""
 
-from __future__ import annotations
-
 import tvm
 from tvm import tir
 from tvm.target import Target
@@ -81,7 +79,7 @@ def apply(  # pylint: disable=too-many-locals,missing-docstring
         func: tir.PrimFunc,
         target: Target,
         _: bool,
-    ) -> tvm.s_tir.Schedule:
+    ) -> "tvm.s_tir.Schedule":
         if target.kind.name == "cuda":
             num_tx = 512
         elif target.kind.name == "opencl":
diff --git a/python/tvm/s_tir/dlight/gpu/transpose.py b/python/tvm/s_tir/dlight/gpu/transpose.py
index 3680729e8052..66f4d613d19e 100644
--- a/python/tvm/s_tir/dlight/gpu/transpose.py
+++ b/python/tvm/s_tir/dlight/gpu/transpose.py
@@ -16,8 +16,6 @@
 # under the License.
 """Reduction rule for operators including softmax, layer norm, RMS norm, etc"""
 
-from typing import List, Union
-
 from tvm import arith, s_tir, tir
 from tvm.s_tir import Schedule
 from tvm.s_tir.schedule import SBlockRV
@@ -47,7 +45,7 @@ def apply(  # pylint: disable=too-many-locals
         func: tir.PrimFunc,
         target: Target,
         _: bool,
-    ) -> Union[None, s_tir.Schedule, List[s_tir.Schedule]]:
+    ) -> None | s_tir.Schedule | list[s_tir.Schedule]:
         # pylint: disable=invalid-name
         if not isinstance(func, tir.PrimFunc) or not self.is_target_available(target):
             return None
diff --git a/python/tvm/s_tir/meta_schedule/arg_info.py b/python/tvm/s_tir/meta_schedule/arg_info.py
index 9e5bb6f2d756..bd52cc7a9680 100644
--- a/python/tvm/s_tir/meta_schedule/arg_info.py
+++ b/python/tvm/s_tir/meta_schedule/arg_info.py
@@ -16,7 +16,7 @@
 # under the License.
 """The argument information"""
 
-from typing import Any, List, Union
+from typing import Any
 
 from tvm_ffi import register_object
 
@@ -53,7 +53,7 @@ def from_json(json_obj: Any) -> "ArgInfo":
         return _ffi_api.ArgInfoFromJSON(json_obj)  # type: ignore # pylint: disable=no-member
 
     @staticmethod
-    def from_prim_func(func: PrimFunc) -> List["ArgInfo"]:
+    def from_prim_func(func: PrimFunc) -> list["ArgInfo"]:
         """Extract a list of the argument information from PrimFunc.
 
         Parameters
@@ -69,7 +69,7 @@ def from_prim_func(func: PrimFunc) -> List["ArgInfo"]:
         return _ffi_api.ArgInfoFromPrimFunc(func)  # type: ignore # pylint: disable=no-member
 
     @staticmethod
-    def from_entry_func(mod: IRModule, remove_preproc: bool = True) -> List["ArgInfo"]:
+    def from_entry_func(mod: IRModule, remove_preproc: bool = True) -> list["ArgInfo"]:
         """Extract a list of the argument information from the entry func of an IRModule.
 
         Parameters
@@ -105,7 +105,7 @@ class TensorInfo(ArgInfo):
     def __init__(
         self,
         dtype: DataType,
-        shape: Union[ShapeTuple, List[int]],
+        shape: ShapeTuple | list[int],
     ) -> None:
         """Constructor
 
diff --git a/python/tvm/s_tir/meta_schedule/builder/builder.py b/python/tvm/s_tir/meta_schedule/builder/builder.py
index 595ac1de1d24..6a413fea4e19 100644
--- a/python/tvm/s_tir/meta_schedule/builder/builder.py
+++ b/python/tvm/s_tir/meta_schedule/builder/builder.py
@@ -17,10 +17,11 @@
 # ruff: noqa: RUF012
 """Meta Schedule builders that translate IRModule to runtime.Module, and then export"""
 
-from typing import Callable, Dict, List, Optional, Union
+from collections.abc import Callable
+from typing import Union
 
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 # isort: on
 from tvm_ffi import register_object
@@ -48,13 +49,13 @@ class BuilderInput(Object):
 
     mod: IRModule
     target: Target
-    params: Optional[Dict[str, Tensor]]
+    params: dict[str, Tensor] | None
 
     def __init__(
         self,
         mod: IRModule,
         target: Target,
-        params: Optional[Dict[str, Tensor]] = None,
+        params: dict[str, Tensor] | None = None,
     ) -> None:
         """Constructor.
 
@@ -87,13 +88,13 @@ class BuilderResult(Object):
         The error message.
     """
 
-    artifact_path: Optional[str]
-    error_msg: Optional[str]
+    artifact_path: str | None
+    error_msg: str | None
 
     def __init__(
         self,
-        artifact_path: Optional[str],
-        error_msg: Optional[str],
+        artifact_path: str | None,
+        error_msg: str | None,
     ) -> None:
         """Constructor.
 
@@ -117,7 +118,7 @@ class Builder(Object):
 
     BuilderType = Union["Builder", Literal["local"]]
 
-    def build(self, build_inputs: List[BuilderInput]) -> List[BuilderResult]:
+    def build(self, build_inputs: list[BuilderInput]) -> list[BuilderResult]:
         """Build the given inputs.
 
         Parameters
@@ -168,7 +169,7 @@ class _PyBuilder(Builder):
     See also: PyBuilder
     """
 
-    def __init__(self, f_build: Optional[Callable] = None):
+    def __init__(self, f_build: Callable | None = None):
         """Constructor."""
 
         self.__init_handle_by_constructor__(
@@ -187,7 +188,7 @@ class PyBuilder:
 
     _tvm_metadata = {"cls": _PyBuilder, "methods": ["build"]}
 
-    def build(self, build_inputs: List[BuilderInput]) -> List[BuilderResult]:
+    def build(self, build_inputs: list[BuilderInput]) -> list[BuilderResult]:
         """Build the given inputs.
 
         Parameters
diff --git a/python/tvm/s_tir/meta_schedule/builder/local_builder.py b/python/tvm/s_tir/meta_schedule/builder/local_builder.py
index df447763e29a..8197f4a01af6 100644
--- a/python/tvm/s_tir/meta_schedule/builder/local_builder.py
+++ b/python/tvm/s_tir/meta_schedule/builder/local_builder.py
@@ -19,7 +19,8 @@
 
 import os
 import tempfile
-from typing import Callable, Dict, List, Optional, Union
+from collections.abc import Callable
+from typing import Optional, Union
 
 from tvm_ffi import register_global_func
 
@@ -36,18 +37,18 @@
 
 
 T_BUILD = Callable[  # pylint: disable=invalid-name
-    [IRModule, Target, Optional[Dict[str, Tensor]]], Module
+    [IRModule, Target, dict[str, Tensor] | None], Module
 ]
 T_EXPORT = Callable[[Module], str]  # pylint: disable=invalid-name
 
 
-def _serialize_params(params: Optional[Dict[str, Tensor]]) -> Optional[bytearray]:
+def _serialize_params(params: dict[str, Tensor] | None) -> bytearray | None:
     if params is None:
         return None
     return save_param_dict(params)
 
 
-def _deserialize_params(params: Optional[bytearray]) -> Optional[Dict[str, Tensor]]:
+def _deserialize_params(params: bytearray | None) -> dict[str, Tensor] | None:
     if params is None:
         return None
     return load_param_dict(params)
@@ -106,18 +107,18 @@ def default_export(mod: Module) -> str:
 
     max_workers: int
     timeout_sec: float
-    initializer: Optional[Callable[[], None]]
-    f_build: Union[None, str, T_BUILD]
-    f_export: Union[None, str, T_EXPORT]
+    initializer: Callable[[], None] | None
+    f_build: None | str | T_BUILD
+    f_export: None | str | T_EXPORT
 
     def __init__(
         self,
         *,
-        max_workers: Optional[int] = None,
+        max_workers: int | None = None,
         timeout_sec: float = 30.0,
-        f_build: Union[None, str, T_BUILD] = None,
-        f_export: Union[None, str, T_EXPORT] = None,
-        initializer: Optional[Callable[[], None]] = None,
+        f_build: None | str | T_BUILD = None,
+        f_export: None | str | T_EXPORT = None,
+        initializer: Callable[[], None] | None = None,
     ) -> None:
         """Constructor.
 
@@ -150,8 +151,8 @@ def __init__(
         self.f_export = f_export
         self._sanity_check()
 
-    def build(self, build_inputs: List[BuilderInput]) -> List[BuilderResult]:
-        results: List[BuilderResult] = []
+    def build(self, build_inputs: list[BuilderInput]) -> list[BuilderResult]:
+        results: list[BuilderResult] = []
         map_result: MapResult
 
         # Here we restart the PopenPool everytime because of a known memory leak issue with the
@@ -215,11 +216,11 @@ def _check(f_build, f_export) -> None:
 
 
 def _worker_func(
-    _f_build: Union[None, str, T_BUILD],
-    _f_export: Union[None, str, T_EXPORT],
+    _f_build: None | str | T_BUILD,
+    _f_export: None | str | T_EXPORT,
     mod: IRModule,
     target: Target,
-    params: Optional[bytearray],
+    params: bytearray | None,
 ) -> str:
     # Step 0. Get the registered functions
     f_build: T_BUILD = get_global_func_with_default_on_worker(
@@ -238,7 +239,7 @@ def _worker_func(
 
 
 @register_global_func("s_tir.meta_schedule.builder.default_build")
-def default_build(mod: IRModule, target: Target, _params: Optional[Dict[str, Tensor]]) -> Module:
+def default_build(mod: IRModule, target: Target, _params: dict[str, Tensor] | None) -> Module:
     """Default build function.
 
     Parameters
diff --git a/python/tvm/s_tir/meta_schedule/cost_model/cost_model.py b/python/tvm/s_tir/meta_schedule/cost_model/cost_model.py
index 0927fd961e1d..31bf26463f6d 100644
--- a/python/tvm/s_tir/meta_schedule/cost_model/cost_model.py
+++ b/python/tvm/s_tir/meta_schedule/cost_model/cost_model.py
@@ -18,10 +18,11 @@
 """Meta Schedule CostModel."""
 
 import ctypes
-from typing import Callable, List, Optional, Union
+from collections.abc import Callable
+from typing import Union
 
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 # isort: on
 
@@ -66,8 +67,8 @@ def save(self, path: str) -> None:
     def update(
         self,
         context: TuneContext,
-        candidates: List[MeasureCandidate],
-        results: List[RunnerResult],
+        candidates: list[MeasureCandidate],
+        results: list[RunnerResult],
     ) -> None:
         """Update the cost model given running results.
 
@@ -82,7 +83,7 @@ def update(
         """
         _ffi_api.CostModelUpdate(self, context, candidates, results)  # type: ignore # pylint: disable=no-member
 
-    def predict(self, context: TuneContext, candidates: List[MeasureCandidate]) -> np.ndarray:
+    def predict(self, context: TuneContext, candidates: list[MeasureCandidate]) -> np.ndarray:
         """Predict normalized score with the cost model.
 
         Parameters
@@ -164,15 +165,15 @@ class _PyCostModel(CostModel):
 
     def __init__(
         self,
-        f_load: Optional[Callable] = None,
-        f_save: Optional[Callable] = None,
-        f_update: Optional[Callable] = None,
-        predict_func: Optional[Callable] = None,
-        f_as_string: Optional[Callable] = None,
+        f_load: Callable | None = None,
+        f_save: Callable | None = None,
+        f_update: Callable | None = None,
+        predict_func: Callable | None = None,
+        f_as_string: Callable | None = None,
     ):
         """Constructor."""
 
-        def f_predict(context: TuneContext, candidates: List[MeasureCandidate], return_ptr) -> None:
+        def f_predict(context: TuneContext, candidates: list[MeasureCandidate], return_ptr) -> None:
             n = len(candidates)
             return_ptr = ctypes.cast(return_ptr, ctypes.POINTER(ctypes.c_double))
             array_wrapper = np.ctypeslib.as_array(return_ptr, shape=(n,))
@@ -228,8 +229,8 @@ def save(self, path: str) -> None:
     def update(
         self,
         context: TuneContext,
-        candidates: List[MeasureCandidate],
-        results: List[RunnerResult],
+        candidates: list[MeasureCandidate],
+        results: list[RunnerResult],
     ) -> None:
         """Update the cost model given running results.
 
@@ -244,7 +245,7 @@ def update(
         """
         raise NotImplementedError
 
-    def predict(self, context: TuneContext, candidates: List[MeasureCandidate]) -> np.ndarray:
+    def predict(self, context: TuneContext, candidates: list[MeasureCandidate]) -> np.ndarray:
         """Predict given the measure candidates.
 
         Parameters
diff --git a/python/tvm/s_tir/meta_schedule/cost_model/mlp_model.py b/python/tvm/s_tir/meta_schedule/cost_model/mlp_model.py
index c9aa71fde945..0feb5af5ebe4 100644
--- a/python/tvm/s_tir/meta_schedule/cost_model/mlp_model.py
+++ b/python/tvm/s_tir/meta_schedule/cost_model/mlp_model.py
@@ -26,7 +26,7 @@
 import tempfile
 from collections import OrderedDict
 from itertools import chain as itertools_chain
-from typing import Dict, List, NamedTuple, Optional, Tuple
+from typing import NamedTuple
 
 import numpy as np  # type: ignore
 import torch  # type: ignore
@@ -153,14 +153,14 @@ class FeatureGroup:
     """
 
     group_hash: str
-    features: List[np.ndarray]
+    features: list[np.ndarray]
     costs: np.ndarray
     min_cost: float
 
     def __init__(
         self,
         group_hash: str,
-        features: List[np.ndarray],
+        features: list[np.ndarray],
         costs: np.ndarray,
     ) -> None:
         self.group_hash = group_hash
@@ -170,7 +170,7 @@ def __init__(
 
     def append(  # pylint: disable=missing-function-docstring
         self,
-        features: List[np.ndarray],
+        features: list[np.ndarray],
         costs: np.ndarray,
     ) -> None:
         self.features.extend(features)
@@ -252,7 +252,7 @@ def _fetch_indices(self, indices):
 def lambda_rank_loss(  # pylint: disable=too-many-locals
     preds: "torch.Tensor",
     labels: "torch.Tensor",
-    k: Optional[int] = None,
+    k: int | None = None,
     eps: float = 1e-10,
     sigma: float = 1.0,
 ) -> "torch.Tensor":
@@ -417,9 +417,9 @@ def forward(  # pylint: disable=missing-function-docstring
 
 def extract_features(
     context: TuneContext,
-    candidates: List[MeasureCandidate],
-    results: Optional[List[RunnerResult]] = None,
-    extractor: Optional[FeatureExtractor] = None,
+    candidates: list[MeasureCandidate],
+    results: list[RunnerResult] | None = None,
+    extractor: FeatureExtractor | None = None,
 ):
     """Extract feature vectors and compute mean costs.
 
@@ -476,14 +476,14 @@ class State:
     """
 
     model: SegmentSumMLP
-    data: Dict[str, FeatureGroup]
+    data: dict[str, FeatureGroup]
     data_size: int
     untrained_size: int
 
     def __init__(
         self,
-        model_config: Optional[SegmentSumMLPConfig] = None,
-        extractor: Optional[FeatureExtractor] = None,
+        model_config: SegmentSumMLPConfig | None = None,
+        extractor: FeatureExtractor | None = None,
     ):
         model_config = model_config or SegmentSumMLPConfig()
         extractor = extractor or PerStoreFeature(extract_workload=True)
@@ -589,7 +589,7 @@ def save(self, path: str) -> None:
 
     def add_to_group(
         self,
-        features: List[np.ndarray],
+        features: list[np.ndarray],
         costs: np.ndarray,
         group_hash: str,
     ):
@@ -667,8 +667,8 @@ class SegmentSumMLPTrainer:
 
     def __init__(
         self,
-        train_config: Optional[TrainerConfig] = None,
-        state: Optional[State] = None,
+        train_config: TrainerConfig | None = None,
+        state: State | None = None,
     ):
         train_config = train_config or TrainerConfig()
         state = state or State()
@@ -682,9 +682,9 @@ def __init__(
 
     def train_step(
         self,
-        data: Tuple["torch.Tensor", "torch.Tensor", "torch.Tensor"],
+        data: tuple["torch.Tensor", "torch.Tensor", "torch.Tensor"],
         batch: int = 0,
-        train_loss: Optional[float] = None,
+        train_loss: float | None = None,
     ) -> float:
         """Helper function for training on a single batch.
 
@@ -729,7 +729,7 @@ def train_step(
 
     def predict_step(
         self,
-        data: Tuple["torch.Tensor", "torch.Tensor", "torch.Tensor"],
+        data: tuple["torch.Tensor", "torch.Tensor", "torch.Tensor"],
     ):
         """Helper function for predicting (validating) on a single batch.
 
@@ -840,7 +840,7 @@ def train_full(self):  # pylint: disable=too-many-locals
 
     def train_incremental(
         self,
-        features: List[np.ndarray],
+        features: list[np.ndarray],
         results: np.ndarray,
     ):
         """Training on incremental data.
@@ -870,8 +870,8 @@ def train_incremental(
 
     def predict_incremental(
         self,
-        features: List[np.ndarray],
-        results: Optional[np.ndarray] = None,
+        features: list[np.ndarray],
+        results: np.ndarray | None = None,
     ) -> np.ndarray:
         """Predicting (validating) on incremental data.
 
@@ -912,7 +912,7 @@ def predict_incremental(
 
     def update(
         self,
-        features: List[np.ndarray],
+        features: list[np.ndarray],
         costs: np.ndarray,
         group_hash: str,
     ):
@@ -951,7 +951,7 @@ class MLPModel(PyCostModel):
     def __init__(
         self,
         *,
-        trainer: Optional[SegmentSumMLPTrainer] = None,
+        trainer: SegmentSumMLPTrainer | None = None,
     ):
         super().__init__()
         self.trainer = trainer or SegmentSumMLPTrainer()
@@ -979,8 +979,8 @@ def save(self, path: str) -> None:
     def update(
         self,
         context: TuneContext,
-        candidates: List[MeasureCandidate],
-        results: List[RunnerResult],
+        candidates: list[MeasureCandidate],
+        results: list[RunnerResult],
     ) -> None:
         """Update the dataset, re-train the cost model if not frozen.
 
@@ -998,7 +998,7 @@ def update(
         )
         self.trainer.update(features, mean_costs, shash2hex(context.mod))
 
-    def predict(self, context: TuneContext, candidates: List[MeasureCandidate]) -> np.ndarray:
+    def predict(self, context: TuneContext, candidates: list[MeasureCandidate]) -> np.ndarray:
         """Predict given the measure candidates.
 
         Parameters
diff --git a/python/tvm/s_tir/meta_schedule/cost_model/random_model.py b/python/tvm/s_tir/meta_schedule/cost_model/random_model.py
index bf616111b56f..292fd4a96417 100644
--- a/python/tvm/s_tir/meta_schedule/cost_model/random_model.py
+++ b/python/tvm/s_tir/meta_schedule/cost_model/random_model.py
@@ -18,8 +18,6 @@
 Random cost model
 """
 
-from typing import List, Optional, Tuple, Union
-
 from ..cost_model import PyCostModel
 from ..runner import RunnerResult
 from ..search_strategy import MeasureCandidate
@@ -47,15 +45,15 @@ class RandomModel(PyCostModel):
 
     import numpy as np  # type: ignore # pylint: disable=import-outside-toplevel
 
-    random_state: Union[Tuple[str, np.ndarray, int, int, float], dict]
-    path: Optional[str]
+    random_state: tuple[str, np.ndarray, int, int, float] | dict
+    path: str | None
 
     def __init__(
         self,
         *,
-        seed: Optional[int] = None,
-        path: Optional[str] = None,
-        max_range: Optional[int] = 100,
+        seed: int | None = None,
+        path: str | None = None,
+        max_range: int | None = 100,
     ):
         import numpy as np  # type: ignore # pylint: disable=import-outside-toplevel
 
@@ -94,8 +92,8 @@ def save(self, path: str) -> None:
     def update(
         self,
         context: TuneContext,
-        candidates: List[MeasureCandidate],
-        results: List[RunnerResult],
+        candidates: list[MeasureCandidate],
+        results: list[RunnerResult],
     ) -> None:
         """Update the cost model given running results.
 
@@ -109,7 +107,7 @@ def update(
             The running results of the measure candidates.
         """
 
-    def predict(self, context: TuneContext, candidates: List[MeasureCandidate]) -> np.ndarray:  # type: ignore # pylint: disable=used-before-assignment
+    def predict(self, context: TuneContext, candidates: list[MeasureCandidate]) -> np.ndarray:  # type: ignore # pylint: disable=used-before-assignment
         """Update the cost model given running results.
 
         Parameters
diff --git a/python/tvm/s_tir/meta_schedule/cost_model/xgb_model.py b/python/tvm/s_tir/meta_schedule/cost_model/xgb_model.py
index c85c8736634f..2bb29eb8cde6 100644
--- a/python/tvm/s_tir/meta_schedule/cost_model/xgb_model.py
+++ b/python/tvm/s_tir/meta_schedule/cost_model/xgb_model.py
@@ -19,11 +19,11 @@
 import os
 import tempfile
 from collections import OrderedDict
+from collections.abc import Callable
 from itertools import chain as itertools_chain
-from typing import TYPE_CHECKING, Any, Callable, Dict, List, NamedTuple, Optional, Tuple
+from typing import TYPE_CHECKING, Any, Literal, NamedTuple, Optional
 
 import numpy as np  # type: ignore
-from typing_extensions import Literal
 
 from ....contrib.tar import tar, untar
 from ....runtime import Tensor
@@ -79,8 +79,8 @@ class PackSum:
 
     def __init__(
         self,
-        xs: List[np.ndarray],  # pylint: disable=invalid-name
-        ys: Optional[np.ndarray],  # pylint: disable=invalid-name
+        xs: list[np.ndarray],  # pylint: disable=invalid-name
+        ys: np.ndarray | None,  # pylint: disable=invalid-name
     ):
         """Create PackSum format given a batch of samples
 
@@ -118,7 +118,7 @@ def predict_with_score(self, pred: np.ndarray) -> np.ndarray:
         """
         return np.bincount(self.ids, weights=pred)
 
-    def obj_square_error(self, ys_pred: np.ndarray) -> Tuple[np.ndarray, np.ndarray]:
+    def obj_square_error(self, ys_pred: np.ndarray) -> tuple[np.ndarray, np.ndarray]:
         """Implement square error loss on pack-sum format as
         a custom objective function for xgboost.
 
@@ -144,7 +144,7 @@ def obj_square_error(self, ys_pred: np.ndarray) -> Tuple[np.ndarray, np.ndarray]
         hessian = np.ones_like(gradient)
         return gradient * ys, hessian * ys
 
-    def rmse(self, ys_pred: np.ndarray) -> Tuple[str, float]:
+    def rmse(self, ys_pred: np.ndarray) -> tuple[str, float]:
         """Evaluate RMSE (rooted mean square error) in the pack-sum format
 
         Parameters
@@ -173,7 +173,7 @@ def average_peak_score(
         self,
         ys_pred: np.ndarray,
         n: int,
-    ) -> Tuple[str, float]:
+    ) -> tuple[str, float]:
         """Evaluate average-peak-score@N in the pack-sum format
 
         Parameters
@@ -230,7 +230,7 @@ class XGBConfig(NamedTuple):
     min_child_weight: float = 0
     eta: float = 0.2
     seed: int = 43
-    nthread: Optional[int] = None
+    nthread: int | None = None
     tree_method: Literal["auto", "exact", "approx", "hist", "gpu_hist"] = "auto"
 
     def to_dict(self):
@@ -263,14 +263,14 @@ class FeatureGroup:
     """
 
     group_hash: str
-    features: List[np.ndarray]
+    features: list[np.ndarray]
     costs: np.ndarray
     min_cost: float
 
     def __init__(
         self,
         group_hash: str,
-        features: List[np.ndarray],
+        features: list[np.ndarray],
         costs: np.ndarray,
     ) -> None:
         self.group_hash = group_hash
@@ -280,7 +280,7 @@ def __init__(
 
     def append(
         self,
-        features: List[np.ndarray],
+        features: list[np.ndarray],
         costs: np.ndarray,
     ) -> None:
         self.features.extend(features)
@@ -322,7 +322,7 @@ class XGBModel(PyCostModel):
     verbose_eval: int
     average_peak_n: int
     # states
-    data: Dict[str, FeatureGroup]
+    data: dict[str, FeatureGroup]
     data_size: int
     booster: Optional["xgb.Booster"]
     # adaptive training
@@ -343,8 +343,8 @@ def __init__(
         verbose_eval: int = 25,
         average_peak_n: int = 32,
         adaptive_training: bool = True,
-        num_tuning_cores: Optional[int] = None,
-        tree_method: Optional[Literal["auto", "exact", "approx", "hist", "gpu_hist"]] = None,
+        num_tuning_cores: int | None = None,
+        tree_method: Literal["auto", "exact", "approx", "hist", "gpu_hist"] | None = None,
     ):
         super().__init__()
         if not isinstance(extractor, FeatureExtractor):
@@ -461,8 +461,8 @@ def save(self, path: str) -> None:
     def update(
         self,
         context: "TuneContext",
-        candidates: List[MeasureCandidate],
-        results: List[RunnerResult],
+        candidates: list[MeasureCandidate],
+        results: list[RunnerResult],
     ) -> None:
         """Update the cost model given running results.
 
@@ -552,7 +552,7 @@ def _mean_cost(x: RunnerResult) -> float:
     def predict(
         self,
         context: "TuneContext",
-        candidates: List[MeasureCandidate],
+        candidates: list[MeasureCandidate],
     ) -> np.ndarray:
         """Predict the normalized score using the cost model.
 
@@ -588,7 +588,7 @@ def predict(
 
     def _train(  # type: ignore # pylint: disable=invalid-name
         self,
-        xs: List[np.ndarray],
+        xs: list[np.ndarray],
         ys: np.ndarray,
     ) -> None:
         import xgboost as xgb  # type: ignore # pylint: disable=import-outside-toplevel
@@ -624,7 +624,7 @@ def avg_peak_score(ys_pred: np.ndarray, d_train: "xgb.DMatrix"):  # type: ignore
 
     def _predict(  # type: ignore # pylint: disable=invalid-name
         self,
-        xs: List[np.ndarray],
+        xs: list[np.ndarray],
     ) -> np.ndarray:
         d_test = PackSum(xs=xs, ys=None)
         pred = self.booster.predict(d_test.dmatrix)
@@ -633,9 +633,9 @@ def _predict(  # type: ignore # pylint: disable=invalid-name
 
     def _validate(  # type: ignore # pylint: disable=invalid-name
         self,
-        xs: List[np.ndarray],
+        xs: list[np.ndarray],
         ys: np.ndarray,
-    ) -> List[Tuple[str, float]]:
+    ) -> list[tuple[str, float]]:
         """Evaluate the score of inputs.
 
         Parameters
@@ -658,7 +658,7 @@ def average_peak_score(ys_pred: np.ndarray):
             return d_valid.average_peak_score(ys_pred, n=self.average_peak_n)
 
         ys_pred = self.booster.predict(d_valid.dmatrix)
-        eval_result: List[Tuple[str, float]] = [
+        eval_result: list[tuple[str, float]] = [
             feval(ys_pred)
             for feval in (
                 average_peak_score,
@@ -672,10 +672,10 @@ def average_peak_score(ys_pred: np.ndarray):
 def _get_custom_call_back(
     early_stopping_rounds: int,
     verbose_eval: int,
-    fevals: List[Callable],
-    evals: List[Tuple["xgb.DMatrix", str]],
+    fevals: list[Callable],
+    evals: list[tuple["xgb.DMatrix", str]],
     focused_metric: str = "tr-p-rmse",
-    cvfolds: Optional[List["xgb.training.CVPack"]] = None,
+    cvfolds: list["xgb.training.CVPack"] | None = None,
 ) -> "TrainingCallback":
     """Get a customized callback function for XGBoost. Work around xgboost import."""
 
@@ -703,16 +703,16 @@ def __init__(
             self,
             early_stopping_rounds: int,
             verbose_eval: int,
-            fevals: List[Callable],
-            evals: List[Tuple["xgb.DMatrix", str]],
+            fevals: list[Callable],
+            evals: list[tuple["xgb.DMatrix", str]],
             focused_metric: str = "tr-p-rmse",
-            cvfolds: Optional[List["xgb.training.CVPack"]] = None,
+            cvfolds: list["xgb.training.CVPack"] | None = None,
         ):
             self.early_stopping_rounds = early_stopping_rounds
             self.verbose_eval = verbose_eval
             self.fevals = fevals
             self.evals = evals
-            self.state: Dict[str, Any] = {}
+            self.state: dict[str, Any] = {}
             self.focused_metric = focused_metric
             self.sort_key = make_metric_sorter(focused_metric=focused_metric)
             self.cvfolds = cvfolds
@@ -739,7 +739,7 @@ def init(self, model: "xgb.Booster"):
                 booster.set_attr(best_iteration=str(self.state["best_iteration"]))
                 booster.set_attr(best_score=str(self.state["best_score"]))
 
-        def after_iteration(self, model: "xgb.Booster", epoch: int, evals_log: Dict):  # pylint: disable = unused-argument
+        def after_iteration(self, model: "xgb.Booster", epoch: int, evals_log: dict):  # pylint: disable = unused-argument
             """Internal function for after_iteration"""
             # pylint:disable = import-outside-toplevel
             try:
@@ -774,10 +774,10 @@ def _fmt_metric(value, show_stdv=True):
                 self.init(model)
             booster: xgb.Booster = model
             iteration: int = epoch
-            cvfolds: List[xgb.training.CVPack] = self.cvfolds
+            cvfolds: list[xgb.training.CVPack] = self.cvfolds
             ##### Evaluation #####
             # `eval_result` is a list of (key, score)
-            eval_result: List[Tuple[str, float]] = []
+            eval_result: list[tuple[str, float]] = []
             if cvfolds is None:
                 eval_result = list(
                     itertools_chain.from_iterable(
diff --git a/python/tvm/s_tir/meta_schedule/database/database.py b/python/tvm/s_tir/meta_schedule/database/database.py
index 3d9d6b1b0fe3..e44d344b8359 100644
--- a/python/tvm/s_tir/meta_schedule/database/database.py
+++ b/python/tvm/s_tir/meta_schedule/database/database.py
@@ -17,10 +17,11 @@
 # ruff: noqa: RUF012
 """TuningRecord database"""
 
-from typing import Any, Callable, List, Optional, Union
+from collections.abc import Callable
+from typing import Any, Optional, Union
 
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 # isort: on
 
@@ -102,17 +103,17 @@ class TuningRecord(Object):
 
     trace: Trace
     workload: Workload
-    run_secs: Optional[List[float]]
-    target: Optional[Target]
-    args_info: Optional[List[ArgInfo]]
+    run_secs: list[float] | None
+    target: Target | None
+    args_info: list[ArgInfo] | None
 
     def __init__(  # type: ignore # pylint: disable=too-many-arguments
         self,
         trace: Trace,
         workload: Workload,
-        run_secs: Optional[List[float]] = None,
-        target: Optional[Target] = None,
-        args_info: Optional[List[ArgInfo]] = None,
+        run_secs: list[float] | None = None,
+        target: Target | None = None,
+        args_info: list[ArgInfo] | None = None,
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.TuningRecord,  # type: ignore # pylint: disable=no-member
@@ -207,7 +208,7 @@ def commit_tuning_record(self, record: TuningRecord) -> None:
         """
         _ffi_api.DatabaseCommitTuningRecord(self, record)  # type: ignore # pylint: disable=no-member
 
-    def get_top_k(self, workload: Workload, top_k: int) -> List[TuningRecord]:
+    def get_top_k(self, workload: Workload, top_k: int) -> list[TuningRecord]:
         """Get the top K valid tuning records of given workload from the database.
 
         Parameters
@@ -224,7 +225,7 @@ def get_top_k(self, workload: Workload, top_k: int) -> List[TuningRecord]:
         """
         return _ffi_api.DatabaseGetTopK(self, workload, top_k)  # type: ignore # pylint: disable=no-member
 
-    def get_all_tuning_records(self) -> List[TuningRecord]:
+    def get_all_tuning_records(self) -> list[TuningRecord]:
         """Get all the tuning records from the database.
 
         Returns
@@ -249,7 +250,7 @@ def query_tuning_record(
         mod: IRModule,
         target: Target,
         workload_name: str,
-    ) -> Optional[TuningRecord]:
+    ) -> TuningRecord | None:
         """Query the best record of the given workload from the database.
 
         Parameters
@@ -273,7 +274,7 @@ def query_schedule(
         mod: IRModule,
         target: Target,
         workload_name: str,
-    ) -> Optional[Schedule]:
+    ) -> Schedule | None:
         """Query the best schedule of the given workload from the database.
 
         Parameters
@@ -297,7 +298,7 @@ def query_ir_module(
         mod: IRModule,
         target: Target,
         workload_name: str,
-    ) -> Optional[IRModule]:
+    ) -> IRModule | None:
         """Query the best IRModule of the given workload from the database.
 
         Parameters
@@ -334,12 +335,8 @@ def query(
         target: Target,
         *,
         workload_name: str = "main",
-        kind: Union[
-            Literal["schedule"],
-            Literal["record"],
-            Literal["ir_module"],
-        ] = "schedule",
-    ) -> Union[Schedule, IRModule, TuningRecord]:
+        kind: Literal["schedule"] | Literal["record"] | Literal["ir_module"] = "schedule",
+    ) -> Schedule | IRModule | TuningRecord:
         """Query the database to retrieve the best optimization outcome of the given workload.
 
         Parameters
@@ -380,15 +377,9 @@ def current() -> Optional["Database"]:
 
     @staticmethod
     def create(  # pylint: disable=keyword-arg-before-vararg
-        kind: Union[
-            Literal[
-                "json",
-                "memory",
-                "union",
-                "ordered_union",
-            ],
-            Callable[[Schedule], bool],
-        ] = "json",
+        kind: (
+            Literal["json", "memory", "union", "ordered_union"] | Callable[[Schedule], bool]
+        ) = "json",
         *args,
         **kwargs,
     ) -> "Database":
@@ -441,15 +432,15 @@ class _PyDatabase(Database):
 
     def __init__(
         self,
-        f_has_workload: Optional[Callable] = None,
-        f_commit_workload: Optional[Callable] = None,
-        f_commit_tuning_record: Optional[Callable] = None,
-        f_get_top_k: Optional[Callable] = None,
-        f_get_all_tuning_records: Optional[Callable] = None,
-        f_query_tuning_record: Optional[Callable] = None,
-        f_query_schedule: Optional[Callable] = None,
-        f_query_ir_module: Optional[Callable] = None,
-        f_size: Optional[Callable] = None,
+        f_has_workload: Callable | None = None,
+        f_commit_workload: Callable | None = None,
+        f_commit_tuning_record: Callable | None = None,
+        f_get_top_k: Callable | None = None,
+        f_get_all_tuning_records: Callable | None = None,
+        f_query_tuning_record: Callable | None = None,
+        f_query_schedule: Callable | None = None,
+        f_query_ir_module: Callable | None = None,
+        f_size: Callable | None = None,
         module_equality: str = "structural",
     ):
         """Constructor."""
@@ -530,7 +521,7 @@ def commit_tuning_record(self, record: TuningRecord) -> None:
         """
         raise NotImplementedError
 
-    def get_top_k(self, workload: Workload, top_k: int) -> List[TuningRecord]:
+    def get_top_k(self, workload: Workload, top_k: int) -> list[TuningRecord]:
         """Get the top K tuning records of given workload from the database.
 
         Parameters
@@ -547,7 +538,7 @@ def get_top_k(self, workload: Workload, top_k: int) -> List[TuningRecord]:
         """
         raise NotImplementedError
 
-    def get_all_tuning_records(self) -> List[TuningRecord]:
+    def get_all_tuning_records(self) -> list[TuningRecord]:
         """Get all the tuning records from the database.
 
         Returns
@@ -558,8 +549,8 @@ def get_all_tuning_records(self) -> List[TuningRecord]:
         raise NotImplementedError
 
     def query_tuning_record(
-        self, mod: IRModule, target: Target, workload_name: Optional[str] = None
-    ) -> Optional[TuningRecord]:
+        self, mod: IRModule, target: Target, workload_name: str | None = None
+    ) -> TuningRecord | None:
         """Query a tuning record from the database.
 
         Parameters
@@ -585,8 +576,8 @@ def query_tuning_record(
         )
 
     def query_schedule(
-        self, mod: IRModule, target: Target, workload_name: Optional[str] = None
-    ) -> Optional[Schedule]:
+        self, mod: IRModule, target: Target, workload_name: str | None = None
+    ) -> Schedule | None:
         """Query a schedule from the database.
 
         Parameters
@@ -612,8 +603,8 @@ def query_schedule(
         )
 
     def query_ir_module(
-        self, mod: IRModule, target: Target, workload_name: Optional[str] = None
-    ) -> Optional[IRModule]:
+        self, mod: IRModule, target: Target, workload_name: str | None = None
+    ) -> IRModule | None:
         """Query an IRModule from the database.
 
         Parameters
diff --git a/python/tvm/s_tir/meta_schedule/database/json_database.py b/python/tvm/s_tir/meta_schedule/database/json_database.py
index b442cb878fbb..8ca76c7e30c1 100644
--- a/python/tvm/s_tir/meta_schedule/database/json_database.py
+++ b/python/tvm/s_tir/meta_schedule/database/json_database.py
@@ -17,7 +17,6 @@
 """The default database that uses a JSON File to store tuning records"""
 
 import os.path as osp
-from typing import Optional
 
 from tvm_ffi import register_object
 
@@ -52,10 +51,10 @@ class JSONDatabase(Database):
 
     def __init__(
         self,
-        path_workload: Optional[str] = None,
-        path_tuning_record: Optional[str] = None,
+        path_workload: str | None = None,
+        path_tuning_record: str | None = None,
         *,
-        work_dir: Optional[str] = None,
+        work_dir: str | None = None,
         allow_missing: bool = True,
         module_equality: str = "structural",
     ) -> None:
diff --git a/python/tvm/s_tir/meta_schedule/database/schedule_fn_database.py b/python/tvm/s_tir/meta_schedule/database/schedule_fn_database.py
index baf58c724521..62057705fc4d 100644
--- a/python/tvm/s_tir/meta_schedule/database/schedule_fn_database.py
+++ b/python/tvm/s_tir/meta_schedule/database/schedule_fn_database.py
@@ -16,7 +16,7 @@
 # under the License.
 """A database for injecting handcrafted schedule functions."""
 
-from typing import Callable
+from collections.abc import Callable
 
 from tvm_ffi import register_object
 
diff --git a/python/tvm/s_tir/meta_schedule/extracted_task.py b/python/tvm/s_tir/meta_schedule/extracted_task.py
index ce1098f9df2b..c0b67ed8f4f3 100644
--- a/python/tvm/s_tir/meta_schedule/extracted_task.py
+++ b/python/tvm/s_tir/meta_schedule/extracted_task.py
@@ -16,8 +16,6 @@
 # under the License.
 """Extracted tasks from high-level IR."""
 
-from typing import List
-
 from tvm_ffi import register_object
 
 from tvm.ir import IRModule
@@ -47,7 +45,7 @@ class ExtractedTask(Object):
 
     task_name: str
     mod: IRModule
-    dispatched: List[IRModule]
+    dispatched: list[IRModule]
     weight: int
 
     def __init__(
@@ -55,7 +53,7 @@ def __init__(
         task_name: str,
         mod: IRModule,
         target: Target,
-        dispatched: List[IRModule],
+        dispatched: list[IRModule],
         weight: int,
     ) -> None:
         self.__init_handle_by_constructor__(
diff --git a/python/tvm/s_tir/meta_schedule/feature_extractor/feature_extractor.py b/python/tvm/s_tir/meta_schedule/feature_extractor/feature_extractor.py
index bab0588f7880..6f04bd2cf263 100644
--- a/python/tvm/s_tir/meta_schedule/feature_extractor/feature_extractor.py
+++ b/python/tvm/s_tir/meta_schedule/feature_extractor/feature_extractor.py
@@ -17,10 +17,11 @@
 # ruff: noqa: RUF012
 """Meta Schedule FeatureExtractor."""
 
-from typing import Callable, List, Optional, Union
+from collections.abc import Callable
+from typing import Union
 
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 # isort: on
 
@@ -42,8 +43,8 @@ class FeatureExtractor(Object):
     FeatureExtractorType = Union[Literal["per-store-feature"], "FeatureExtractor"]
 
     def extract_from(
-        self, context: TuneContext, candidates: List[MeasureCandidate]
-    ) -> List[Tensor]:
+        self, context: TuneContext, candidates: list[MeasureCandidate]
+    ) -> list[Tensor]:
         """Extract features from the given measure candidate.
 
         Parameters
@@ -86,7 +87,7 @@ class _PyFeatureExtractor(FeatureExtractor):
     See also: PyFeatureExtractor
     """
 
-    def __init__(self, f_extract_from: Callable, f_as_string: Optional[Callable] = None):
+    def __init__(self, f_extract_from: Callable, f_as_string: Callable | None = None):
         """Constructor."""
 
         self.__init_handle_by_constructor__(
@@ -110,8 +111,8 @@ class PyFeatureExtractor:
     }
 
     def extract_from(
-        self, context: TuneContext, candidates: List[MeasureCandidate]
-    ) -> List[Tensor]:
+        self, context: TuneContext, candidates: list[MeasureCandidate]
+    ) -> list[Tensor]:
         """Extract features from the given measure candidate.
 
         Parameters
diff --git a/python/tvm/s_tir/meta_schedule/feature_extractor/random_feature_extractor.py b/python/tvm/s_tir/meta_schedule/feature_extractor/random_feature_extractor.py
index 9ea41d1c16c9..8cf7e2f2bfc3 100644
--- a/python/tvm/s_tir/meta_schedule/feature_extractor/random_feature_extractor.py
+++ b/python/tvm/s_tir/meta_schedule/feature_extractor/random_feature_extractor.py
@@ -16,8 +16,6 @@
 # under the License.
 """Random Feature Extractor."""
 
-from typing import List, Tuple, Union
-
 import numpy as np  # type: ignore
 
 import tvm.runtime
@@ -44,7 +42,7 @@ class RandomFeatureExtractor(PyFeatureExtractor):
 
     feature_size: int
     max_block_num: int
-    random_state: Union[Tuple[str, np.ndarray, int, int, float], dict]
+    random_state: tuple[str, np.ndarray, int, int, float] | dict
 
     def __init__(self, *, feature_size: int = 30, max_block_num: int = 5, seed=0):
         super().__init__()
@@ -55,8 +53,8 @@ def __init__(self, *, feature_size: int = 30, max_block_num: int = 5, seed=0):
         self.random_state = np.random.get_state()
 
     def extract_from(
-        self, context: TuneContext, candidates: List[MeasureCandidate]
-    ) -> List[tvm.runtime.Tensor]:
+        self, context: TuneContext, candidates: list[MeasureCandidate]
+    ) -> list[tvm.runtime.Tensor]:
         np.random.set_state(self.random_state)
         result = [
             np.random.rand(np.random.randint(1, self.max_block_num + 1), self.feature_size)
diff --git a/python/tvm/s_tir/meta_schedule/logging.py b/python/tvm/s_tir/meta_schedule/logging.py
index b39a362ec86d..04710cc53626 100644
--- a/python/tvm/s_tir/meta_schedule/logging.py
+++ b/python/tvm/s_tir/meta_schedule/logging.py
@@ -20,8 +20,9 @@
 import logging.config
 import os
 import os.path as osp
+from collections.abc import Callable
 from logging import Logger
-from typing import Any, Callable, Dict, List, Optional
+from typing import Any
 
 
 def get_logger(name: str) -> Logger:
@@ -40,7 +41,7 @@ def get_logger(name: str) -> Logger:
     return logging.getLogger(name)
 
 
-def get_logging_func(logger: Logger) -> Optional[Callable[[int, str, int, str], None]]:
+def get_logging_func(logger: Logger) -> Callable[[int, str, int, str], None] | None:
     """Get the logging function.
 
     Parameters
@@ -78,8 +79,8 @@ def logging_func(level: int, filename: str, lineo: int, msg: str):
 
 def create_loggers(
     log_dir: str,
-    params: List[Dict[str, Any]],
-    logger_config: Optional[Dict[str, Any]] = None,
+    params: list[dict[str, Any]],
+    logger_config: dict[str, Any] | None = None,
     disable_existing_loggers: bool = False,
 ):
     """Create loggers from configuration"""
@@ -176,9 +177,9 @@ def create_loggers(
 
 
 def _batch_parameterize_config(
-    config: Dict[str, Any],
-    params: List[Dict[str, str]],
-) -> Dict[str, Any]:
+    config: dict[str, Any],
+    params: list[dict[str, str]],
+) -> dict[str, Any]:
     """Parameterize the given configuration with multiple parameters sets.
 
     Parameters
@@ -204,9 +205,9 @@ def _batch_parameterize_config(
 
 
 def _parameterize_config(
-    config: Dict[str, Any],
-    params: Dict[str, str],
-) -> Dict[str, Any]:
+    config: dict[str, Any],
+    params: dict[str, str],
+) -> dict[str, Any]:
     """Parameterize the given configuration.
 
     Parameters
@@ -237,8 +238,8 @@ def _parameterize_config(
 
 def get_loggers_from_work_dir(
     work_dir: str,
-    task_names: List[str],
-) -> List[Logger]:
+    task_names: list[str],
+) -> list[Logger]:
     """Create loggers from work directory
 
     Parameters
diff --git a/python/tvm/s_tir/meta_schedule/measure_callback/measure_callback.py b/python/tvm/s_tir/meta_schedule/measure_callback/measure_callback.py
index 045d17c230f4..0b7d39916603 100644
--- a/python/tvm/s_tir/meta_schedule/measure_callback/measure_callback.py
+++ b/python/tvm/s_tir/meta_schedule/measure_callback/measure_callback.py
@@ -17,10 +17,11 @@
 # ruff: noqa: RUF012
 """Meta Schedule MeasureCallback."""
 
-from typing import TYPE_CHECKING, Callable, List, Optional, Union
+from collections.abc import Callable
+from typing import TYPE_CHECKING, Union
 
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 # isort: on
 
@@ -42,15 +43,15 @@
 class MeasureCallback(Object):
     """Rules to apply after measure results is available."""
 
-    CallbackListType = Union[List["MeasureCallback"], "MeasureCallback", Literal["default"]]
+    CallbackListType = Union[list["MeasureCallback"], "MeasureCallback", Literal["default"]]
 
     def apply(
         self,
         task_scheduler: "TaskScheduler",
         task_id: int,
-        measure_candidates: List[MeasureCandidate],
-        builder_results: List[BuilderResult],
-        runner_results: List[RunnerResult],
+        measure_candidates: list[MeasureCandidate],
+        builder_results: list[BuilderResult],
+        runner_results: list[RunnerResult],
     ) -> None:
         """Apply a measure callback to the given schedule.
 
@@ -77,7 +78,7 @@ def apply(
         )
 
     @staticmethod
-    def create(kind: Literal["default"]) -> List["MeasureCallback"]:
+    def create(kind: Literal["default"]) -> list["MeasureCallback"]:
         """Create a list of measure callbacks."""
         if kind == "default":
             return _ffi_api.MeasureCallbackDefault()  # type: ignore # pylint: disable=no-member
@@ -93,7 +94,7 @@ class _PyMeasureCallback(MeasureCallback):
     See also: PyMeasureCallback
     """
 
-    def __init__(self, f_apply: Callable, f_as_string: Optional[Callable] = None):
+    def __init__(self, f_apply: Callable, f_as_string: Callable | None = None):
         """Constructor."""
 
         self.__init_handle_by_constructor__(
@@ -120,9 +121,9 @@ def apply(
         self,
         task_scheduler: "TaskScheduler",
         task_id: int,
-        measure_candidates: List[MeasureCandidate],
-        builder_results: List[BuilderResult],
-        runner_results: List[RunnerResult],
+        measure_candidates: list[MeasureCandidate],
+        builder_results: list[BuilderResult],
+        runner_results: list[RunnerResult],
     ) -> None:
         """Apply a measure callback to the given schedule.
 
diff --git a/python/tvm/s_tir/meta_schedule/mutator/mutator.py b/python/tvm/s_tir/meta_schedule/mutator/mutator.py
index 0dcfda80a187..fb81f4af5bca 100644
--- a/python/tvm/s_tir/meta_schedule/mutator/mutator.py
+++ b/python/tvm/s_tir/meta_schedule/mutator/mutator.py
@@ -17,10 +17,11 @@
 # ruff: noqa: RUF012
 """Meta Schedule Mutator."""
 
-from typing import TYPE_CHECKING, Callable, Dict, Optional
+from collections.abc import Callable
+from typing import TYPE_CHECKING
 
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 # isort: on
 
@@ -51,7 +52,7 @@ def _initialize_with_tune_context(self, context: "TuneContext") -> None:
             self, context
         )
 
-    def apply(self, trace: Trace) -> Optional[Trace]:
+    def apply(self, trace: Trace) -> Trace | None:
         """Apply the mutator function to the given trace.
 
         Parameters
@@ -84,7 +85,7 @@ def create(
             "cuda-tensorcore",
             "hexagon",
         ],
-    ) -> Dict["Mutator", float]:
+    ) -> dict["Mutator", float]:
         """Create a list of default mutators.
 
         Parameters
@@ -125,10 +126,10 @@ class _PyMutator(Mutator):
 
     def __init__(
         self,
-        f_initialize_with_tune_context: Optional[Callable] = None,
-        f_apply: Optional[Callable] = None,
-        f_clone: Optional[Callable] = None,
-        f_as_string: Optional[Callable] = None,
+        f_initialize_with_tune_context: Callable | None = None,
+        f_apply: Callable | None = None,
+        f_clone: Callable | None = None,
+        f_as_string: Callable | None = None,
     ):
         """Constructor."""
 
@@ -164,7 +165,7 @@ def _initialize_with_tune_context(self, context: "TuneContext") -> None:
         """
         raise NotImplementedError
 
-    def apply(self, trace: Trace, _) -> Optional[Trace]:
+    def apply(self, trace: Trace, _) -> Trace | None:
         """Apply the mutator function to the given trace.
 
         Parameters
diff --git a/python/tvm/s_tir/meta_schedule/post_optimization/space.py b/python/tvm/s_tir/meta_schedule/post_optimization/space.py
index 6b46cabca504..5cae34ef9211 100644
--- a/python/tvm/s_tir/meta_schedule/post_optimization/space.py
+++ b/python/tvm/s_tir/meta_schedule/post_optimization/space.py
@@ -19,7 +19,7 @@
 import json
 import random
 from copy import deepcopy
-from typing import Any, Dict, List
+from typing import Any
 
 import numpy as np  # type: ignore
 
@@ -52,9 +52,9 @@ def __init__(self, data: Any, workload: Any, target: Target):
         self.target = target
         self.dev = self.get_device_type(target)
         self.total_dims = 0
-        self.dims: List[int] = []
-        self.start: List[int] = []
-        self.config_space: Dict[str, List[int]] = dict()
+        self.dims: list[int] = []
+        self.start: list[int] = []
+        self.config_space: dict[str, list[int]] = dict()
         self.create_space()
 
     def __repr__(self) -> str:
@@ -77,7 +77,7 @@ def get_value(self, key, pos):
         """Return the space"""
         return self.config_space[key][pos]
 
-    def add_space(self, space_list: list, element_list: list, limit=10000) -> List[int]:
+    def add_space(self, space_list: list, element_list: list, limit=10000) -> list[int]:
         """Return a list without repeat and with limited value"""
         new_list = element_list
         for elem in space_list:
diff --git a/python/tvm/s_tir/meta_schedule/post_optimization/utils.py b/python/tvm/s_tir/meta_schedule/post_optimization/utils.py
index afee5eda50f5..e17915293091 100644
--- a/python/tvm/s_tir/meta_schedule/post_optimization/utils.py
+++ b/python/tvm/s_tir/meta_schedule/post_optimization/utils.py
@@ -18,7 +18,6 @@
 
 import json
 import os
-from typing import Dict
 
 import numpy as np  # type: ignore
 
@@ -82,7 +81,7 @@ def get_time(log: str) -> list:
     return best_time
 
 
-def read_cfg_file(path_tuning_file: str, path_workload_file: str) -> Dict[int, list]:
+def read_cfg_file(path_tuning_file: str, path_workload_file: str) -> dict[int, list]:
     """Colect the info from meta logfile
 
     Parameters
@@ -100,7 +99,7 @@ def read_cfg_file(path_tuning_file: str, path_workload_file: str) -> Dict[int, l
         for line in log_file.readlines():
             workload_list.append(json.loads(line))
 
-    cfg: Dict[int, list] = dict()
+    cfg: dict[int, list] = dict()
     with open(path_tuning_file, encoding="utf-8") as log_file:
         for line in log_file.readlines():
             data = json.loads(line)
diff --git a/python/tvm/s_tir/meta_schedule/postproc/postproc.py b/python/tvm/s_tir/meta_schedule/postproc/postproc.py
index c78fdd5e4b62..a1f17f734ccd 100644
--- a/python/tvm/s_tir/meta_schedule/postproc/postproc.py
+++ b/python/tvm/s_tir/meta_schedule/postproc/postproc.py
@@ -17,10 +17,11 @@
 # ruff: noqa: RUF012
 """Meta Schedule Postproc."""
 
-from typing import TYPE_CHECKING, Callable, List, Optional
+from collections.abc import Callable
+from typing import TYPE_CHECKING
 
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 # isort: on
 
@@ -78,7 +79,7 @@ def clone(self) -> "Postproc":
         return _ffi_api.PostprocClone(self)  # type: ignore # pylint: disable=no-member
 
     @staticmethod
-    def create(kind: Literal["llvm", "cuda", "cuda-tensorcore", "hexagon"]) -> List["Postproc"]:
+    def create(kind: Literal["llvm", "cuda", "cuda-tensorcore", "hexagon"]) -> list["Postproc"]:
         """Create a list of default postprocessors.
 
         Parameters
@@ -119,10 +120,10 @@ class _PyPostproc(Postproc):
 
     def __init__(
         self,
-        f_initialize_with_tune_context: Optional[Callable] = None,
-        f_apply: Optional[Callable] = None,
-        f_clone: Optional[Callable] = None,
-        f_as_string: Optional[Callable] = None,
+        f_initialize_with_tune_context: Callable | None = None,
+        f_apply: Callable | None = None,
+        f_clone: Callable | None = None,
+        f_as_string: Callable | None = None,
     ):
         """Constructor."""
 
diff --git a/python/tvm/s_tir/meta_schedule/profiler.py b/python/tvm/s_tir/meta_schedule/profiler.py
index a27e200bf4d9..993e15a85604 100644
--- a/python/tvm/s_tir/meta_schedule/profiler.py
+++ b/python/tvm/s_tir/meta_schedule/profiler.py
@@ -18,7 +18,7 @@
 """A context manager that profiles tuning time cost for different parts."""
 
 from contextlib import contextmanager
-from typing import Dict, Optional
+from typing import Optional
 
 from tvm_ffi import register_object
 
@@ -36,7 +36,7 @@ def __init__(self) -> None:
             _ffi_api.Profiler,  # type: ignore # pylint: disable=no-member
         )
 
-    def get(self) -> Dict[str, float]:
+    def get(self) -> dict[str, float]:
         """Get the profiling results in seconds"""
         return _ffi_api.ProfilerGet(self)  # type: ignore # pylint: disable=no-member
 
diff --git a/python/tvm/s_tir/meta_schedule/relax_integration.py b/python/tvm/s_tir/meta_schedule/relax_integration.py
index e1199ee1e166..161e159a9a42 100644
--- a/python/tvm/s_tir/meta_schedule/relax_integration.py
+++ b/python/tvm/s_tir/meta_schedule/relax_integration.py
@@ -17,10 +17,10 @@
 """Meta schedule integration with high-level IR"""
 
 import warnings
-from typing import TYPE_CHECKING, Dict, List, Optional, Tuple, Union
+from typing import TYPE_CHECKING, Union
 
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 # isort: on
 
@@ -58,9 +58,9 @@
 def extract_tasks(
     mod: Union[IRModule, "relax.Function"],
     target: Target,
-    params: Optional[Dict[str, Tensor]] = None,
+    params: dict[str, Tensor] | None = None,
     module_equality: str = "structural",
-) -> List[ExtractedTask]:
+) -> list[ExtractedTask]:
     """Extract tuning tasks from a relax program.
 
     Parameters
@@ -102,13 +102,13 @@ def extract_tasks(
 
 
 def extracted_tasks_to_tune_contexts(
-    extracted_tasks: List[ExtractedTask],
+    extracted_tasks: list[ExtractedTask],
     work_dir: str,
     space: SpaceGenerator.SpaceGeneratorType = "post-order-apply",
     strategy: SearchStrategy.SearchStrategyType = "evolutionary",
-    num_threads: Union[Literal["physical", "logical"], int] = "physical",
-    seed: Optional[int] = None,
-) -> Tuple[List[TuneContext], List[float]]:
+    num_threads: Literal["physical", "logical"] | int = "physical",
+    seed: int | None = None,
+) -> tuple[list[TuneContext], list[float]]:
     """Convert ExtractedTask to TuneContext.
 
     Parameters
@@ -133,8 +133,8 @@ def extracted_tasks_to_tune_contexts(
     task_weights : List[float]
         The weights of the tasks
     """
-    tasks: List[TuneContext] = []
-    task_weights: List[float] = []
+    tasks: list[TuneContext] = []
+    task_weights: list[float] = []
     for task, logger, rand_state in zip(
         extracted_tasks,
         get_loggers_from_work_dir(work_dir, [t.task_name for t in extracted_tasks]),
@@ -161,12 +161,12 @@ def extracted_tasks_to_tune_contexts(
 
 def tune_relax(
     mod: Union[IRModule, "relax.Function"],
-    params: Dict[str, Tensor],
-    target: Union[str, Target],
+    params: dict[str, Tensor],
+    target: str | Target,
     work_dir: str,
     max_trials_global: int,
-    max_trials_per_task: Optional[int] = None,
-    op_names: Optional[List[str]] = None,
+    max_trials_per_task: int | None = None,
+    op_names: list[str] | None = None,
     *,
     num_trials_per_iter: int = 64,
     builder: Builder.BuilderType = "local",
@@ -177,7 +177,7 @@ def tune_relax(
     task_scheduler: TaskScheduler.TaskSchedulerType = "gradient",
     space: SpaceGenerator.SpaceGeneratorType = "post-order-apply",
     strategy: SearchStrategy.SearchStrategyType = "evolutionary",
-    seed: Optional[int] = None,
+    seed: int | None = None,
     module_equality: str = "structural",
 ) -> Database:
     """Tune a Relax program.
@@ -274,12 +274,12 @@ def tune_relax(
 @register_global_func("tvm.s_tir.meta_schedule.tune_relax")
 def _tune_relax(
     mod: Union[IRModule, "relax.Function"],
-    params: Dict[str, Tensor],
-    target: Union[str, Target],
+    params: dict[str, Tensor],
+    target: str | Target,
     work_dir: str,
     max_trials_global: int,
-    max_trials_per_task: Optional[int] = None,
-    op_names: Optional[List[str]] = None,
+    max_trials_per_task: int | None = None,
+    op_names: list[str] | None = None,
     *,
     num_trials_per_iter: int = 64,
     builder: Builder.BuilderType = "local",
@@ -290,7 +290,7 @@ def _tune_relax(
     task_scheduler: TaskScheduler.TaskSchedulerType = "gradient",
     space: SpaceGenerator.SpaceGeneratorType = "post-order-apply",
     strategy: SearchStrategy.SearchStrategyType = "evolutionary",
-    seed: Optional[int] = None,
+    seed: int | None = None,
     module_equality: str = "structural",
 ) -> Database:
     """Interface with tuning api to tune a Relax program.
@@ -381,8 +381,8 @@ def _tune_relax(
 def compile_relax(
     database: Database,
     mod: IRModule,
-    target: Union[Target, str],
-    params: Optional[Dict[str, Tensor]],
+    target: Target | str,
+    params: dict[str, Tensor] | None,
     enable_warning: bool = False,
 ) -> "relax.VMExecutable":
     """Compile a relax program with a MetaSchedule database.
diff --git a/python/tvm/s_tir/meta_schedule/runner/config.py b/python/tvm/s_tir/meta_schedule/runner/config.py
index d9125580a1f1..13e3b0b713e7 100644
--- a/python/tvm/s_tir/meta_schedule/runner/config.py
+++ b/python/tvm/s_tir/meta_schedule/runner/config.py
@@ -18,7 +18,7 @@
 
 import os
 from threading import Thread
-from typing import NamedTuple, Optional, Union
+from typing import NamedTuple, Optional
 
 from tvm import rpc
 
@@ -84,9 +84,9 @@ class RPCConfig(NamedTuple):
         Priority of the RPC session
     """
 
-    tracker_host: Optional[str] = None
-    tracker_port: Union[None, int, str] = None
-    tracker_key: Optional[str] = None
+    tracker_host: str | None = None
+    tracker_port: None | int | str = None
+    tracker_key: str | None = None
     session_priority: int = 1
     session_timeout_sec: int = 10
 
@@ -129,7 +129,7 @@ def connect_tracker(self) -> rpc.TrackerSession:
         tracker : TrackerSession
             The connected tracker session
         """
-        tracker: Optional[rpc.TrackerSession] = None
+        tracker: rpc.TrackerSession | None = None
 
         def _connect():
             nonlocal tracker
diff --git a/python/tvm/s_tir/meta_schedule/runner/local_runner.py b/python/tvm/s_tir/meta_schedule/runner/local_runner.py
index 5457201ba8b5..2e73f6b695a3 100644
--- a/python/tvm/s_tir/meta_schedule/runner/local_runner.py
+++ b/python/tvm/s_tir/meta_schedule/runner/local_runner.py
@@ -18,8 +18,8 @@
 
 import logging
 import subprocess
+from collections.abc import Callable
 from contextlib import contextmanager
-from typing import Callable, List, Optional, Union
 
 import tvm
 
@@ -46,16 +46,16 @@
         T_ARG_INFO_JSON_OBJ_LIST,  # The metadata information of the arguments to be allocated
         int,  # The number of repeated allocations to be done
     ],
-    List[T_ARGUMENT_LIST],  # A list of argument lists
+    list[T_ARGUMENT_LIST],  # A list of argument lists
 ]
 T_RUN_EVALUATOR = Callable[  # pylint: disable=invalid-name
     [
         Module,  # The Module opened on the remote
         Device,  # The device on the remote
         EvaluatorConfig,  # The evaluator configuration
-        List[T_ARGUMENT_LIST],  # A list of argument lists
+        list[T_ARGUMENT_LIST],  # A list of argument lists
     ],
-    List[float],  # A list of running time
+    list[float],  # A list of running time
 ]
 T_CLEANUP = Callable[  # pylint: disable=invalid-name
     [],
@@ -80,12 +80,10 @@ class LocalRunnerFuture(PyRunnerFuture):
     of LocalRunnerFuture object
     """
 
-    res: Optional[List[float]]
-    error_message: Optional[str]
+    res: list[float] | None
+    error_message: str | None
 
-    def __init__(
-        self, res: Optional[List[float]] = None, error_message: Optional[str] = None
-    ) -> None:
+    def __init__(self, res: list[float] | None = None, error_message: str | None = None) -> None:
         """Constructor
 
         Parameters
@@ -117,15 +115,15 @@ def result(self) -> RunnerResult:
 
 
 def _worker_func(
-    _f_alloc_argument: Optional[str],
-    _f_run_evaluator: Optional[str],
-    _f_cleanup: Optional[str],
+    _f_alloc_argument: str | None,
+    _f_run_evaluator: str | None,
+    _f_cleanup: str | None,
     evaluator_config: EvaluatorConfig,
     alloc_repeat: int,
     artifact_path: str,
     device_type: str,
     args_info: T_ARG_INFO_JSON_OBJ_LIST,
-) -> List[float]:
+) -> list[float]:
     f_alloc_argument: T_ALLOC_ARGUMENT = get_global_func_with_default_on_worker(
         _f_alloc_argument, default_alloc_argument
     )
@@ -150,14 +148,14 @@ def resource_handler():
         # Step 2: Allocate input arguments
         with Profiler.timeit("LocalRunner/alloc_argument"):
             device = tvm.runtime.device(device_type, 0)
-            repeated_args: List[T_ARGUMENT_LIST] = f_alloc_argument(
+            repeated_args: list[T_ARGUMENT_LIST] = f_alloc_argument(
                 device,
                 args_info,
                 alloc_repeat,
             )
         # Step 3: Run time_evaluator
         with Profiler.timeit("LocalRunner/run_evaluator"):
-            costs: List[float] = f_run_evaluator(
+            costs: list[float] = f_run_evaluator(
                 rt_mod,
                 device,
                 evaluator_config,
@@ -228,22 +226,22 @@ def default_cleanup() -> None:
     cooldown_sec: float
     alloc_repeat: int
 
-    f_alloc_argument: Union[T_ALLOC_ARGUMENT, str, None]
-    f_run_evaluator: Union[T_RUN_EVALUATOR, str, None]
-    f_cleanup: Union[T_CLEANUP, str, None]
+    f_alloc_argument: T_ALLOC_ARGUMENT | str | None
+    f_run_evaluator: T_RUN_EVALUATOR | str | None
+    f_cleanup: T_CLEANUP | str | None
 
     pool: PopenPoolExecutor
 
     def __init__(
         self,
         timeout_sec: float = 30,
-        evaluator_config: Optional[EvaluatorConfig] = None,
+        evaluator_config: EvaluatorConfig | None = None,
         cooldown_sec: float = 0.0,
         alloc_repeat: int = 1,
-        f_alloc_argument: Union[T_ALLOC_ARGUMENT, str, None] = None,
-        f_run_evaluator: Union[T_RUN_EVALUATOR, str, None] = None,
-        f_cleanup: Union[T_CLEANUP, str, None] = None,
-        initializer: Optional[Callable[[], None]] = None,
+        f_alloc_argument: T_ALLOC_ARGUMENT | str | None = None,
+        f_run_evaluator: T_RUN_EVALUATOR | str | None = None,
+        f_cleanup: T_CLEANUP | str | None = None,
+        initializer: Callable[[], None] | None = None,
     ) -> None:
         """Constructor
 
@@ -288,8 +286,8 @@ def __init__(
         )
         self._sanity_check()
 
-    def run(self, runner_inputs: List[RunnerInput]) -> List[RunnerFuture]:
-        results: List[RunnerFuture] = []
+    def run(self, runner_inputs: list[RunnerInput]) -> list[RunnerFuture]:
+        results: list[RunnerFuture] = []
         for runner_input in runner_inputs:
             future = self.pool.submit(
                 _worker_func,
@@ -303,7 +301,7 @@ def run(self, runner_inputs: List[RunnerInput]) -> List[RunnerFuture]:
                 tuple(arg_info.as_json() for arg_info in runner_input.args_info),
             )
             try:
-                result: List[float] = future.result()
+                result: list[float] = future.result()
                 error_message: str = None
             except TimeoutError:
                 result = None
@@ -338,7 +336,7 @@ def default_alloc_argument(
     device: Device,
     args_info: T_ARG_INFO_JSON_OBJ_LIST,
     alloc_repeat: int,
-) -> List[T_ARGUMENT_LIST]:
+) -> list[T_ARGUMENT_LIST]:
     """Default function to allocate the arguments
 
     Parameters
@@ -365,8 +363,8 @@ def default_run_evaluator(
     rt_mod: Module,
     device: Device,
     evaluator_config: EvaluatorConfig,
-    repeated_args: List[T_ARGUMENT_LIST],
-) -> List[float]:
+    repeated_args: list[T_ARGUMENT_LIST],
+) -> list[float]:
     """Default function to run the evaluator
 
     Parameters
diff --git a/python/tvm/s_tir/meta_schedule/runner/rpc_runner.py b/python/tvm/s_tir/meta_schedule/runner/rpc_runner.py
index d0d34e4dd1d7..ebfdd57715ff 100644
--- a/python/tvm/s_tir/meta_schedule/runner/rpc_runner.py
+++ b/python/tvm/s_tir/meta_schedule/runner/rpc_runner.py
@@ -18,8 +18,8 @@
 
 import concurrent.futures
 import os.path as osp
+from collections.abc import Callable
 from contextlib import contextmanager
-from typing import Callable, List, Optional, Union
 
 from tvm.contrib.popen_pool import PopenPoolExecutor
 from tvm.rpc import RPCSession
@@ -63,7 +63,7 @@
         T_ARG_INFO_JSON_OBJ_LIST,  # The metadata information of the arguments to be allocated
         int,  # The number of repeated allocations to be done
     ],
-    List[T_ARGUMENT_LIST],  # A list of argument lists
+    list[T_ARGUMENT_LIST],  # A list of argument lists
 ]
 T_RUN_EVALUATOR = Callable[  # pylint: disable=invalid-name
     [
@@ -71,14 +71,14 @@
         Module,  # The Module opened on the remote
         Device,  # The device on the remote
         EvaluatorConfig,  # The evaluator configuration
-        List[T_ARGUMENT_LIST],  # A list of argument lists
+        list[T_ARGUMENT_LIST],  # A list of argument lists
     ],
-    List[float],  # A list of running time
+    list[float],  # A list of running time
 ]
 T_CLEANUP = Callable[  # pylint: disable=invalid-name
     [
-        Optional[RPCSession],  # The RPC Session to be cleaned up
-        Optional[str],  # remote path to the artifact
+        RPCSession | None,  # The RPC Session to be cleaned up
+        str | None,  # remote path to the artifact
     ],
     None,
 ]
@@ -118,7 +118,7 @@ def done(self) -> bool:
 
     def result(self) -> RunnerResult:
         try:
-            run_secs: List[float] = self.future.result()
+            run_secs: list[float] = self.future.result()
         except TimeoutError:
             return RunnerResult(
                 None,
@@ -225,27 +225,27 @@ def default_cleanup(
     cooldown_sec: float
     alloc_repeat: int
 
-    f_create_session: Union[T_CREATE_SESSION, str, None]
-    f_upload_module: Union[T_UPLOAD_MODULE, str, None]
-    f_alloc_argument: Union[T_ALLOC_ARGUMENT, str, None]
-    f_run_evaluator: Union[T_RUN_EVALUATOR, str, None]
-    f_cleanup: Union[T_CLEANUP, str, None]
+    f_create_session: T_CREATE_SESSION | str | None
+    f_upload_module: T_UPLOAD_MODULE | str | None
+    f_alloc_argument: T_ALLOC_ARGUMENT | str | None
+    f_run_evaluator: T_RUN_EVALUATOR | str | None
+    f_cleanup: T_CLEANUP | str | None
 
     pool: PopenPoolExecutor
 
     def __init__(
         self,
-        rpc_config: Optional[RPCConfig] = None,
-        evaluator_config: Optional[EvaluatorConfig] = None,
+        rpc_config: RPCConfig | None = None,
+        evaluator_config: EvaluatorConfig | None = None,
         cooldown_sec: float = 0.0,
         alloc_repeat: int = 1,
-        f_create_session: Union[T_CREATE_SESSION, str, None] = None,
-        f_upload_module: Union[T_UPLOAD_MODULE, str, None] = None,
-        f_alloc_argument: Union[T_ALLOC_ARGUMENT, str, None] = None,
-        f_run_evaluator: Union[T_RUN_EVALUATOR, str, None] = None,
-        f_cleanup: Union[T_CLEANUP, str, None] = None,
-        max_workers: Optional[int] = None,
-        initializer: Optional[Callable[[], None]] = None,
+        f_create_session: T_CREATE_SESSION | str | None = None,
+        f_upload_module: T_UPLOAD_MODULE | str | None = None,
+        f_alloc_argument: T_ALLOC_ARGUMENT | str | None = None,
+        f_run_evaluator: T_RUN_EVALUATOR | str | None = None,
+        f_cleanup: T_CLEANUP | str | None = None,
+        max_workers: int | None = None,
+        initializer: Callable[[], None] | None = None,
     ) -> None:
         """Constructor
 
@@ -293,8 +293,8 @@ def __init__(
         )
         self._sanity_check()
 
-    def run(self, runner_inputs: List[RunnerInput]) -> List[RunnerFuture]:
-        results: List[RunnerFuture] = []
+    def run(self, runner_inputs: list[RunnerInput]) -> list[RunnerFuture]:
+        results: list[RunnerFuture] = []
         for runner_input in runner_inputs:
             future = RPCRunnerFuture(
                 future=self.pool.submit(
@@ -342,18 +342,18 @@ def _check(
 
 
 def _worker_func(
-    _f_create_session: Union[T_CREATE_SESSION, str, None],
-    _f_upload_module: Union[T_UPLOAD_MODULE, str, None],
-    _f_alloc_argument: Union[T_ALLOC_ARGUMENT, str, None],
-    _f_run_evaluator: Union[T_RUN_EVALUATOR, str, None],
-    _f_cleanup: Union[T_CLEANUP, str, None],
+    _f_create_session: T_CREATE_SESSION | str | None,
+    _f_upload_module: T_UPLOAD_MODULE | str | None,
+    _f_alloc_argument: T_ALLOC_ARGUMENT | str | None,
+    _f_run_evaluator: T_RUN_EVALUATOR | str | None,
+    _f_cleanup: T_CLEANUP | str | None,
     rpc_config: RPCConfig,
     evaluator_config: EvaluatorConfig,
     alloc_repeat: int,
     artifact_path: str,
     device_type: str,
     args_info: T_ARG_INFO_JSON_OBJ_LIST,
-) -> List[float]:
+) -> list[float]:
     # Step 0. Get the registered functions
     f_create_session: T_CREATE_SESSION = get_global_func_with_default_on_worker(
         _f_create_session, default_create_session
@@ -369,8 +369,8 @@ def _worker_func(
     )
     f_cleanup: T_CLEANUP = get_global_func_with_default_on_worker(_f_cleanup, default_cleanup)
     # Managed resources
-    session: Optional[RPCSession] = None
-    remote_path: Optional[str] = None
+    session: RPCSession | None = None
+    remote_path: str | None = None
 
     @contextmanager
     def resource_handler():
@@ -393,7 +393,7 @@ def resource_handler():
             rt_mod: Module = f_upload_module(session, local_path, remote_path)
         # Step 3: Allocate input arguments
         with Profiler.timeit("RPCRunner/alloc_argument"):
-            repeated_args: List[T_ARGUMENT_LIST] = f_alloc_argument(
+            repeated_args: list[T_ARGUMENT_LIST] = f_alloc_argument(
                 session,
                 device,
                 args_info,
@@ -401,7 +401,7 @@ def resource_handler():
             )
         # Step 4: Run time_evaluator
         with Profiler.timeit("LocalRunner/run_evaluator"):
-            costs: List[float] = f_run_evaluator(
+            costs: list[float] = f_run_evaluator(
                 session,
                 rt_mod,
                 device,
@@ -458,7 +458,7 @@ def default_alloc_argument(
     device: Device,
     args_info: T_ARG_INFO_JSON_OBJ_LIST,
     alloc_repeat: int,
-) -> List[T_ARGUMENT_LIST]:
+) -> list[T_ARGUMENT_LIST]:
     """Default function to allocate the arguments
 
     Parameters
@@ -491,8 +491,8 @@ def default_run_evaluator(
     rt_mod: Module,
     device: Device,
     evaluator_config: EvaluatorConfig,
-    repeated_args: List[T_ARGUMENT_LIST],
-) -> List[float]:
+    repeated_args: list[T_ARGUMENT_LIST],
+) -> list[float]:
     """Default function to run the evaluator
 
     Parameters
@@ -517,8 +517,8 @@ def default_run_evaluator(
 
 
 def default_cleanup(
-    session: Optional[RPCSession],
-    remote_path: Optional[str],
+    session: RPCSession | None,
+    remote_path: str | None,
 ) -> None:
     """Default function to clean up the session
 
diff --git a/python/tvm/s_tir/meta_schedule/runner/runner.py b/python/tvm/s_tir/meta_schedule/runner/runner.py
index 7680143871c0..c2c49bf70f0b 100644
--- a/python/tvm/s_tir/meta_schedule/runner/runner.py
+++ b/python/tvm/s_tir/meta_schedule/runner/runner.py
@@ -17,10 +17,11 @@
 # ruff: noqa: RUF012
 """Runners"""
 
-from typing import Callable, List, Optional, Union
+from collections.abc import Callable
+from typing import Union
 
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 # isort: on
 
@@ -48,13 +49,13 @@ class RunnerInput(Object):
 
     artifact_path: str
     device_type: str
-    args_info: List[ArgInfo]
+    args_info: list[ArgInfo]
 
     def __init__(
         self,
         artifact_path: str,
         device_type: str,
-        args_info: List[ArgInfo],
+        args_info: list[ArgInfo],
     ) -> None:
         """Constructor
 
@@ -87,13 +88,13 @@ class RunnerResult(Object):
         The error message, if any.
     """
 
-    run_secs: Optional[List[float]]
-    error_msg: Optional[str]
+    run_secs: list[float] | None
+    error_msg: str | None
 
     def __init__(
         self,
-        run_secs: Optional[List[float]],
-        error_msg: Optional[str],
+        run_secs: list[float] | None,
+        error_msg: str | None,
     ) -> None:
         """Constructor
 
@@ -121,7 +122,7 @@ class RunnerFuture(Object):
     See also: PyRunnerFuture
     """
 
-    def __init__(self, f_done: Callable, f_result: Optional[Callable] = None) -> None:
+    def __init__(self, f_done: Callable, f_result: Callable | None = None) -> None:
         """Constructor"""
 
         self.__init_handle_by_constructor__(
@@ -172,7 +173,7 @@ class Runner(Object):
 
     RunnerType = Union["Runner", Literal["local", "rpc"]]
 
-    def run(self, runner_inputs: List[RunnerInput]) -> List[RunnerFuture]:
+    def run(self, runner_inputs: list[RunnerInput]) -> list[RunnerFuture]:
         """Run the built artifact and get runner futures.
 
         Parameters
@@ -217,7 +218,7 @@ class _PyRunner(Runner):
     See also: PyRunner
     """
 
-    def __init__(self, f_run: Optional[Callable] = None) -> None:
+    def __init__(self, f_run: Callable | None = None) -> None:
         """Constructor"""
 
         self.__init_handle_by_constructor__(
@@ -239,7 +240,7 @@ class PyRunner:
         "methods": ["run"],
     }
 
-    def run(self, runner_inputs: List[RunnerInput]) -> List[RunnerFuture]:
+    def run(self, runner_inputs: list[RunnerInput]) -> list[RunnerFuture]:
         """Run the built artifact and get runner futures.
 
         Parameters
diff --git a/python/tvm/s_tir/meta_schedule/runner/utils.py b/python/tvm/s_tir/meta_schedule/runner/utils.py
index 78f5534808a9..1a18d8032843 100644
--- a/python/tvm/s_tir/meta_schedule/runner/utils.py
+++ b/python/tvm/s_tir/meta_schedule/runner/utils.py
@@ -17,17 +17,18 @@
 """Runner utility functions"""
 
 import itertools
-from typing import Any, Callable, Dict, List
+from collections.abc import Callable
+from typing import Any
 
 import tvm.runtime
 
 from ....runtime import Device, Module
 from .config import EvaluatorConfig
 
-T_ARG_INFO_JSON_OBJ = List[Any]  # pylint: disable=invalid-name
-T_ARG_INFO_JSON_OBJ_LIST = List[T_ARG_INFO_JSON_OBJ]  # pylint: disable=invalid-name
+T_ARG_INFO_JSON_OBJ = list[Any]  # pylint: disable=invalid-name
+T_ARG_INFO_JSON_OBJ_LIST = list[T_ARG_INFO_JSON_OBJ]  # pylint: disable=invalid-name
 T_ARGUMENT = Any  # pylint: disable=invalid-name
-T_ARGUMENT_LIST = List[T_ARGUMENT]  # pylint: disable=invalid-name
+T_ARGUMENT_LIST = list[T_ARGUMENT]  # pylint: disable=invalid-name
 
 
 def alloc_argument_common(
@@ -35,7 +36,7 @@ def alloc_argument_common(
     device: Device,
     args_info: T_ARG_INFO_JSON_OBJ_LIST,
     alloc_repeat: int,
-) -> List[T_ARGUMENT_LIST]:
+) -> list[T_ARGUMENT_LIST]:
     """Common function to allocate the arguments
 
     Parameters
@@ -63,12 +64,12 @@ def alloc_tensor(_, dtype, shape) -> tvm.runtime.Tensor:
     def alloc_fail(*arg_info) -> None:
         raise NotImplementedError(arg_info)
 
-    dispatcher: Dict[Any, Callable] = {
+    dispatcher: dict[Any, Callable] = {
         "TENSOR": alloc_tensor,
         None: alloc_fail,
     }
 
-    repeated_args: List[T_ARGUMENT_LIST] = []
+    repeated_args: list[T_ARGUMENT_LIST] = []
     for _ in range(alloc_repeat):
         args: T_ARGUMENT_LIST = []
         arg_info: T_ARG_INFO_JSON_OBJ
@@ -84,8 +85,8 @@ def run_evaluator_common(
     rt_mod: Module,
     device: Device,
     evaluator_config: EvaluatorConfig,
-    repeated_args: List[T_ARGUMENT_LIST],
-) -> List[float]:
+    repeated_args: list[T_ARGUMENT_LIST],
+) -> list[float]:
     """Common function to run the evaluator
 
     Parameters
@@ -114,7 +115,7 @@ def run_evaluator_common(
         if evaluator_config.enable_cpu_cache_flush
         else "",
     )
-    repeated_costs: List[List[float]] = []
+    repeated_costs: list[list[float]] = []
     for args in repeated_args:
         device.sync()
         profile_result = evaluator(*args)
diff --git a/python/tvm/s_tir/meta_schedule/schedule/cuda/layout_transform.py b/python/tvm/s_tir/meta_schedule/schedule/cuda/layout_transform.py
index 1b0c39e82150..7f34d81c208b 100644
--- a/python/tvm/s_tir/meta_schedule/schedule/cuda/layout_transform.py
+++ b/python/tvm/s_tir/meta_schedule/schedule/cuda/layout_transform.py
@@ -19,7 +19,6 @@
 
 import math
 from collections import deque
-from typing import List, Optional, Tuple, Union
 
 import tvm
 from tvm.s_tir import meta_schedule
@@ -90,9 +89,9 @@ def tile_layout_transform(
     block_write: SBlockRV,
     src_layout: str,
     dst_layout: str,
-    input_shape: List[int],
+    input_shape: list[int],
     tile_size: ExprRV,
-) -> Tuple[SBlockRV, SBlockRV]:
+) -> tuple[SBlockRV, SBlockRV]:
     """
     High level tiling for layout transform block. Mutates sch in place.
 
@@ -140,7 +139,7 @@ def pad_dimension_to_at_least_number(loop: LoopRV, requested_size: int):
 
     def pad_dimension_to_factor_of_tile_size(
         loop: LoopRV, initial_size: int, tile_size: int = tile_size
-    ) -> Tuple[LoopRV, int]:
+    ) -> tuple[LoopRV, int]:
         """
         Pads loop of given size until it is divisible into tile_size.
         If the given size of the loop is greater than tile size. Do not pad.
@@ -167,8 +166,8 @@ def pad_dimension_to_factor_of_tile_size(
         return pad_dimension_to_at_least_number(loop, size), int(size)
 
     def spin_out_factor(
-        loops: List[LoopRV], loop_extants: List[int], index: int, factor_needed: int
-    ) -> Tuple[List[LoopRV], List[int], int]:
+        loops: list[LoopRV], loop_extants: list[int], index: int, factor_needed: int
+    ) -> tuple[list[LoopRV], list[int], int]:
         """
         Factor out the requested loop's dimensions to reach the requested factor and
         places the requested factor as the innermost loop.
@@ -219,11 +218,11 @@ def spin_out_factor(
         return loops, loop_extants, factor_needed
 
     def factor_dim_in_order(
-        indices: List[int],
-        loops: List[LoopRV],
-        cur_loop_extants: List[int],
+        indices: list[int],
+        loops: list[LoopRV],
+        cur_loop_extants: list[int],
         work_needed_inner_loop: int = tile_size,
-    ) -> Tuple[List[LoopRV], List[int]]:
+    ) -> tuple[list[LoopRV], list[int]]:
         """Factors out the loops in the order of indices until we reach needed work.
 
         Adds new loop factors to the back in reverse order of access. Returns new list
@@ -238,7 +237,7 @@ def factor_dim_in_order(
         return loops, cur_loop_extants
 
     def get_high_level_loop_structure(
-        block_read: SBlockRV, input_shape: List[int], src_layout: str, dst_layout: str
+        block_read: SBlockRV, input_shape: list[int], src_layout: str, dst_layout: str
     ):
         """Runs the factorization described above."""
         # index 0 ... rank - 1 will always correspond to original loops
@@ -329,10 +328,10 @@ def get_high_level_loop_structure(
 def create_cached_read(
     sch: Schedule,
     block_write: SBlockRV,
-    orig_input_shape: List[int],
+    orig_input_shape: list[int],
     orig_src_layout: str,
     orig_dst_layout: str,
-) -> Tuple[SBlockRV, List[int], str, str]:
+) -> tuple[SBlockRV, list[int], str, str]:
     """
     Creates the cached read block with expected structure.
 
@@ -378,7 +377,7 @@ def create_cached_read(
         the new src_layout, and new dst_layout string.
     """
     # Figure out split dimensions, entries are (loop index in src_layout, split amount)
-    split_dimensions: List[Tuple[int, int]] = []
+    split_dimensions: list[tuple[int, int]] = []
 
     # This is without numeric digits, e.g. NCHW4c -> NCHWc
     new_dst_layout = []
@@ -403,8 +402,8 @@ def create_cached_read(
 
     # Calculate final input shapes, each of these are a single element for unsplit dims
     # and tuples for split dims associated with the two new axis
-    input_shape: List[Union[int, Tuple]] = list(orig_input_shape)
-    new_src_layout: List[Union[str, Tuple]] = list(orig_src_layout)
+    input_shape: list[int | tuple] = list(orig_input_shape)
+    new_src_layout: list[str | tuple] = list(orig_src_layout)
     for src_layout_split_index, split_factor in split_dimensions:
         dimension_name = orig_src_layout[src_layout_split_index]
         new_src_layout[src_layout_split_index] = (dimension_name, dimension_name.lower())
@@ -414,8 +413,8 @@ def create_cached_read(
         )
 
     # Unpack any tuples introduced via appending
-    def unpack_list(target_list) -> List:
-        output: List = []
+    def unpack_list(target_list) -> list:
+        output: list = []
         for ele in target_list:
             if isinstance(ele, tuple):
                 output.extend(ele)
@@ -502,8 +501,8 @@ def get_max_tile_size() -> int:
 
 @tvm.register_global_func("s_tir.meta_schedule.cuda.layout_transform")
 def cuda_layout_transform_schedule_rule(
-    sch: Schedule, block: SBlockRV, testing_tile_sizes: Optional[List[int]] = None
-) -> List[Schedule]:
+    sch: Schedule, block: SBlockRV, testing_tile_sizes: list[int] | None = None
+) -> list[Schedule]:
     """
     Applies tiling scheme to layout transform task (potentially fused with other injective funcs).
 
diff --git a/python/tvm/s_tir/meta_schedule/schedule_rule/add_rfactor.py b/python/tvm/s_tir/meta_schedule/schedule_rule/add_rfactor.py
index 0d71d9d17a88..01afa4063ec8 100644
--- a/python/tvm/s_tir/meta_schedule/schedule_rule/add_rfactor.py
+++ b/python/tvm/s_tir/meta_schedule/schedule_rule/add_rfactor.py
@@ -16,8 +16,6 @@
 # under the License.
 """Add-rfactor Rule that add-rfactor to some blocks if needed"""
 
-from typing import Optional
-
 from tvm_ffi import register_object
 
 from .. import _ffi_api
@@ -41,7 +39,7 @@ class AddRFactor(ScheduleRule):
     def __init__(
         self,
         max_jobs_per_core: int = 16,
-        max_innermost_factor: Optional[int] = None,
+        max_innermost_factor: int | None = None,
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.ScheduleRuleAddRFactor,  # type: ignore # pylint: disable=no-member
diff --git a/python/tvm/s_tir/meta_schedule/schedule_rule/auto_bind.py b/python/tvm/s_tir/meta_schedule/schedule_rule/auto_bind.py
index 08a71769c930..49bfd7fabad9 100644
--- a/python/tvm/s_tir/meta_schedule/schedule_rule/auto_bind.py
+++ b/python/tvm/s_tir/meta_schedule/schedule_rule/auto_bind.py
@@ -16,8 +16,6 @@
 # under the License.
 """Auto-bind Rule that binds blocks to threads if needed"""
 
-from typing import List, Optional
-
 from tvm_ffi import register_object
 
 from .. import _ffi_api
@@ -41,7 +39,7 @@ class AutoBind(ScheduleRule):
     def __init__(
         self,
         max_threadblocks: int = 256,
-        thread_extents: Optional[List[int]] = None,
+        thread_extents: list[int] | None = None,
         max_threads_per_block: int = -1,
     ) -> None:
         if thread_extents is None:
diff --git a/python/tvm/s_tir/meta_schedule/schedule_rule/auto_inline.py b/python/tvm/s_tir/meta_schedule/schedule_rule/auto_inline.py
index 573093962f10..47f17d913e71 100644
--- a/python/tvm/s_tir/meta_schedule/schedule_rule/auto_inline.py
+++ b/python/tvm/s_tir/meta_schedule/schedule_rule/auto_inline.py
@@ -16,8 +16,6 @@
 # under the License.
 """Auto-Inline. Rule that inlines spatial blocks if it satisfies some conditions"""
 
-from typing import List, Optional
-
 from tvm_ffi import register_object
 
 from .. import _ffi_api
@@ -54,7 +52,7 @@ def __init__(
         disallow_if_then_else: bool,
         require_injective: bool,
         require_ordered: bool,
-        disallow_op: Optional[List[str]] = None,
+        disallow_op: list[str] | None = None,
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.ScheduleRuleAutoInline,  # type: ignore # pylint: disable=no-member
diff --git a/python/tvm/s_tir/meta_schedule/schedule_rule/cross_thread_reduction.py b/python/tvm/s_tir/meta_schedule/schedule_rule/cross_thread_reduction.py
index b3366c63fbbc..a87192c770f2 100644
--- a/python/tvm/s_tir/meta_schedule/schedule_rule/cross_thread_reduction.py
+++ b/python/tvm/s_tir/meta_schedule/schedule_rule/cross_thread_reduction.py
@@ -16,8 +16,6 @@
 # under the License.
 """Rules which apply cross-thread reduction to some reduction blocks correspondingly when needed"""
 
-from typing import List
-
 from tvm_ffi import register_object
 
 from .. import _ffi_api
@@ -35,7 +33,7 @@ class CrossThreadReduction(ScheduleRule):
         Candidates of thread axis extent (values are required to be positive).
     """
 
-    def __init__(self, thread_extents: List[int]) -> None:
+    def __init__(self, thread_extents: list[int]) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.ScheduleRuleCrossThreadReduction,  # type: ignore # pylint: disable=no-member
             thread_extents,
diff --git a/python/tvm/s_tir/meta_schedule/schedule_rule/multi_level_tiling.py b/python/tvm/s_tir/meta_schedule/schedule_rule/multi_level_tiling.py
index 00acda8b9134..59501f16ba1c 100644
--- a/python/tvm/s_tir/meta_schedule/schedule_rule/multi_level_tiling.py
+++ b/python/tvm/s_tir/meta_schedule/schedule_rule/multi_level_tiling.py
@@ -16,7 +16,8 @@
 # under the License.
 """Multi-level tiling with reuse."""
 
-from typing import Any, Callable, Dict, List, Mapping, NamedTuple, Optional
+from collections.abc import Callable, Mapping
+from typing import Any, NamedTuple
 
 from tvm_ffi import register_object
 
@@ -30,10 +31,10 @@ class ReuseType(NamedTuple):
     """Reuse type."""
 
     req: str
-    levels: List[int]
+    levels: list[int]
     scope: str
 
-    def as_dict(self) -> Dict[str, Any]:
+    def as_dict(self) -> dict[str, Any]:
         """Return the dict representation of the reuse type."""
         return {
             "req": self.req,
@@ -75,12 +76,12 @@ class MultiLevelTiling(ScheduleRule):
     def __init__(
         self,
         structure: str,
-        tile_binds: Optional[List[str]] = None,
-        max_innermost_factor: Optional[int] = None,
-        vector_load_lens: Optional[List[int]] = None,
-        reuse_read: Optional[ReuseType] = None,
-        reuse_write: Optional[ReuseType] = None,
-        filter_fn: Optional[Callable[[Schedule, SBlockRV], bool]] = None,
+        tile_binds: list[str] | None = None,
+        max_innermost_factor: int | None = None,
+        vector_load_lens: list[int] | None = None,
+        reuse_read: ReuseType | None = None,
+        reuse_write: ReuseType | None = None,
+        filter_fn: Callable[[Schedule, SBlockRV], bool] | None = None,
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.ScheduleRuleMultiLevelTiling,  # type: ignore # pylint: disable=no-member
@@ -125,11 +126,11 @@ def __init__(
         self,
         intrin_name: str,
         structure: str,
-        tile_binds: Optional[List[str]] = None,
-        max_innermost_factor: Optional[int] = None,
-        vector_load_lens: Optional[List[int]] = None,
-        reuse_read: Optional[ReuseType] = None,
-        reuse_write: Optional[ReuseType] = None,
+        tile_binds: list[str] | None = None,
+        max_innermost_factor: int | None = None,
+        vector_load_lens: list[int] | None = None,
+        reuse_read: ReuseType | None = None,
+        reuse_write: ReuseType | None = None,
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.ScheduleRuleMultiLevelTilingWithIntrin,  # type: ignore # pylint: disable=no-member
@@ -177,13 +178,13 @@ class MultiLevelTilingTensorCore(ScheduleRule):
 
     def __init__(
         self,
-        intrin_groups: List[Mapping[str, str]],
+        intrin_groups: list[Mapping[str, str]],
         structure: str,
-        tile_binds: Optional[List[str]] = None,
-        max_innermost_factor: Optional[int] = None,
-        vector_load_lens: Optional[List[int]] = None,
-        reuse_read: Optional[ReuseType] = None,
-        reuse_write: Optional[ReuseType] = None,
+        tile_binds: list[str] | None = None,
+        max_innermost_factor: int | None = None,
+        vector_load_lens: list[int] | None = None,
+        reuse_read: ReuseType | None = None,
+        reuse_write: ReuseType | None = None,
         use_software_pipeline: bool = False,
     ) -> None:
         self.__init_handle_by_constructor__(
@@ -222,9 +223,9 @@ def __init__(
         self,
         structure: str,
         vector_length_in_bits: int,
-        max_innermost_factor: Optional[int] = None,
-        reuse_read: Optional[ReuseType] = None,
-        reuse_write: Optional[ReuseType] = None,
+        max_innermost_factor: int | None = None,
+        reuse_read: ReuseType | None = None,
+        reuse_write: ReuseType | None = None,
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.ScheduleRuleMultiLevelTilingWideVector,  # type: ignore # pylint: disable=no-member
diff --git a/python/tvm/s_tir/meta_schedule/schedule_rule/parallel_vectorize_unroll.py b/python/tvm/s_tir/meta_schedule/schedule_rule/parallel_vectorize_unroll.py
index 089a6c0678c8..8ec460babd33 100644
--- a/python/tvm/s_tir/meta_schedule/schedule_rule/parallel_vectorize_unroll.py
+++ b/python/tvm/s_tir/meta_schedule/schedule_rule/parallel_vectorize_unroll.py
@@ -17,8 +17,6 @@
 """Rule that mark parallelize, vectorize and unroll to the root block. The mark will be applied to
 each block in a follow-up post processor"""
 
-from typing import List, Optional
-
 from tvm_ffi import register_object
 
 from .. import _ffi_api
@@ -51,7 +49,7 @@ def __init__(
         self,
         max_jobs_per_core: int = 16,
         max_vectorize_extent: int = 16,
-        unroll_max_steps: Optional[List[int]] = None,
+        unroll_max_steps: list[int] | None = None,
         unroll_explicit: bool = True,
     ) -> None:
         if unroll_max_steps is None:
diff --git a/python/tvm/s_tir/meta_schedule/schedule_rule/schedule_rule.py b/python/tvm/s_tir/meta_schedule/schedule_rule/schedule_rule.py
index 923e6d7f56fb..127579d2a5c0 100644
--- a/python/tvm/s_tir/meta_schedule/schedule_rule/schedule_rule.py
+++ b/python/tvm/s_tir/meta_schedule/schedule_rule/schedule_rule.py
@@ -20,10 +20,11 @@
 blocks in a schedule. See also PostOrderApply.
 """
 
-from typing import TYPE_CHECKING, Callable, List, Optional
+from collections.abc import Callable
+from typing import TYPE_CHECKING
 
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 # isort: on
 
@@ -55,7 +56,7 @@ def _initialize_with_tune_context(self, context: "TuneContext") -> None:
             self, context
         )
 
-    def apply(self, sch: Schedule, block: SBlockRV) -> List[Schedule]:
+    def apply(self, sch: Schedule, block: SBlockRV) -> list[Schedule]:
         """Apply a schedule rule to the specific block in the given schedule.
 
         Parameters
@@ -85,7 +86,7 @@ def clone(self) -> "ScheduleRule":
         return _ffi_api.ScheduleRuleClone(self)  # type: ignore # pylint: disable=no-member
 
     @staticmethod
-    def create(kind: Literal["llvm", "cuda", "cuda-tensorcore", "hexagon"]) -> List["ScheduleRule"]:
+    def create(kind: Literal["llvm", "cuda", "cuda-tensorcore", "hexagon"]) -> list["ScheduleRule"]:
         """Create a list of schedule rules for the given kind.
 
         Parameters
@@ -126,10 +127,10 @@ class _PyScheduleRule(ScheduleRule):
 
     def __init__(
         self,
-        f_initialize_with_tune_context: Optional[Callable] = None,
-        f_apply: Optional[Callable] = None,
-        f_clone: Optional[Callable] = None,
-        f_as_string: Optional[Callable] = None,
+        f_initialize_with_tune_context: Callable | None = None,
+        f_apply: Callable | None = None,
+        f_clone: Callable | None = None,
+        f_as_string: Callable | None = None,
     ):
         """Constructor."""
 
@@ -165,7 +166,7 @@ def _initialize_with_tune_context(self, context: "TuneContext") -> None:
         """
         raise NotImplementedError
 
-    def apply(self, sch: Schedule, block: SBlockRV) -> List[Schedule]:
+    def apply(self, sch: Schedule, block: SBlockRV) -> list[Schedule]:
         """Apply a schedule rule to the specific block in the given schedule.
 
         Parameters
diff --git a/python/tvm/s_tir/meta_schedule/search_strategy/search_strategy.py b/python/tvm/s_tir/meta_schedule/search_strategy/search_strategy.py
index b82feffbc44e..b03208783e64 100644
--- a/python/tvm/s_tir/meta_schedule/search_strategy/search_strategy.py
+++ b/python/tvm/s_tir/meta_schedule/search_strategy/search_strategy.py
@@ -20,10 +20,11 @@
 candidates for measurement.
 """
 
-from typing import TYPE_CHECKING, Callable, List, Optional, Union
+from collections.abc import Callable
+from typing import TYPE_CHECKING, Optional, Union
 
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 # isort: on
 from tvm_ffi import register_object
@@ -54,12 +55,12 @@ class MeasureCandidate(Object):
     """
 
     sch: Schedule
-    args_info: List[ArgInfo]
+    args_info: list[ArgInfo]
 
     def __init__(
         self,
         sch: Schedule,
-        args_info: List[ArgInfo],
+        args_info: list[ArgInfo],
     ) -> None:
         """Constructor.
 
@@ -121,7 +122,7 @@ def pre_tuning(
         self,
         max_trials: int,
         num_trials_per_iter: int,
-        design_spaces: List[Schedule],
+        design_spaces: list[Schedule],
         database: Optional["Database"] = None,
         cost_model: Optional["CostModel"] = None,
     ) -> None:
@@ -153,7 +154,7 @@ def post_tuning(self) -> None:
         """Post-tuning for the search strategy."""
         _ffi_api.SearchStrategyPostTuning(self)  # type: ignore # pylint: disable=no-member
 
-    def generate_measure_candidates(self) -> Optional[List[MeasureCandidate]]:
+    def generate_measure_candidates(self) -> list[MeasureCandidate] | None:
         """Generate measure candidates from design spaces for measurement.
 
         Returns
@@ -165,8 +166,8 @@ def generate_measure_candidates(self) -> Optional[List[MeasureCandidate]]:
 
     def notify_runner_results(
         self,
-        measure_candidates: List[MeasureCandidate],
-        results: List[RunnerResult],
+        measure_candidates: list[MeasureCandidate],
+        results: list[RunnerResult],
     ) -> None:
         """Update the search strategy with profiling results.
 
@@ -233,12 +234,12 @@ class _PySearchStrategy(SearchStrategy):
 
     def __init__(
         self,
-        f_initialize_with_tune_context: Optional[Callable] = None,
-        f_pre_tuning: Optional[Callable] = None,
-        f_post_tuning: Optional[Callable] = None,
-        f_generate_measure_candidates: Optional[Callable] = None,
-        f_notify_runner_results: Optional[Callable] = None,
-        f_clone: Optional[Callable] = None,
+        f_initialize_with_tune_context: Callable | None = None,
+        f_pre_tuning: Callable | None = None,
+        f_post_tuning: Callable | None = None,
+        f_generate_measure_candidates: Callable | None = None,
+        f_notify_runner_results: Callable | None = None,
+        f_clone: Callable | None = None,
     ):
         """Constructor."""
 
@@ -287,7 +288,7 @@ def pre_tuning(
         self,
         max_trials: int,
         num_trials_per_iter: int,
-        design_spaces: List[Schedule],
+        design_spaces: list[Schedule],
         database: Optional["Database"] = None,
         cost_model: Optional["CostModel"] = None,
     ) -> None:
@@ -304,7 +305,7 @@ def post_tuning(self) -> None:
         """Post-tuning for the search strategy."""
         raise NotImplementedError
 
-    def generate_measure_candidates(self) -> Optional[List[MeasureCandidate]]:
+    def generate_measure_candidates(self) -> list[MeasureCandidate] | None:
         """Generate measure candidates from design spaces for measurement.
 
         Returns
@@ -316,8 +317,8 @@ def generate_measure_candidates(self) -> Optional[List[MeasureCandidate]]:
 
     def notify_runner_results(
         self,
-        measure_candidates: List[MeasureCandidate],
-        results: List[RunnerResult],
+        measure_candidates: list[MeasureCandidate],
+        results: list[RunnerResult],
     ) -> None:
         """Update the search strategy with profiling results.
 
diff --git a/python/tvm/s_tir/meta_schedule/space_generator/space_generator.py b/python/tvm/s_tir/meta_schedule/space_generator/space_generator.py
index ceee06cd00d9..73068a0d4214 100644
--- a/python/tvm/s_tir/meta_schedule/space_generator/space_generator.py
+++ b/python/tvm/s_tir/meta_schedule/space_generator/space_generator.py
@@ -20,10 +20,11 @@
 space for generation of measure candidates.
 """
 
-from typing import TYPE_CHECKING, Callable, Dict, List, Optional, Tuple, Union
+from collections.abc import Callable
+from typing import TYPE_CHECKING, Union
 
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 # isort: on
 from tvm_ffi import register_object
@@ -45,11 +46,11 @@
 class SpaceGenerator(Object):
     """The abstract design space generator interface."""
 
-    ScheduleFnType = Union[
-        Callable[[Schedule], None],  # No output
-        Callable[[Schedule], Schedule],  # Single output
-        Callable[[Schedule], List[Schedule]],  # Multiple outputs
-    ]
+    ScheduleFnType = (
+        Callable[[Schedule], None]  # No output
+        | Callable[[Schedule], Schedule]  # Single output
+        | Callable[[Schedule], list[Schedule]]  # Multiple outputs
+    )
 
     SpaceGeneratorType = Union[
         "SpaceGenerator",
@@ -57,9 +58,9 @@ class SpaceGenerator(Object):
         Literal["post-order-apply", "union"],
     ]
 
-    sch_rules: Optional[List["ScheduleRule"]]
-    postprocs: Optional[List["Postproc"]]
-    mutator_probs: Optional[Dict["Mutator", float]]
+    sch_rules: list["ScheduleRule"] | None
+    postprocs: list["Postproc"] | None
+    mutator_probs: dict["Mutator", float] | None
 
     def _initialize_with_tune_context(self, context: "TuneContext") -> None:
         """Initialize the design space generator with tuning context.
@@ -73,7 +74,7 @@ def _initialize_with_tune_context(self, context: "TuneContext") -> None:
             self, context
         )
 
-    def generate_design_space(self, mod: IRModule) -> List[Schedule]:
+    def generate_design_space(self, mod: IRModule) -> list[Schedule]:
         """Generate design spaces given a module.
 
         Parameters
@@ -100,10 +101,7 @@ def clone(self) -> "SpaceGenerator":
 
     @staticmethod
     def create(  # pylint: disable=keyword-arg-before-vararg
-        kind: Union[
-            Literal["post-order-apply", "union"],
-            ScheduleFnType,
-        ] = "post-order-apply",
+        kind: Literal["post-order-apply", "union"] | ScheduleFnType = "post-order-apply",
         *args,
         **kwargs,
     ) -> "SpaceGenerator":
@@ -135,18 +133,15 @@ def create_schedule_fn(
 
 
 ScheduleFnType = SpaceGenerator.ScheduleFnType
-ScheduleRuleType = Union[
-    List["ScheduleRule"],
-    Literal["llvm", "cuda", "cuda-tensorcore", "hexagon", "from-target"],
-]
-PostprocType = Union[
-    List["Postproc"],
-    Literal["llvm", "cuda", "cuda-tensorcore", "hexagon", "from-target"],
-]
-MutatorProbType = Union[
-    Dict["Mutator", float],
-    Literal["llvm", "cuda", "cuda-tensorcore", "hexagon", "from-target"],
-]
+ScheduleRuleType = (
+    list["ScheduleRule"] | Literal["llvm", "cuda", "cuda-tensorcore", "hexagon", "from-target"]
+)
+PostprocType = (
+    list["Postproc"] | Literal["llvm", "cuda", "cuda-tensorcore", "hexagon", "from-target"]
+)
+MutatorProbType = (
+    dict["Mutator", float] | Literal["llvm", "cuda", "cuda-tensorcore", "hexagon", "from-target"]
+)
 create = SpaceGenerator.create  # pylint: disable=invalid-name
 
 
@@ -154,10 +149,10 @@ def _normalize_rules(
     sch_rules: ScheduleRuleType,
     postprocs: PostprocType,
     mutator_probs: MutatorProbType,
-) -> Tuple[
-    Optional[List["ScheduleRule"]],
-    Optional[List["Postproc"]],
-    Optional[Dict["Mutator", float]],
+) -> tuple[
+    list["ScheduleRule"] | None,
+    list["Postproc"] | None,
+    dict["Mutator", float] | None,
 ]:
     # pylint: disable=import-outside-toplevel
     from ..mutator import Mutator
@@ -201,9 +196,9 @@ def __init__(
         sch_rules: ScheduleRuleType = "from-target",
         postprocs: PostprocType = "from-target",
         mutator_probs: MutatorProbType = "from-target",
-        f_initialize_with_tune_context: Optional[Callable] = None,
-        f_generate_design_space: Optional[Callable] = None,
-        f_clone: Optional[Callable] = None,
+        f_initialize_with_tune_context: Callable | None = None,
+        f_generate_design_space: Callable | None = None,
+        f_clone: Callable | None = None,
     ):
         """Constructor."""
         sch_rules, postprocs, mutator_probs = _normalize_rules(sch_rules, postprocs, mutator_probs)
@@ -243,7 +238,7 @@ def _initialize_with_tune_context(self, context: "TuneContext") -> None:
         """
         raise NotImplementedError
 
-    def generate_design_space(self, mod: IRModule) -> List[Schedule]:
+    def generate_design_space(self, mod: IRModule) -> list[Schedule]:
         """Generate design spaces given a module.
 
         Parameters
diff --git a/python/tvm/s_tir/meta_schedule/space_generator/space_generator_union.py b/python/tvm/s_tir/meta_schedule/space_generator/space_generator_union.py
index 0e4fa0c7430f..097822646fd0 100644
--- a/python/tvm/s_tir/meta_schedule/space_generator/space_generator_union.py
+++ b/python/tvm/s_tir/meta_schedule/space_generator/space_generator_union.py
@@ -16,8 +16,6 @@
 # under the License.
 """Union of meta Schedule design space generators."""
 
-from typing import List
-
 from tvm_ffi import register_object
 
 from .. import _ffi_api
@@ -36,7 +34,7 @@ class SpaceGeneratorUnion(SpaceGenerator):
 
     def __init__(
         self,
-        space_generators: List[SpaceGenerator],
+        space_generators: list[SpaceGenerator],
         sch_rules: ScheduleRuleType = "from-target",
         postprocs: PostprocType = "from-target",
         mutator_probs: MutatorProbType = "from-target",
diff --git a/python/tvm/s_tir/meta_schedule/task_scheduler/task_scheduler.py b/python/tvm/s_tir/meta_schedule/task_scheduler/task_scheduler.py
index 7e32a50bbf6e..1510d021a1c2 100644
--- a/python/tvm/s_tir/meta_schedule/task_scheduler/task_scheduler.py
+++ b/python/tvm/s_tir/meta_schedule/task_scheduler/task_scheduler.py
@@ -17,10 +17,11 @@
 # ruff: noqa: RUF012
 """Auto-tuning Task Scheduler"""
 
-from typing import Callable, List, Optional, Union
+from collections.abc import Callable
+from typing import Union
 
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 # isort: on
 
@@ -51,19 +52,19 @@ class TaskRecord(Object):
     is_terminated: bool
     build_error_count: int
     run_error_count: int
-    measure_candidates: List[MeasureCandidate]
-    builder_results: List[BuilderResult]
-    runner_results: List[RunnerResult]
+    measure_candidates: list[MeasureCandidate]
+    builder_results: list[BuilderResult]
+    runner_results: list[RunnerResult]
 
 
 @register_object("s_tir.meta_schedule.TaskScheduler")
 class TaskScheduler(Object):
     """The abstract task scheduler interface."""
 
-    tasks_: List[TaskRecord]
-    measure_callbacks_: List[MeasureCallback]
-    database_: Optional[Database]
-    cost_model_: Optional[CostModel]
+    tasks_: list[TaskRecord]
+    measure_callbacks_: list[MeasureCallback]
+    database_: Database | None
+    cost_model_: CostModel | None
     remaining_tasks_: int
 
     TaskSchedulerType = Union["TaskScheduler", Literal["gradient", "round-robin"]]
@@ -78,7 +79,7 @@ def next_task_id(self) -> int:
         """
         return _ffi_api.TaskSchedulerNextTaskId(self)  # type: ignore # pylint: disable=no-member
 
-    def join_running_task(self, task_id: int) -> List[RunnerResult]:
+    def join_running_task(self, task_id: int) -> list[RunnerResult]:
         """Wait until the task is finished.
 
         Parameters
@@ -95,16 +96,16 @@ def join_running_task(self, task_id: int) -> List[RunnerResult]:
 
     def tune(
         self,
-        tasks: List[TuneContext],
-        task_weights: List[float],
+        tasks: list[TuneContext],
+        task_weights: list[float],
         max_trials_global: int,
         max_trials_per_task: int,
         num_trials_per_iter: int,
         builder: Builder,
         runner: Runner,
-        measure_callbacks: List[MeasureCallback],
-        database: Optional[Database],
-        cost_model: Optional[CostModel],
+        measure_callbacks: list[MeasureCallback],
+        database: Database | None,
+        cost_model: CostModel | None,
     ) -> None:
         """Auto-tuning.
 
@@ -236,15 +237,15 @@ def __init__(self): ...
 
     def tune(
         self,
-        tasks: List[TuneContext],
-        task_weights: List[float],
+        tasks: list[TuneContext],
+        task_weights: list[float],
         max_trials_global: int,
         max_trials_per_task: int,
         builder: Builder,
         runner: Runner,
-        measure_callbacks: List[MeasureCallback],
-        database: Optional[Database],
-        cost_model: Optional[CostModel],
+        measure_callbacks: list[MeasureCallback],
+        database: Database | None,
+        cost_model: CostModel | None,
     ) -> None:
         """Auto-tuning."""
         # Using self._outer to replace the self pointer
@@ -271,7 +272,7 @@ def next_task_id(self) -> int:
         """
         raise NotImplementedError
 
-    def join_running_task(self, task_id: int) -> List[RunnerResult]:
+    def join_running_task(self, task_id: int) -> list[RunnerResult]:
         """Wait until the task is finished.
 
         Parameters
diff --git a/python/tvm/s_tir/meta_schedule/testing/custom_builder_runner.py b/python/tvm/s_tir/meta_schedule/testing/custom_builder_runner.py
index ba04cf2794f7..ef193b338b54 100644
--- a/python/tvm/s_tir/meta_schedule/testing/custom_builder_runner.py
+++ b/python/tvm/s_tir/meta_schedule/testing/custom_builder_runner.py
@@ -17,7 +17,7 @@
 """Customized builder and runner methods"""
 # pylint: disable=import-outside-toplevel
 
-from typing import Callable, Dict, Union
+from collections.abc import Callable
 
 import numpy as np  # type: ignore
 
@@ -27,9 +27,9 @@
 
 def run_module_via_rpc(
     rpc_config: RPCConfig,
-    lib: Union[Module, Executable],
+    lib: Module | Executable,
     dev_type: str,
-    args: Union[Dict[int, np.ndarray], Dict[str, np.ndarray]],
+    args: dict[int, np.ndarray] | dict[str, np.ndarray],
     continuation: Callable,
 ):
     """Execute a tvm.runtime.Module on RPC remote"""
diff --git a/python/tvm/s_tir/meta_schedule/testing/dummy_object.py b/python/tvm/s_tir/meta_schedule/testing/dummy_object.py
index 489c513a2305..007de8a9de0a 100644
--- a/python/tvm/s_tir/meta_schedule/testing/dummy_object.py
+++ b/python/tvm/s_tir/meta_schedule/testing/dummy_object.py
@@ -17,7 +17,6 @@
 """Dummy objects for testing."""
 
 import random
-from typing import List, Optional
 
 from tvm.s_tir.schedule import Trace
 
@@ -40,13 +39,13 @@ def result(self) -> RunnerResult:
 
 @derived_object
 class DummyBuilder(PyBuilder):
-    def build(self, build_inputs: List[BuilderInput]) -> List[BuilderResult]:
+    def build(self, build_inputs: list[BuilderInput]) -> list[BuilderResult]:
         return [BuilderResult("test_path", None) for _ in build_inputs]
 
 
 @derived_object
 class DummyRunner(PyRunner):
-    def run(self, runner_inputs: List[RunnerInput]) -> List[RunnerFuture]:
+    def run(self, runner_inputs: list[RunnerInput]) -> list[RunnerFuture]:
         return [DummyRunnerFuture() for _ in runner_inputs]  # type: ignore
 
 
@@ -57,7 +56,7 @@ class DummyMutator(PyMutator):
     def _initialize_with_tune_context(self, context: "TuneContext") -> None:
         pass
 
-    def apply(self, trace: Trace, _) -> Optional[Trace]:
+    def apply(self, trace: Trace, _) -> Trace | None:
         return Trace(trace.insts, {})
 
     def clone(self):
diff --git a/python/tvm/s_tir/meta_schedule/testing/space_generation.py b/python/tvm/s_tir/meta_schedule/testing/space_generation.py
index 444e9b15833c..b2a5046f65ec 100644
--- a/python/tvm/s_tir/meta_schedule/testing/space_generation.py
+++ b/python/tvm/s_tir/meta_schedule/testing/space_generation.py
@@ -15,10 +15,9 @@
 # specific language governing permissions and limitations
 # under the License.
 # pylint: disable=missing-module-docstring,missing-function-docstring,missing-class-docstring
-from typing import List, Optional, Tuple, Union
 
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 # isort: on
 
@@ -32,8 +31,8 @@
 
 def get_rules(
     kind: Literal["llvm", "cuda", "cuda-tensorcore", "hexagon"],
-    types: Union[type, Tuple[type, ...]],
-) -> List[ms.ScheduleRule]:
+    types: type | tuple[type, ...],
+) -> list[ms.ScheduleRule]:
     """Get default schedule rules"""
     rules = ms.ScheduleRule.create(kind)
     return [rule for rule in rules if isinstance(rule, types)]
@@ -59,9 +58,9 @@ def generate_design_space(
     kind: Literal["llvm", "cuda", "cuda-tensorcore", "hexagon"],
     mod: IRModule,
     target: Target,
-    types: Union[type, Tuple[type, ...]],
-    sch_rules: Optional[List[ms.ScheduleRule]] = None,
-) -> List[Schedule]:
+    types: type | tuple[type, ...],
+    sch_rules: list[ms.ScheduleRule] | None = None,
+) -> list[Schedule]:
     if sch_rules is None:
         sch_rules = get_rules(kind, types)
     else:
@@ -80,12 +79,12 @@ def generate_design_space(
 
 def _find_match_sketch_id(
     mod: IRModule,
-    sketches: List[Schedule],
+    sketches: list[Schedule],
     expected_mod: IRModule,
-    expected_decision: List[Tuple[str, List[int]]],
+    expected_decision: list[tuple[str, list[int]]],
     *,
     debug_mask="all",
-) -> Optional[int]:
+) -> int | None:
     for sketch_id, sketch in enumerate(sketches):
         i = 0
         new_decisions = {}
@@ -111,9 +110,9 @@ def _find_match_sketch_id(
 
 def check_sketches(
     mod: IRModule,
-    sketches: List[Schedule],
-    expected_mods: List[IRModule],
-    expected_decisions: List[List[Tuple[str, List[int]]]],
+    sketches: list[Schedule],
+    expected_mods: list[IRModule],
+    expected_decisions: list[list[tuple[str, list[int]]]],
     *,
     debug_mask="all",
 ):
@@ -140,7 +139,7 @@ def check_sketches(
         sketches.pop(sketch_id)
 
 
-def print_sketches(sketches: List[Schedule]):
+def print_sketches(sketches: list[Schedule]):
     for i, sch in enumerate(sketches):
         print(f"###### {i}")
         sch.mod.show(black_format=False)
diff --git a/python/tvm/s_tir/meta_schedule/testing/te_workload.py b/python/tvm/s_tir/meta_schedule/testing/te_workload.py
index 8db117f26445..ae40249d2799 100644
--- a/python/tvm/s_tir/meta_schedule/testing/te_workload.py
+++ b/python/tvm/s_tir/meta_schedule/testing/te_workload.py
@@ -18,7 +18,6 @@
 """Workloads in TE"""
 
 # pylint: disable=missing-docstring
-from typing import Tuple
 
 from tvm import te, tir, topi
 from tvm.target import Target
@@ -31,7 +30,7 @@ def batch_matmul_nkkm(  # pylint: disable=invalid-name,missing-docstring
     K: int,
     in_dtype: str = "float32",
     out_dtype: str = "float32",
-) -> Tuple[te.Tensor, te.Tensor, te.Tensor]:
+) -> tuple[te.Tensor, te.Tensor, te.Tensor]:
     x = te.placeholder((B, N, K), name="X", dtype=in_dtype)
     y = te.placeholder((B, K, M), name="Y", dtype=in_dtype)
     k = te.reduce_axis((0, K), name="k")
@@ -58,7 +57,7 @@ def conv1d_nlc(  # pylint: disable=invalid-name,missing-docstring
     groups: int = 1,
     in_dtype: str = "float32",
     out_dtype: str = "float32",
-) -> Tuple[te.Tensor, te.Tensor, te.Tensor]:
+) -> tuple[te.Tensor, te.Tensor, te.Tensor]:
     inputs = te.placeholder((N, L, CI), name="inputs", dtype=in_dtype)
     weight = te.placeholder((kernel_size, CI // groups, CO), name="weight", dtype=in_dtype)
 
@@ -101,7 +100,7 @@ def conv2d_nhwc(  # pylint: disable=invalid-name,missing-docstring
     groups: int = 1,
     in_dtype: str = "float32",
     out_dtype: str = "float32",
-) -> Tuple[te.Tensor, te.Tensor, te.Tensor]:
+) -> tuple[te.Tensor, te.Tensor, te.Tensor]:
     inputs = te.placeholder((N, H, W, CI), name="inputs", dtype=in_dtype)
     weight = te.placeholder(
         (kernel_size, kernel_size, CI // groups, CO), name="weight", dtype=in_dtype
@@ -150,7 +149,7 @@ def conv3d_ndhwc(  # pylint: disable=invalid-name,missing-docstring
     groups: int = 1,
     in_dtype: str = "float32",
     out_dtype: str = "float32",
-) -> Tuple[te.Tensor, te.Tensor, te.Tensor]:
+) -> tuple[te.Tensor, te.Tensor, te.Tensor]:
     inputs = te.placeholder((N, D, H, W, CI), name="inputs", dtype=in_dtype)
     weight = te.placeholder(
         (kernel_size, kernel_size, kernel_size, CI // groups, CO), name="weight", dtype=in_dtype
@@ -200,7 +199,7 @@ def depthwise_conv2d_nhwc(  # pylint: disable=invalid-name,missing-docstring
     factor: int = 1,
     in_dtype: str = "float32",
     out_dtype: str = "float32",
-) -> Tuple[te.Tensor, te.Tensor, te.Tensor]:
+) -> tuple[te.Tensor, te.Tensor, te.Tensor]:
     inputs = te.placeholder((N, H, W, C), dtype=in_dtype)
     weight = te.placeholder((factor, kernel_size, kernel_size, C), dtype=in_dtype)
     batch_size, in_h, in_w, in_channel = inputs.shape
@@ -242,7 +241,7 @@ def conv2d_transpose_nhwc(  # pylint: disable=invalid-name,missing-docstring
     padding: int = 0,
     in_dtype: str = "float32",
     out_dtype: str = "float32",
-) -> Tuple[te.Tensor, te.Tensor, te.Tensor]:
+) -> tuple[te.Tensor, te.Tensor, te.Tensor]:
     inputs = te.placeholder((N, H, W, CI), name="inputs", dtype=in_dtype)
     weight = te.placeholder((kernel_size, kernel_size, CI, CO), name="weight", dtype=in_dtype)
 
@@ -333,7 +332,7 @@ def conv2d_capsule_nhwijc(  # pylint: disable=invalid-name,missing-docstring
     capsule_size: int = 4,
     in_dtype: str = "float32",
     out_dtype: str = "float32",
-) -> Tuple[te.Tensor, te.Tensor, te.Tensor]:
+) -> tuple[te.Tensor, te.Tensor, te.Tensor]:
     inputs = te.placeholder(
         (N, H, W, capsule_size, capsule_size, CI), name="inputs", dtype=in_dtype
     )
@@ -372,7 +371,7 @@ def norm_bmn(  # pylint: disable=invalid-name,missing-docstring
     B: int,
     M: int,
     N: int,
-) -> Tuple[te.Tensor, te.Tensor]:
+) -> tuple[te.Tensor, te.Tensor]:
     a = te.placeholder((B, M, N), name="A")
     i = te.reduce_axis((0, M), name="i")
     j = te.reduce_axis((0, N), name="j")
@@ -459,7 +458,7 @@ def conv2d_nhwc_bn_relu(  # pylint: disable=invalid-name,missing-docstring
     dilation: int = 1,
     in_dtype: str = "float32",
     out_dtype: str = "float32",
-) -> Tuple[te.Tensor, te.Tensor, te.Tensor, te.Tensor, te.Tensor, te.Tensor]:
+) -> tuple[te.Tensor, te.Tensor, te.Tensor, te.Tensor, te.Tensor, te.Tensor]:
     data = te.placeholder((N, H, W, CI), name="data", dtype=in_dtype)
     kernel = te.placeholder((kernel_size, kernel_size, CI, CO), name="kernel", dtype=in_dtype)
     bias = te.placeholder((CO,), name="bias")
@@ -488,7 +487,7 @@ def transpose_batch_matmul(  # pylint: disable=invalid-name,missing-docstring
     n_dim: int,
     in_dtype: str = "float32",
     out_dtype: str = "float32",
-) -> Tuple[te.Tensor, te.Tensor, te.Tensor]:
+) -> tuple[te.Tensor, te.Tensor, te.Tensor]:
     query = te.placeholder((batch, seq_len, n_head, n_dim), name="query", dtype=in_dtype)
     value = te.placeholder((batch, seq_len, n_head, n_dim), name="value", dtype=in_dtype)
     query_T = te.compute(
@@ -523,7 +522,7 @@ def conv2d_winograd_nhwc(  # pylint: disable=invalid-name,missing-docstring
     padding: int = 0,
     dilation: int = 1,
     tile_size: int = 4,
-) -> Tuple[te.Tensor, te.Tensor, te.Tensor]:
+) -> tuple[te.Tensor, te.Tensor, te.Tensor]:
     from tvm.topi.nn.conv2d import (  # pylint: disable=import-outside-toplevel
         _conv2d_winograd_nhwc_impl,
     )
@@ -553,7 +552,7 @@ def conv2d_winograd_nhwc(  # pylint: disable=invalid-name,missing-docstring
 
 def matmul(
     n: int, m: int, k: int, in_dtype: str = "float32", out_dtype: str = "float32"
-) -> Tuple[te.Tensor, te.Tensor, te.Tensor]:
+) -> tuple[te.Tensor, te.Tensor, te.Tensor]:
     a = te.placeholder((n, k), name="A", dtype=in_dtype)
     b = te.placeholder((k, m), name="B", dtype=in_dtype)
     k = te.reduce_axis((0, k), name="k")
@@ -567,7 +566,7 @@ def matmul(
 
 def matmul_relu(
     n: int, m: int, k: int, in_dtype: str = "float32", out_dtype: str = "float32"
-) -> Tuple[te.Tensor, te.Tensor, te.Tensor]:
+) -> tuple[te.Tensor, te.Tensor, te.Tensor]:
     a = te.placeholder((n, k), name="A", dtype=in_dtype)
     b = te.placeholder((k, m), name="B", dtype=in_dtype)
     k = te.reduce_axis((0, k), name="k")
@@ -593,7 +592,7 @@ def conv2d_nchw(  # pylint: disable=invalid-name
     dilation: int = 1,
     in_dtype: str = "float32",
     out_dtype: str = "float32",
-) -> Tuple[te.Tensor, te.Tensor, te.Tensor]:
+) -> tuple[te.Tensor, te.Tensor, te.Tensor]:
     x = te.placeholder((n, ci, h, w), name="X", dtype=in_dtype)
     w = te.placeholder((co, ci, kh, kw), name="W", dtype=in_dtype)
     y = topi.nn.conv2d_nchw(
@@ -615,7 +614,7 @@ def conv2d_nchw_bias_bn_relu(  # pylint: disable=invalid-name
     dilation: int = 1,
     in_dtype: str = "float32",
     out_dtype: str = "float32",
-) -> Tuple[te.Tensor, te.Tensor, te.Tensor, te.Tensor, te.Tensor, te.Tensor]:
+) -> tuple[te.Tensor, te.Tensor, te.Tensor, te.Tensor, te.Tensor, te.Tensor]:
     oh = (h + 2 * padding - (kh - 1) * dilation - 1) // stride + 1  # pylint: disable=invalid-name
     ow = (w + 2 * padding - (kw - 1) * dilation - 1) // stride + 1  # pylint: disable=invalid-name
     x = te.placeholder((n, ci, h, w), name="X", dtype=in_dtype)
@@ -643,13 +642,13 @@ def max_pool2d_nchw(  # pylint: disable=invalid-name
     w: int,
     ci: int,
     padding: int,
-) -> Tuple[te.Tensor, te.Tensor]:  # pylint: disable=invalid-name
+) -> tuple[te.Tensor, te.Tensor]:  # pylint: disable=invalid-name
     x = te.placeholder((n, ci, h, w), name="X")
     y = topi.nn.pool2d(x, [2, 2], [1, 1], [1, 1], [padding, padding, padding, padding], "max")
     return (x, y)
 
 
-def softmax_mn(m, n) -> Tuple[te.Tensor, te.Tensor]:  # pylint: disable=invalid-name
+def softmax_mn(m, n) -> tuple[te.Tensor, te.Tensor]:  # pylint: disable=invalid-name
     a = te.placeholder((m, n), name="A")
     b = topi.nn.softmax(a, axis=1)
 
diff --git a/python/tvm/s_tir/meta_schedule/testing/tune_te.py b/python/tvm/s_tir/meta_schedule/testing/tune_te.py
index 4b6caf189fcb..959155c3a599 100644
--- a/python/tvm/s_tir/meta_schedule/testing/tune_te.py
+++ b/python/tvm/s_tir/meta_schedule/testing/tune_te.py
@@ -18,7 +18,6 @@
 # ruff: noqa: F821
 import argparse
 import logging
-from typing import Optional
 
 import tvm
 from tvm.s_tir import meta_schedule as ms
@@ -114,7 +113,7 @@ def main():
     describe()
     print(f"Workload: {ARGS.workload}")
     with ms.Profiler() as profiler:
-        sch: Optional[s_tir.Schedule] = ms.tir_integration.tune_tir(
+        sch: s_tir.Schedule | None = ms.tir_integration.tune_tir(
             mod=create_te_workload(ARGS.workload, 0),
             target=ARGS.target,
             work_dir=ARGS.work_dir,
diff --git a/python/tvm/s_tir/meta_schedule/testing/tune_utils.py b/python/tvm/s_tir/meta_schedule/testing/tune_utils.py
index c8502981981e..b93f381048c9 100644
--- a/python/tvm/s_tir/meta_schedule/testing/tune_utils.py
+++ b/python/tvm/s_tir/meta_schedule/testing/tune_utils.py
@@ -16,7 +16,7 @@
 # under the License.
 """Testing utility functions in meta schedule"""
 
-from typing import Callable, Dict, List, Optional
+from collections.abc import Callable
 
 import numpy as np  # type: ignore
 
@@ -25,11 +25,11 @@
 
 
 def generate_input_data(
-    input_shape: List[int],
+    input_shape: list[int],
     input_dtype: str,
     *,
-    low: Optional[int] = None,
-    high: Optional[int] = None,
+    low: int | None = None,
+    high: int | None = None,
 ) -> np.ndarray:
     """Generate input date with given shape and data type.
 
@@ -83,8 +83,8 @@ def create_calculator(backend: str) -> Callable:
     def f_calculator(
         rt_mod: tvm.runtime.Module,
         dev: tvm.runtime.Device,  # pylint: disable=unused-argument
-        input_data: Dict[str, Tensor],
-    ) -> List[Tensor]:
+        input_data: dict[str, Tensor],
+    ) -> list[Tensor]:
         """Fetch the result of running the given runtime module.
 
         Parameters
diff --git a/python/tvm/s_tir/meta_schedule/testing/validate_database.py b/python/tvm/s_tir/meta_schedule/testing/validate_database.py
index a9483513525e..f266e6ac3add 100644
--- a/python/tvm/s_tir/meta_schedule/testing/validate_database.py
+++ b/python/tvm/s_tir/meta_schedule/testing/validate_database.py
@@ -21,8 +21,9 @@
 import itertools
 import logging
 import warnings
+from collections.abc import Callable
 from statistics import mean
-from typing import Any, Callable, List, Optional, Tuple, Union
+from typing import Any
 
 import numpy as np  # type: ignore
 from tvm_ffi import get_global_func, register_global_func
@@ -182,7 +183,7 @@ def get_runtime_device(target: Target) -> tvm.runtime.Device:
         raise RuntimeError(f"Unsupported target kind for runtime device: {target.kind.name}")
 
 
-def check_and_run(func: Union[str, Callable], *args, **kwargs) -> Any:
+def check_and_run(func: str | Callable, *args, **kwargs) -> Any:
     """Check if the function is a string or a callable, and run it."""
     if isinstance(func, str):
         func = get_global_func(func)
@@ -207,7 +208,7 @@ def initializer() -> None:
 
     @register_global_func("tvm.s_tir.meta_schedule.testing.default_check_metric")
     def default_check_metric(  # pylint: disable=unused-variable,unreachable-code
-        lhs: List[tvm.runtime.Tensor], rhs: List[tvm.runtime.Tensor]
+        lhs: list[tvm.runtime.Tensor], rhs: list[tvm.runtime.Tensor]
     ) -> bool:
         """Check if the outputs are equal
 
@@ -234,7 +235,7 @@ def default_check_metric(  # pylint: disable=unused-variable,unreachable-code
 @register_global_func("tvm.s_tir.meta_schedule.testing.default_input_generator")
 def default_input_generator(  # pylint: disable=unused-variable
     mod: IRModule,
-) -> List[tvm.runtime.Tensor]:
+) -> list[tvm.runtime.Tensor]:
     """Default input generator function
 
     Parameters
@@ -258,7 +259,7 @@ def default_input_generator(  # pylint: disable=unused-variable
     return inputs
 
 
-def to_numpy(a: List[tvm.runtime.Tensor]) -> List[np.ndarray]:
+def to_numpy(a: list[tvm.runtime.Tensor]) -> list[np.ndarray]:
     """Convert a list of TVM Tensor to a list of numpy array
 
     Parameters
@@ -275,7 +276,7 @@ def to_numpy(a: List[tvm.runtime.Tensor]) -> List[np.ndarray]:
     return [x.numpy() for x in a]
 
 
-def to_tvm_tensor(a: List[np.ndarray]) -> List[tvm.runtime.Tensor]:
+def to_tvm_tensor(a: list[np.ndarray]) -> list[tvm.runtime.Tensor]:
     """Convert a list of numpy array to a list of TVM Tensor
 
     Parameters
@@ -328,13 +329,13 @@ def print_result(
         *,
         original_mod: IRModule = None,
         scheduled_mod: IRModule = None,
-        inputs: Optional[List[np.ndarray]] = None,
-        original_res: Optional[List[np.ndarray]] = None,
-        scheduled_res: Optional[List[np.ndarray]] = None,
-        original_run_secs: Optional[List[float]] = None,
-        scheduled_run_secs: Optional[List[float]] = None,
-        exception: Optional[Exception] = None,
-        trace: Optional[str] = None,
+        inputs: list[np.ndarray] | None = None,
+        original_res: list[np.ndarray] | None = None,
+        scheduled_res: list[np.ndarray] | None = None,
+        original_run_secs: list[float] | None = None,
+        scheduled_run_secs: list[float] | None = None,
+        exception: Exception | None = None,
+        trace: str | None = None,
     ) -> None:
         """Print the validation result."""
         status = f"Progress {counter: 6d} / {total: 6d} (estimated) checked, result: {result:>10}, "
@@ -381,14 +382,14 @@ def print_result(
 
 
 def make_alloc_arg_and_check(
-    inputs: List[np.ndarray],
+    inputs: list[np.ndarray],
     original_mod: IRModule,
     scheduled_mod: IRModule,
     trace: str,
-    original_res: List[np.ndarray],
-    original_run_secs: List[float],
+    original_res: list[np.ndarray],
+    original_run_secs: list[float],
     print_result: Callable,
-) -> Tuple[Callable, Callable]:
+) -> tuple[Callable, Callable]:
     """Make alloc_arg and check functions for the given inputs and collect results.
 
     Parameters
@@ -421,7 +422,7 @@ def f_with_args_alloc_argument_common(
         device: tvm.runtime.Device,
         args_info: ms.runner.rpc_runner.T_ARG_INFO_JSON_OBJ_LIST,  # pylint: disable=unused-argument
         alloc_repeat: int,
-    ) -> List[ms.runner.rpc_runner.T_ARGUMENT_LIST]:
+    ) -> list[ms.runner.rpc_runner.T_ARGUMENT_LIST]:
         """Allocate arguments using the given inputs.
 
         Parameters
@@ -448,8 +449,8 @@ def f_with_args_run_evaluator_common(
         rt_mod: tvm.runtime.Module,
         device: tvm.runtime.Device,
         evaluator_config: ms.runner.EvaluatorConfig,
-        repeated_args: List[ms.runner.rpc_runner.T_ARGUMENT_LIST],
-    ) -> List[float]:
+        repeated_args: list[ms.runner.rpc_runner.T_ARGUMENT_LIST],
+    ) -> list[float]:
         """With args function to run the evaluator
 
         Parameters
@@ -481,7 +482,7 @@ def f_with_args_run_evaluator_common(
             ),
         )
 
-        repeated_costs: List[List[float]] = []
+        repeated_costs: list[list[float]] = []
         for args in repeated_args:
             device.sync()
             profile_result = evaluator(*args)
@@ -516,7 +517,7 @@ def f_with_args_alloc_argument_rpc(
         device: tvm.runtime.Device,
         args_info: ms.runner.rpc_runner.T_ARG_INFO_JSON_OBJ_LIST,
         alloc_repeat: int,
-    ) -> List[ms.runner.rpc_runner.T_ARGUMENT_LIST]:
+    ) -> list[ms.runner.rpc_runner.T_ARGUMENT_LIST]:
         return f_with_args_alloc_argument_common(device, args_info, alloc_repeat)
 
     def f_with_args_run_evaluator_rpc(
@@ -524,8 +525,8 @@ def f_with_args_run_evaluator_rpc(
         rt_mod: tvm.runtime.Module,
         device: tvm.runtime.Device,
         evaluator_config: ms.runner.EvaluatorConfig,
-        repeated_args: List[ms.runner.rpc_runner.T_ARGUMENT_LIST],
-    ) -> List[float]:
+        repeated_args: list[ms.runner.rpc_runner.T_ARGUMENT_LIST],
+    ) -> list[float]:
         return f_with_args_run_evaluator_common(rt_mod, device, evaluator_config, repeated_args)
 
     if ARGS.rpc_config is None:
@@ -538,8 +539,8 @@ def local_build_and_run(
     mod: IRModule,
     target: Target,
     device: tvm.runtime.Device,
-    inputs: List[np.ndarray],
-) -> Tuple[List[np.ndarray], List[float]]:
+    inputs: list[np.ndarray],
+) -> tuple[list[np.ndarray], list[float]]:
     """Build and run the module locally.
 
     Parameters
@@ -604,8 +605,8 @@ def _apply_trace(mod: IRModule, trace: Trace) -> IRModule:
 
 
 def _build_all_mods(
-    mods: List[IRModule], builder: ms.builder.Builder, target: Target
-) -> List[ms.builder.BuilderResult]:
+    mods: list[IRModule], builder: ms.builder.Builder, target: Target
+) -> list[ms.builder.BuilderResult]:
     """Build all the modules.
 
     Parameters
diff --git a/python/tvm/s_tir/meta_schedule/tir_integration.py b/python/tvm/s_tir/meta_schedule/tir_integration.py
index 5978c0a5ff41..3fe6d090a4de 100644
--- a/python/tvm/s_tir/meta_schedule/tir_integration.py
+++ b/python/tvm/s_tir/meta_schedule/tir_integration.py
@@ -16,10 +16,10 @@
 # under the License.
 """MetaSchedule-TIR integration"""
 
-from typing import List, Mapping, Optional, Tuple, Union
+from collections.abc import Mapping
 
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 from tvm_ffi import register_global_func
 
 # isort: on
@@ -43,12 +43,12 @@
 
 
 def tune_tir(  # pylint: disable=too-many-locals
-    mod: Union[ir.IRModule, tir.PrimFunc],
-    target: Union[str, Target],
+    mod: ir.IRModule | tir.PrimFunc,
+    target: str | Target,
     work_dir: str,
     max_trials_global: int,
     *,
-    max_trials_per_task: Optional[int] = None,
+    max_trials_per_task: int | None = None,
     num_trials_per_iter: int = 64,
     builder: Builder.BuilderType = "local",
     runner: Runner.RunnerType = "local",
@@ -58,11 +58,11 @@ def tune_tir(  # pylint: disable=too-many-locals
     task_scheduler: TaskScheduler.TaskSchedulerType = "gradient",
     space: SpaceGenerator.SpaceGeneratorType = "post-order-apply",
     strategy: SearchStrategy.SearchStrategyType = "evolutionary",
-    num_tuning_cores: Union[Literal["physical", "logical"], int] = "physical",
-    seed: Optional[int] = None,
+    num_tuning_cores: Literal["physical", "logical"] | int = "physical",
+    seed: int | None = None,
     module_equality: str = "structural",
-    special_space: Optional[Mapping[str, SpaceGenerator.SpaceGeneratorType]] = None,
-    post_optimization: Optional[bool] = False,
+    special_space: Mapping[str, SpaceGenerator.SpaceGeneratorType] | None = None,
+    post_optimization: bool | None = False,
 ) -> Database:
     """Tune a TIR function or an IRModule of TIR functions.
 
@@ -113,14 +113,14 @@ def tune_tir(  # pylint: disable=too-many-locals
     if isinstance(mod, tir.PrimFunc):
         mod = _normalize_mod(mod)
 
-    named_tasks: List[Tuple[str, tir.PrimFunc]] = []
+    named_tasks: list[tuple[str, tir.PrimFunc]] = []
     for gv, func in mod.functions_items():  # pylint: disable=invalid-name
         if isinstance(func, tir.PrimFunc):
             named_tasks.append((gv.name_hint, func))
     named_tasks.sort(key=lambda x: x[0])
 
     task_names = [x for x, _ in named_tasks]
-    tasks: List[TuneContext] = []
+    tasks: list[TuneContext] = []
     for task_name, task_func, logger, rand_state in zip(
         task_names,
         [x for _, x in named_tasks],
@@ -165,8 +165,8 @@ def tune_tir(  # pylint: disable=too-many-locals
 
 @register_global_func("tvm.s_tir.meta_schedule.tune_tir")
 def _tune_tir(
-    mod: Union[ir.IRModule, tir.PrimFunc],
-    target: Union[str, Target],
+    mod: ir.IRModule | tir.PrimFunc,
+    target: str | Target,
     work_dir: str,
     max_trials_global: int,
     *,
@@ -179,8 +179,8 @@ def _tune_tir(
     task_scheduler: TaskScheduler.TaskSchedulerType = "round-robin",
     space: SpaceGenerator.SpaceGeneratorType = "post-order-apply",
     strategy: SearchStrategy.SearchStrategyType = "evolutionary",
-    num_tuning_cores: Union[Literal["physical", "logical"], int] = "physical",
-    seed: Optional[int] = None,
+    num_tuning_cores: Literal["physical", "logical"] | int = "physical",
+    seed: int | None = None,
 ) -> Database:
     """Interface with tuning api to tune a TIR program.
 
@@ -248,8 +248,8 @@ def _tune_tir(
 
 def compile_tir(
     database: Database,
-    mod: Union[ir.IRModule, tir.PrimFunc],
-    target: Union[Target, str],
+    mod: ir.IRModule | tir.PrimFunc,
+    target: Target | str,
 ) -> _Schedule:
     """Compile a TIR to s_tir.Schedule, according to the records in the database.
 
diff --git a/python/tvm/s_tir/meta_schedule/tune.py b/python/tvm/s_tir/meta_schedule/tune.py
index 20de6e08018d..e5f4c5d2dd4f 100644
--- a/python/tvm/s_tir/meta_schedule/tune.py
+++ b/python/tvm/s_tir/meta_schedule/tune.py
@@ -16,8 +16,6 @@
 # under the License.
 """The core tuning API"""
 
-from typing import List, Optional
-
 from .builder import Builder
 from .cost_model import CostModel
 from .database import Database
@@ -30,11 +28,11 @@
 
 def tune_tasks(
     *,
-    tasks: List[TuneContext],
-    task_weights: List[float],
+    tasks: list[TuneContext],
+    task_weights: list[float],
     work_dir: str,
     max_trials_global: int,
-    max_trials_per_task: Optional[int] = None,
+    max_trials_per_task: int | None = None,
     num_trials_per_iter: int = 64,
     builder: Builder.BuilderType = "local",
     runner: Runner.RunnerType = "local",
@@ -43,7 +41,7 @@ def tune_tasks(
     measure_callbacks: MeasureCallback.CallbackListType = "default",
     task_scheduler: TaskScheduler.TaskSchedulerType = "gradient",
     module_equality: str = "structural",
-    post_optimization: Optional[bool] = False,
+    post_optimization: bool | None = False,
 ) -> Database:
     """Tune a list of tasks. Using a task scheduler.
 
diff --git a/python/tvm/s_tir/meta_schedule/tune_context.py b/python/tvm/s_tir/meta_schedule/tune_context.py
index 81f2ada1be0b..09de3afc5ebf 100644
--- a/python/tvm/s_tir/meta_schedule/tune_context.py
+++ b/python/tvm/s_tir/meta_schedule/tune_context.py
@@ -17,10 +17,10 @@
 # ruff: noqa: F401
 """Meta Schedule tuning context."""
 
-from typing import TYPE_CHECKING, List, Optional, Union
+from typing import TYPE_CHECKING, Optional, Union
 
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 from tvm_ffi import register_object, register_global_func
 
@@ -45,7 +45,7 @@
 
 
 @register_global_func("tvm.s_tir.meta_schedule.normalize_mod")
-def _normalize_mod(mod: Union[PrimFunc, IRModule]) -> IRModule:
+def _normalize_mod(mod: PrimFunc | IRModule) -> IRModule:
     """Normalize the input to an IRModule"""
     if isinstance(mod, PrimFunc):
         if not (mod.attrs and "global_symbol" in mod.attrs):
@@ -86,26 +86,26 @@ class TuneContext(Object):
         The number of threads to be used, None means using the logical cpu count.
     """
 
-    mod: Optional[IRModule]
-    target: Optional[Target]
+    mod: IRModule | None
+    target: Target | None
     space_generator: Optional["SpaceGenerator"]
     search_strategy: Optional["SearchStrategy"]
     task_name: str
-    logger: Optional[Logger]
+    logger: Logger | None
     rand_state: int
     num_threads: int
 
     def __init__(
         self,
-        mod: Optional[IRModule] = None,
+        mod: IRModule | None = None,
         *,
-        target: Union[Target, str, None] = None,
+        target: Target | str | None = None,
         space_generator: Union["SpaceGenerator.SpaceGeneratorType", None] = None,
         search_strategy: Union["SearchStrategy.SearchStrategyType", None] = None,
         task_name: str = "main",
         rand_state: int = -1,
-        num_threads: Union[int, Literal["physical", "logical"]] = "physical",
-        logger: Optional[Logger] = None,
+        num_threads: int | Literal["physical", "logical"] = "physical",
+        logger: Logger | None = None,
     ):
         # pylint: disable=import-outside-toplevel
         import tvm.s_tir.tensor_intrin  # pylint: disable=unused-import
@@ -159,7 +159,7 @@ def __init__(
         )
         _ffi_api.TuneContextInitialize(self)  # type: ignore # pylint: disable=no-member
 
-    def generate_design_space(self) -> List[Schedule]:
+    def generate_design_space(self) -> list[Schedule]:
         """Generate design spaces given a module.
 
         Delegated to self.space_generator.generate_design_space with self.mod
@@ -181,7 +181,7 @@ def pre_tuning(
         self,
         max_trials: int,
         num_trials_per_iter: int = 64,
-        design_spaces: Optional[List[Schedule]] = None,
+        design_spaces: list[Schedule] | None = None,
         database: Optional["Database"] = None,
         cost_model: Optional["CostModel"] = None,
     ) -> None:
@@ -245,7 +245,7 @@ def post_tuning(self) -> None:
             )
         return self.search_strategy.post_tuning()
 
-    def generate_measure_candidates(self) -> Optional[List["MeasureCandidate"]]:
+    def generate_measure_candidates(self) -> list["MeasureCandidate"] | None:
         """Generate a batch of measure candidates from design spaces for measurement.
 
         Delegated to self.search_strategy.generate_measure_candidates.
@@ -263,8 +263,8 @@ def generate_measure_candidates(self) -> Optional[List["MeasureCandidate"]]:
 
     def notify_runner_results(
         self,
-        measure_candidates: List["MeasureCandidate"],
-        results: List["RunnerResult"],
+        measure_candidates: list["MeasureCandidate"],
+        results: list["RunnerResult"],
     ) -> None:
         """Update the state in SearchStrategy with profiling results.
 
diff --git a/python/tvm/s_tir/meta_schedule/utils.py b/python/tvm/s_tir/meta_schedule/utils.py
index 6c9642813980..10c8327b2bbe 100644
--- a/python/tvm/s_tir/meta_schedule/utils.py
+++ b/python/tvm/s_tir/meta_schedule/utils.py
@@ -19,7 +19,8 @@
 import ctypes
 import os
 import shutil
-from typing import Any, Callable, List, Optional, Union
+from collections.abc import Callable
+from typing import Any
 
 import numpy as np  # type: ignore
 import psutil  # type: ignore
@@ -158,7 +159,7 @@ def __setattr__(self, name, value):
     TVMDerivedObject.__doc__ = cls.__doc__
     TVMDerivedObject.__module__ = cls.__module__
     for key, value in cls.__dict__.items():
-        if isinstance(value, (classmethod, staticmethod)):
+        if isinstance(value, classmethod | staticmethod):
             setattr(TVMDerivedObject, key, value)
     return TVMDerivedObject
 
@@ -260,7 +261,7 @@ def print_interactive_table(data: str) -> None:
 
 
 def get_global_func_with_default_on_worker(
-    name: Union[None, str, Callable],
+    name: None | str | Callable,
     default: Callable,
 ) -> Callable:
     """Get the registered global function on the worker process.
@@ -299,7 +300,7 @@ def get_global_func_with_default_on_worker(
 def get_global_func_on_rpc_session(
     session: RPCSession,
     name: str,
-    extra_error_msg: Optional[str] = None,
+    extra_error_msg: str | None = None,
 ) -> PackedFunc:
     """Get a PackedFunc from the global registry from an RPCSession.
 
@@ -348,11 +349,11 @@ def _json_de_tvm(obj: Any) -> Any:
     """
     if obj is None:
         return None
-    if isinstance(obj, (int, float)):
+    if isinstance(obj, int | float):
         return obj
-    if isinstance(obj, (IntImm, FloatImm)):
+    if isinstance(obj, IntImm | FloatImm):
         return obj.value
-    if isinstance(obj, (str,)):
+    if isinstance(obj, str):
         return str(obj)
     if isinstance(obj, Array):
         return [_json_de_tvm(i) for i in obj]
@@ -401,7 +402,7 @@ def _to_hex_address(handle: ctypes.c_void_p) -> str:
     return hex(ctypes.cast(handle, ctypes.c_void_p).value)
 
 
-def fork_seed(seed: Optional[int], n: int) -> List[int]:
+def fork_seed(seed: int | None, n: int) -> list[int]:
     # fmt: off
     return np.random.RandomState(seed=seed).randint(1, 2 ** 30, size=n).tolist()
     # fmt: on
diff --git a/python/tvm/s_tir/sblock_dependence_info.py b/python/tvm/s_tir/sblock_dependence_info.py
index 4bf1b6de13fa..968533a7c999 100644
--- a/python/tvm/s_tir/sblock_dependence_info.py
+++ b/python/tvm/s_tir/sblock_dependence_info.py
@@ -17,8 +17,6 @@
 """Define SBlockDependenceInfoNode that uses the SBlockScope and StmtSRef objects
 to store the block level dependences"""
 
-from typing import Optional, Union
-
 from tvm_ffi import register_object
 
 from tvm.ir.module import IRModule
@@ -47,7 +45,7 @@ class SBlockDependenceInfo(Object):
 
     mod: IRModule
 
-    def __init__(self, mod: Union[IRModule, PrimFunc]):
+    def __init__(self, mod: IRModule | PrimFunc):
         if isinstance(mod, PrimFunc):
             mod = IRModule({"main": mod})
         if not isinstance(mod, IRModule):
@@ -57,7 +55,7 @@ def __init__(self, mod: Union[IRModule, PrimFunc]):
             mod,
         )
 
-    def get_sref(self, block: SBlock) -> Optional[StmtSRef]:
+    def get_sref(self, block: SBlock) -> StmtSRef | None:
         """Return the corresponding sref that points to the block
 
         Parameters
diff --git a/python/tvm/s_tir/sblock_scope.py b/python/tvm/s_tir/sblock_scope.py
index 9d4fd483cb30..4406cd4481c5 100644
--- a/python/tvm/s_tir/sblock_scope.py
+++ b/python/tvm/s_tir/sblock_scope.py
@@ -17,7 +17,7 @@
 """Definition of two pillar data structure for TensorIR scheduling: StmtSRef, SBlockScope."""
 
 from enum import IntEnum
-from typing import List, Optional, Union
+from typing import Optional
 
 from tvm_ffi import register_object
 
@@ -45,7 +45,7 @@ class StmtSRef(Object):
     seq_index: int
 
     @property
-    def stmt(self) -> Optional[Union[SBlock, For]]:
+    def stmt(self) -> SBlock | For | None:
         """The block/for stmt the object refers to"""
         return _ffi_api.StmtSRefStmt(self)  # type: ignore # pylint: disable=no-member
 
@@ -126,7 +126,7 @@ class SBlockScope(Object):
     - Child SBlock: The scope leaf SBlocks under the scope root or a specific internal sref
     """
 
-    def get_deps_by_src(self, block: StmtSRef) -> List[Dependency]:
+    def get_deps_by_src(self, block: StmtSRef) -> list[Dependency]:
         """Get all dependencies whose `src` is the target`block`.
 
         Parameters
@@ -141,7 +141,7 @@ def get_deps_by_src(self, block: StmtSRef) -> List[Dependency]:
         """
         return _ffi_api.SBlockScopeGetDepsBySrc(self, block)  # type: ignore # pylint: disable=no-member
 
-    def get_deps_by_dst(self, block: StmtSRef) -> List[Dependency]:
+    def get_deps_by_dst(self, block: StmtSRef) -> list[Dependency]:
         """Get all dependencies whose `dst` is the target `block`.
 
         Parameters
diff --git a/python/tvm/s_tir/schedule/_type_checker.py b/python/tvm/s_tir/schedule/_type_checker.py
index f1e7804b93a4..57e612688bb0 100644
--- a/python/tvm/s_tir/schedule/_type_checker.py
+++ b/python/tvm/s_tir/schedule/_type_checker.py
@@ -21,8 +21,10 @@
 import collections.abc
 import functools
 import inspect
+import types
 import typing
-from typing import Any, Callable, Dict, List, Optional, Tuple, TypeVar, Union
+from collections.abc import Callable
+from typing import Any, TypeVar, Union
 
 
 def _is_none_type(type_: Any) -> bool:
@@ -64,10 +66,13 @@ def _origin(type_: Any) -> Any:
         def list_(type_: Any) -> Any:
             if _Subtype._origin(type_) is list:
                 if hasattr(typing, "get_args"):
-                    (subtype,) = typing.get_args(type_)  # type: ignore
+                    args = typing.get_args(type_)  # type: ignore
                 else:
-                    (subtype,) = type_.__args__
-                return [subtype]
+                    args = type_.__args__
+                if len(args) == 1:
+                    return [args[0]]
+                # Handle list[X | Y] where get_args may return individual types
+                return [Union[args]]  # noqa: UP007 (runtime use)
             return None
 
         @staticmethod
@@ -81,7 +86,7 @@ def dict_(type_: Any) -> Any:
             return None
 
         @staticmethod
-        def tuple_(type_: Any) -> Optional[List[type]]:
+        def tuple_(type_: Any) -> list[type] | None:
             if _Subtype._origin(type_) is tuple:
                 subtypes = _get_subtypes(type_)
                 return subtypes
@@ -90,23 +95,33 @@ def tuple_(type_: Any) -> Optional[List[type]]:
         @staticmethod
         def optional(  # pylint: disable=missing-function-docstring
             type_: Any,
-        ) -> Optional[List[type]]:
+        ) -> list[type] | None:
             if _Subtype._origin(type_) is Union:
                 subtypes = _get_subtypes(type_)
                 if len(subtypes) == 2 and _is_none_type(subtypes[1]):
                     return [subtypes[0]]
+            # PEP 604: X | None
+            if isinstance(type_, types.UnionType):
+                subtypes = type_.__args__
+                if len(subtypes) == 2 and _is_none_type(subtypes[1]):
+                    return [subtypes[0]]
             return None
 
         @staticmethod
-        def union(type_: Any) -> Optional[List[type]]:  # pylint: disable=missing-function-docstring
+        def union(type_: Any) -> list[type] | None:  # pylint: disable=missing-function-docstring
             if _Subtype._origin(type_) is Union:
                 subtypes = _get_subtypes(type_)
                 if len(subtypes) != 2 or not _is_none_type(subtypes[1]):
                     return list(subtypes)
+            # PEP 604: X | Y
+            if isinstance(type_, types.UnionType):
+                subtypes = type_.__args__
+                if len(subtypes) != 2 or not _is_none_type(subtypes[1]):
+                    return list(subtypes)
             return None
 
         @staticmethod
-        def callable(type_: Any) -> Optional[List[type]]:
+        def callable(type_: Any) -> list[type] | None:
             if _Subtype._origin(type_) is collections.abc.Callable:
                 subtypes = _get_subtypes(type_)
                 return subtypes
@@ -117,15 +132,18 @@ def callable(type_: Any) -> Optional[List[type]]:
 
     class _Subtype:  # type: ignore
         @staticmethod
-        def list_(type_: Any) -> Optional[List[type]]:
+        def list_(type_: Any) -> list[type] | None:
             if isinstance(type_, typing.GenericMeta):  # type: ignore # pylint: disable=no-member
                 if type_.__name__ == "List":
-                    (subtype,) = type_.__args__  # type: ignore # pylint: disable=no-member
-                    return [subtype]
+                    args = type_.__args__  # type: ignore # pylint: disable=no-member
+                    if len(args) == 1:
+                        return [args[0]]
+                    # Handle list[X | Y] where args may return individual types
+                    return [Union[args]]  # noqa: UP007 (runtime use)
             return None
 
         @staticmethod
-        def dict_(type_: Any) -> Optional[List[type]]:
+        def dict_(type_: Any) -> list[type] | None:
             if isinstance(type_, typing.GenericMeta):  # type: ignore # pylint: disable=no-member
                 if type_.__name__ == "Dict":
                     (ktype, vtype) = type_.__args__  # type: ignore # pylint: disable=no-member
@@ -133,7 +151,7 @@ def dict_(type_: Any) -> Optional[List[type]]:
             return None
 
         @staticmethod
-        def tuple_(type_: Any) -> Optional[List[type]]:
+        def tuple_(type_: Any) -> list[type] | None:
             if isinstance(type_, typing.GenericMeta):  # type: ignore # pylint: disable=no-member
                 if type_.__name__ == "Tuple":
                     subtypes = type_.__args__  # type: ignore # pylint: disable=no-member
@@ -141,7 +159,7 @@ def tuple_(type_: Any) -> Optional[List[type]]:
             return None
 
         @staticmethod
-        def optional(type_: Any) -> Optional[List[type]]:
+        def optional(type_: Any) -> list[type] | None:
             if isinstance(type_, typing._Union):  # type: ignore # pylint: disable=no-member,protected-access
                 subtypes = type_.__args__
                 if len(subtypes) == 2 and _is_none_type(subtypes[1]):
@@ -149,7 +167,7 @@ def optional(type_: Any) -> Optional[List[type]]:
             return None
 
         @staticmethod
-        def union(type_: Any) -> Optional[List[type]]:
+        def union(type_: Any) -> list[type] | None:
             if isinstance(type_, typing._Union):  # type: ignore # pylint: disable=no-member,protected-access
                 subtypes = type_.__args__
                 if len(subtypes) != 2 or not _is_none_type(subtypes[1]):
@@ -157,14 +175,14 @@ def union(type_: Any) -> Optional[List[type]]:
             return None
 
         @staticmethod
-        def callable(type_: Any) -> Optional[List[type]]:
+        def callable(type_: Any) -> list[type] | None:
             if isinstance(type_, typing.CallableMeta):  # type: ignore # pylint: disable=no-member,protected-access
                 subtypes = type_.__args__
                 return subtypes
             return None
 
 
-def _dispatcher(type_: Any) -> Tuple[str, List[type]]:
+def _dispatcher(type_: Any) -> tuple[str, list[type]]:
     if _is_none_type(type_):
         return "none", []
 
@@ -205,7 +223,7 @@ def callable_str(*subtypes):
         return "Callable"
 
 
-_TYPE2STR: Dict[Any, Callable] = {
+_TYPE2STR: dict[Any, Callable] = {
     "none": lambda: "None",
     "atomic": lambda t: str(t.__name__),
     "callable": callable_str,
@@ -226,13 +244,13 @@ def _val2type(value: Any):
     if isinstance(value, list):
         types = set(_val2type(x) for x in value)
         if len(types) == 1:
-            return List[types.pop()]  # type: ignore
+            return list[types.pop()]  # type: ignore
 
-        return List[Union[tuple(types)]]  # type: ignore
+        return list[tuple(types)]  # type: ignore
 
     if isinstance(value, tuple):
         types = tuple(_val2type(x) for x in value)  # type: ignore
-        return Tuple[types]
+        return tuple[types]
 
     return type(value)
 
@@ -245,21 +263,21 @@ def _type_check_err(x: Any, name: str, expected: Any) -> str:
     )
 
 
-def _type_check_vtable() -> Dict[str, Callable]:
-    def _type_check_none(v: Any, name: str) -> Optional[str]:
+def _type_check_vtable() -> dict[str, Callable]:
+    def _type_check_none(v: Any, name: str) -> str | None:
         return None if v is None else _type_check_err(v, name, None)
 
-    def _type_check_atomic(v: Any, name: str, type_: Any) -> Optional[str]:
+    def _type_check_atomic(v: Any, name: str, type_: Any) -> str | None:
         return None if isinstance(v, type_) else _type_check_err(v, name, type_)
 
-    def _type_check_callable(v: Any, name: str, *_subtypes: Any) -> Optional[str]:
+    def _type_check_callable(v: Any, name: str, *_subtypes: Any) -> str | None:
         # Current implementation only validates that the argument is
         # callable, and doesn't validate the arguments accepted by the
         # callable, if any.
         return None if callable(v) else _type_check_err(v, name, Callable)
 
-    def _type_check_list(v: List[Any], name: str, type_: Any) -> Optional[str]:
-        if not isinstance(v, (list, tuple)):
+    def _type_check_list(v: list[Any], name: str, type_: Any) -> str | None:
+        if not isinstance(v, list | tuple):
             return _type_check_err(v, name, list)
         for i, x in enumerate(v):
             error_msg = _type_check(x, f"{name}[{i}]", type_)
@@ -267,7 +285,7 @@ def _type_check_list(v: List[Any], name: str, type_: Any) -> Optional[str]:
                 return error_msg
         return None
 
-    def _type_check_dict(dict_obj: Dict[Any, Any], name: str, *types: Any) -> Optional[str]:
+    def _type_check_dict(dict_obj: dict[Any, Any], name: str, *types: Any) -> str | None:
         ktype_, vtype_ = types
         if not isinstance(dict_obj, dict):
             return _type_check_err(dict_obj, name, dict)
@@ -280,26 +298,26 @@ def _type_check_dict(dict_obj: Dict[Any, Any], name: str, *types: Any) -> Option
                 return error_msg
         return None
 
-    def _type_check_tuple(v: Any, name: str, *types: Any) -> Optional[str]:
+    def _type_check_tuple(v: Any, name: str, *types: Any) -> str | None:
         if not isinstance(v, tuple):
-            return _type_check_err(v, name, Tuple[types])
+            return _type_check_err(v, name, tuple[types])
         if len(types) != len(v):
-            return _type_check_err(v, name, Tuple[types])
+            return _type_check_err(v, name, tuple[types])
         for i, (x, type_) in enumerate(zip(v, types)):
             error_msg = _type_check(x, f"{name}[{i}]", type_)
             if error_msg is not None:
                 return error_msg
         return None
 
-    def _type_check_optional(v: Any, name: str, type_: Any) -> Optional[str]:
+    def _type_check_optional(v: Any, name: str, type_: Any) -> str | None:
         return None if v is None else _type_check(v, name, type_)
 
-    def _type_check_union(v: Any, name: str, *types: Any) -> Optional[str]:
+    def _type_check_union(v: Any, name: str, *types: Any) -> str | None:
         for type_ in types:
             error_msg = _type_check(v, name, type_)
             if error_msg is None:
                 return None
-        return _type_check_err(v, name, Union[types])
+        return _type_check_err(v, name, Union[types])  # noqa: UP007 (runtime use)
 
     return {
         "none": _type_check_none,
@@ -313,10 +331,10 @@ def _type_check_union(v: Any, name: str, *types: Any) -> Optional[str]:
     }
 
 
-_TYPE_CHECK: Dict[Any, Callable] = _type_check_vtable()
+_TYPE_CHECK: dict[Any, Callable] = _type_check_vtable()
 
 
-def _type_check(v: Any, name: str, type_: Any) -> Optional[str]:
+def _type_check(v: Any, name: str, type_: Any) -> str | None:
     key, subtypes = _dispatcher(type_)
     return _TYPE_CHECK[key](v, name, *subtypes)
 
diff --git a/python/tvm/s_tir/schedule/analysis.py b/python/tvm/s_tir/schedule/analysis.py
index 2862110a97d4..0b4da193504d 100644
--- a/python/tvm/s_tir/schedule/analysis.py
+++ b/python/tvm/s_tir/schedule/analysis.py
@@ -16,8 +16,6 @@
 # under the License.
 """Analysis used in TensorIR scheduling"""
 
-from typing import List, Optional
-
 import tvm_ffi
 
 from tvm.runtime import Object
@@ -32,10 +30,10 @@
 
 def suggest_index_map(
     buffer: Buffer,
-    indices: List[PrimExpr],
-    loops: List[For],
+    indices: list[PrimExpr],
+    loops: list[For],
     predicate: PrimExpr,
-) -> Optional[IndexMap]:
+) -> IndexMap | None:
     """Provided the access pattern to a buffer, suggest one of the possible layout
     transformation to maximize the locality of the access pattern.
 
@@ -70,7 +68,7 @@ class TensorizeInfo(Object):
 
 def get_tensorize_loop_mapping(
     sch: Schedule, block: SBlockRV, desc_func: PrimFunc, allow_padding: bool = False
-) -> Optional[TensorizeInfo]:
+) -> TensorizeInfo | None:
     """Establish a mapping between loops in a target block and an intrinsic description
 
     Parameters
@@ -98,7 +96,7 @@ class AutoTensorizeMappingInfo(Object):
 
 def get_auto_tensorize_mapping_info(
     sch: Schedule, block: SBlockRV, desc_func: PrimFunc
-) -> Optional[AutoTensorizeMappingInfo]:
+) -> AutoTensorizeMappingInfo | None:
     """Get mapping info between a target block and an intrinsic description including layout
     transformations to apply.
 
diff --git a/python/tvm/s_tir/schedule/instruction.py b/python/tvm/s_tir/schedule/instruction.py
index cc11ae71ea28..5e918206f9f6 100644
--- a/python/tvm/s_tir/schedule/instruction.py
+++ b/python/tvm/s_tir/schedule/instruction.py
@@ -16,7 +16,7 @@
 # under the License.
 """Schedule instructions each corresponds to a schedule primitive"""
 
-from typing import TYPE_CHECKING, Any, List, Union
+from typing import TYPE_CHECKING, Any
 
 from tvm_ffi import register_object as _register_object
 
@@ -27,8 +27,8 @@
 if TYPE_CHECKING:
     from .schedule import RAND_VAR_TYPE
 
-    INPUT_RV_TYPE = Union[RAND_VAR_TYPE, float, int, str, None]  # pylint: disable=invalid-name
-    OUTPUT_RV_TYPE = Union[RAND_VAR_TYPE]  # pylint: disable=invalid-name
+    INPUT_RV_TYPE = RAND_VAR_TYPE | float | int | str | None  # pylint: disable=invalid-name
+    OUTPUT_RV_TYPE = RAND_VAR_TYPE  # pylint: disable=invalid-name
     ATTR_TYPE = Any
 else:
     INPUT_RV_TYPE = OUTPUT_RV_TYPE = ATTR_TYPE = Any
@@ -121,16 +121,16 @@ class Instruction(Object):
     """
 
     kind: InstructionKind
-    inputs: List[INPUT_RV_TYPE]
-    attrs: List[ATTR_TYPE]
-    outputs: List[OUTPUT_RV_TYPE]
+    inputs: list[INPUT_RV_TYPE]
+    attrs: list[ATTR_TYPE]
+    outputs: list[OUTPUT_RV_TYPE]
 
     def __init__(
         self,
         kind: InstructionKind,
-        inputs: List[INPUT_RV_TYPE],
-        attrs: List[ATTR_TYPE],
-        outputs: List[OUTPUT_RV_TYPE],
+        inputs: list[INPUT_RV_TYPE],
+        attrs: list[ATTR_TYPE],
+        outputs: list[OUTPUT_RV_TYPE],
     ) -> None:
         """Constructor
 
diff --git a/python/tvm/s_tir/schedule/schedule.py b/python/tvm/s_tir/schedule/schedule.py
index cf232acb5ee6..0f07249ffadc 100644
--- a/python/tvm/s_tir/schedule/schedule.py
+++ b/python/tvm/s_tir/schedule/schedule.py
@@ -17,7 +17,8 @@
 """The TensorIR schedule class"""
 
 import inspect
-from typing import Callable, Dict, List, Literal, Optional, Tuple, Union
+from collections.abc import Callable
+from typing import Literal
 
 from tvm_ffi import register_object as _register_object
 
@@ -64,11 +65,11 @@ def __init__(self) -> None:
 # This feature is not supported until python 3.10:
 # https://docs.python.org/3.10/whatsnew/3.10.html#pep-613-typealias
 # A random variable that evaluates to an integer
-ExprRV = Union[PrimExpr]  # pylint: disable=invalid-name
+ExprRV = PrimExpr  # pylint: disable=invalid-name
 
-RAND_VAR_TYPE = Union[ExprRV, SBlockRV, LoopRV]  # pylint: disable=invalid-name
+RAND_VAR_TYPE = ExprRV | SBlockRV | LoopRV  # pylint: disable=invalid-name
 
-_ERROR_RENDER_LEVEL: Dict[Literal["detail", "fast", "none"], int] = {
+_ERROR_RENDER_LEVEL: dict[Literal["detail", "fast", "none"], int] = {
     "detail": 0,
     "fast": 1,
     "none": 2,
@@ -90,7 +91,7 @@ def _parse_enable_checks(enable_checks: bool) -> bool:
     return enable_checks
 
 
-def _parse_seed(seed: Optional[int]) -> int:
+def _parse_seed(seed: int | None) -> int:
     if seed is None:
         return -1
     if not isinstance(seed, int):
@@ -127,10 +128,10 @@ class Schedule(Object):
     @type_checked
     def __init__(
         self,
-        mod: Union[PrimFunc, IRModule],
+        mod: PrimFunc | IRModule,
         *,
-        seed: Optional[int] = None,
-        debug_mask: Union[str, int] = "none",
+        seed: int | None = None,
+        debug_mask: str | int = "none",
         error_render_level: str = "detail",
         enable_check: bool = True,
     ) -> None:
@@ -138,13 +139,13 @@ def __init__(
 
         Parameters
         ----------
-        mod : Union[PrimFunc, IRModule]
+        mod : PrimFunc | IRModule
             The IRModule or PrimFunc to be scheduled
         seed: Optional[int]
             The seed value for schedule's random state
             Note that None and -1 means use device random, otherwise only integer between 1 and
             2147483647 is allowed.
-        debug_mask : Union[str, int]
+        debug_mask : str | int
             Do extra correctness checking after the class creation and each time
             after calling the Replace method.
             Possible choices of `debug_mask`:
@@ -184,10 +185,10 @@ def __init__(
 
     @staticmethod
     def _create_non_traced(
-        mod: Union[PrimFunc, IRModule],
+        mod: PrimFunc | IRModule,
         *,
-        seed: Optional[int] = None,
-        debug_mask: Union[str, int] = "none",
+        seed: int | None = None,
+        debug_mask: str | int = "none",
         error_render_level: str = "detail",
         enable_check: bool = True,
     ) -> "Schedule":
@@ -213,12 +214,12 @@ def state(self) -> ScheduleState:
         return _ffi_api.ScheduleGetState(self)  # type: ignore # pylint: disable=no-member
 
     @property
-    def trace(self) -> Optional[Trace]:
+    def trace(self) -> Trace | None:
         """Returns the internally maintained trace of scheduling program execution"""
         return _ffi_api.ScheduleGetTrace(self)  # type: ignore # pylint: disable=no-member
 
     @property
-    def func_working_on(self) -> Optional[GlobalVar]:
+    def func_working_on(self) -> GlobalVar | None:
         """Returns the GlobalVar of the func that the schedule is currently working on"""
         return _ffi_api.ScheduleGetFuncWorkingOn(self)  # type: ignore # pylint: disable=no-member
 
@@ -297,9 +298,7 @@ def show(self, *args, **kwargs) -> None:
     ########## Lookup ##########
 
     @type_checked
-    def get(
-        self, rand_var_or_sref: Union[RAND_VAR_TYPE, StmtSRef]
-    ) -> Optional[Union[int, SBlock, For]]:
+    def get(self, rand_var_or_sref: RAND_VAR_TYPE | StmtSRef) -> int | SBlock | For | None:
         """Returns:
         - the corresponding SBlock that a SBlockRV evaluates to;
         - the corresponding For that a LoopRV evaluates to;
@@ -309,12 +308,12 @@ def get(
 
         Parameters
         ----------
-        rand_var_or_sref : Union[ExprRV, SBlockRV, LoopRV, StmtSRef]
+        rand_var_or_sref : ExprRV | SBlockRV | LoopRV | StmtSRef
             The random variable / sref to be evaluated
 
         Returns
         -------
-        result : Optional[Union[int, SBlock, For]]
+        result : Optional[int | SBlock | For]
             The corresponding result
         """
         if isinstance(rand_var_or_sref, StmtSRef):
@@ -326,9 +325,7 @@ def get(
         return result
 
     @type_checked
-    def get_sref(
-        self, rand_var_or_stmt: Union[SBlockRV, LoopRV, SBlock, For]
-    ) -> Optional[StmtSRef]:
+    def get_sref(self, rand_var_or_stmt: SBlockRV | LoopRV | SBlock | For) -> StmtSRef | None:
         """Returns the corresponding sref to the given
         1) LoopRV
         2) SBlockRV
@@ -337,7 +334,7 @@ def get_sref(
 
         Parameters
         ----------
-        rand_var_or_stmt : Union[SBlockRV, LoopRV, SBlock, For]
+        rand_var_or_stmt : SBlockRV | LoopRV | SBlock | For
             The random variable / sref to be evaluated
 
         Returns
@@ -355,7 +352,7 @@ def remove_rv(self, rand_var: RAND_VAR_TYPE) -> None:
 
         Parameters
         ----------
-        rand_var : Union[SBlockRV, LoopRV, ExprRV]
+        rand_var : SBlockRV | LoopRV | ExprRV
             The random variable to be removed
         """
         return _ffi_api.ScheduleRemoveRV(self, rand_var)  # type: ignore # pylint: disable=no-member
@@ -364,7 +361,7 @@ def remove_rv(self, rand_var: RAND_VAR_TYPE) -> None:
 
     @type_checked
     def sample_categorical(
-        self, candidates: List[int], probs: List[float], decision: Optional[int] = None
+        self, candidates: list[int], probs: list[float], decision: int | None = None
     ) -> ExprRV:
         """Sample an integer given the probability distribution
 
@@ -392,8 +389,8 @@ def sample_perfect_tile(
         loop: LoopRV,
         n: int,
         max_innermost_factor: int = 16,
-        decision: Optional[List[int]] = None,
-    ) -> List[ExprRV]:
+        decision: list[int] | None = None,
+    ) -> list[ExprRV]:
         """Sample the factors to perfect tile a specific loop
 
         Parameters
@@ -425,8 +422,8 @@ def sample_partitioned_tile(
         n: int,
         partition_pos: int = 0,
         innerpart_factor: int = 1,
-        decision: Optional[List[int]] = None,
-    ) -> List[ExprRV]:
+        decision: list[int] | None = None,
+    ) -> list[ExprRV]:
         """Sample the factors to a partitioned tile for a specific loop
 
         Parameters
@@ -459,14 +456,12 @@ def sample_partitioned_tile(
         )
 
     @type_checked
-    def sample_compute_location(
-        self, block: Union[SBlockRV, str], decision: Optional[int] = None
-    ) -> LoopRV:
+    def sample_compute_location(self, block: SBlockRV | str, decision: int | None = None) -> LoopRV:
         """Sample a compute-at location of the given block
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The block whose compute-at location is to be sampled
         decision : Optional[int]
             The sampling decision
@@ -484,7 +479,7 @@ def sample_compute_location(
 
     ########## Schedule: Get blocks & loops ##########
     @type_checked
-    def get_sblock(self, name: str, func_name: Optional[str] = None) -> SBlockRV:
+    def get_sblock(self, name: str, func_name: str | None = None) -> SBlockRV:
         """Retrieve a block in a specific function with its name
 
         By default, if `func_name` is not specified, the schedule will search for the block in the
@@ -509,12 +504,12 @@ def get_sblock(self, name: str, func_name: Optional[str] = None) -> SBlockRV:
         )
 
     @type_checked
-    def get_loops(self, block: Union[SBlockRV, str]) -> List[LoopRV]:
+    def get_loops(self, block: SBlockRV | str) -> list[LoopRV]:
         """Get the parent loops of the block in its scope, from outer to inner
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The query block
 
         Returns
@@ -527,12 +522,12 @@ def get_loops(self, block: Union[SBlockRV, str]) -> List[LoopRV]:
         return list(_ffi_api.ScheduleGetLoops(self, block))  # type: ignore
 
     @type_checked
-    def get_child_blocks(self, block_or_loop: Union[SBlockRV, LoopRV]) -> List[SBlockRV]:
+    def get_child_blocks(self, block_or_loop: SBlockRV | LoopRV) -> list[SBlockRV]:
         """Get the leaf blocks of a specific block/loop
 
         Parameters
         ----------
-        block_or_loop : Union[SBlockRV, LoopRV]
+        block_or_loop : SBlockRV | LoopRV
             The query block/loop
 
         Returns
@@ -544,12 +539,12 @@ def get_child_blocks(self, block_or_loop: Union[SBlockRV, LoopRV]) -> List[SBloc
         return list(_ffi_api.ScheduleGetChildBlocks(self, block_or_loop))  # type: ignore
 
     @type_checked
-    def get_producers(self, block: Union[SBlockRV, str]) -> List[SBlockRV]:
+    def get_producers(self, block: SBlockRV | str) -> list[SBlockRV]:
         """Get the producers of a specific block
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The block in the query
 
         Returns
@@ -562,12 +557,12 @@ def get_producers(self, block: Union[SBlockRV, str]) -> List[SBlockRV]:
         return list(_ffi_api.ScheduleGetProducers(self, block))  # type: ignore
 
     @type_checked
-    def get_consumers(self, block: Union[SBlockRV, str]) -> List[SBlockRV]:
+    def get_consumers(self, block: SBlockRV | str) -> list[SBlockRV]:
         """Get the consumers of a specific block
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The block in the query
 
         Returns
@@ -580,14 +575,14 @@ def get_consumers(self, block: Union[SBlockRV, str]) -> List[SBlockRV]:
         return list(_ffi_api.ScheduleGetConsumers(self, block))  # type: ignore
 
     @type_checked
-    def get_output_blocks(self, scope_block: Union[SBlockRV, str]) -> List[SBlockRV]:
+    def get_output_blocks(self, scope_block: SBlockRV | str) -> list[SBlockRV]:
         """Get the list of output blocks within the given scope
         An output block is a block which has atleast one buffer being written
         to, but is not allocated within the PrimFunc
 
         Parameters
         ----------
-        scope_block : Union[SBlockRV, str],
+        scope_block : SBlockRV | str,
             The scope block from which output blocks are collected
 
         Returns
@@ -602,7 +597,7 @@ def get_output_blocks(self, scope_block: Union[SBlockRV, str]) -> List[SBlockRV]
 
     ########## Schedule: Transform loops ##########
     @type_checked
-    def merge(self, *loops: List[LoopRV]) -> LoopRV:
+    def merge(self, *loops: list[LoopRV]) -> LoopRV:
         """Merge a list of loops into one. The loops under their LCA requires:
         1) Under the same scope.
         2) Can't have annotations or thread bindings.
@@ -677,7 +672,7 @@ def after_fuse(a: T.handle, b: T.handle, c: T.handle) -> None:
         return _ffi_api.ScheduleMerge(self, loops)  # type: ignore # pylint: disable=no-member
 
     @type_checked
-    def fuse(self, *loops: List[LoopRV], preserve_unit_iters: bool = True) -> LoopRV:
+    def fuse(self, *loops: list[LoopRV], preserve_unit_iters: bool = True) -> LoopRV:
         """Fuse a list of consecutive loops into one. It requires:
         1) The loops can't have annotations or thread bindings.
         2) The (i+1)-th loop must be the only child of the i-th loop.
@@ -742,10 +737,10 @@ def after_fuse(a: T.handle, b: T.handle) -> None:
     def split(
         self,
         loop: LoopRV,
-        factors: List[Union[int, ExprRV, None]],
+        factors: list[int | ExprRV | None],
         preserve_unit_iters: bool = True,
         disable_predication: bool = False,
-    ) -> List[LoopRV]:
+    ) -> list[LoopRV]:
         """Split a loop into a list of consecutive loops. It requires:
         1) The loop can't have annotation or thread binding.
         2) The loop must start with 0.
@@ -758,7 +753,7 @@ def split(
         loop : LoopRV
             The loop to be split
 
-        factors: List[Union[int, ExprRV, None]]
+        factors: List[int | ExprRV | None]
             The splitting factors
             Potential inputs are:
             - None
@@ -838,9 +833,9 @@ def after_split(a: T.handle, b: T.handle) -> None:
     def loop_partition(
         self,
         loop: LoopRV,
-        factors: List[Union[int, ExprRV, None]],
+        factors: list[int | ExprRV | None],
         preserve_unit_iters: bool = True,
-    ) -> List[LoopRV]:
+    ) -> list[LoopRV]:
         """Partition a loop into a list of consecutive loops. It requires:
         1) The loop can't have annotation or thread binding.
         Predicates may be added to ensure the total loop numbers keeps unchanged.
@@ -852,7 +847,7 @@ def loop_partition(
         loop : LoopRV
             The loop to be partition
 
-        factors: List[Union[int, ExprRV, None]]
+        factors: List[int | ExprRV | None]
             The partitioning factors
             Potential inputs are:
             - None
@@ -943,7 +938,7 @@ def after_partition(a: T.handle, b: T.handle) -> None:
         )
 
     @type_checked
-    def reorder(self, *ordered_loops: List[LoopRV]) -> None:
+    def reorder(self, *ordered_loops: list[LoopRV]) -> None:
         """
         Reorder a list of loops. It doesn't require the loops to be consecutive.
         It requires:
@@ -1003,7 +998,7 @@ def after_reorder(a: T.handle, b: T.handle) -> None:
         _ffi_api.ScheduleReorder(self, ordered_loops)  # type: ignore # pylint: disable=no-member
 
     @type_checked
-    def reorder_block_iter_var(self, block: SBlockRV, new_order: List[int]) -> None:
+    def reorder_block_iter_var(self, block: SBlockRV, new_order: list[int]) -> None:
         """Reorder the itervars inside a given block.
 
         Parameters
@@ -1068,12 +1063,12 @@ def matmul_after_reorder_block_iter_var(
         _ffi_api.ScheduleReorderBlockIterVar(self, block, new_order)  # type: ignore
 
     @type_checked
-    def add_unit_loop(self, block_or_loop: Union[LoopRV, SBlockRV]) -> LoopRV:
+    def add_unit_loop(self, block_or_loop: LoopRV | SBlockRV) -> LoopRV:
         """Create a new unit loop on top of the specific block or loop.
 
         Parameters
         ----------
-        block_or_loop : Union[LoopRV, SBlockRV]
+        block_or_loop : LoopRV | SBlockRV
             The block above which the new loop is created
 
         Returns
@@ -1355,10 +1350,10 @@ def after_unroll(a: T.handle, b: T.handle) -> None:
     @type_checked
     def cache_read(
         self,
-        block: Union[SBlockRV, str],
-        read_buffer_index: Union[int, str, Buffer],
+        block: SBlockRV | str,
+        read_buffer_index: int | str | Buffer,
         storage_scope: str,
-        consumer_blocks: Optional[List[Union[SBlockRV, str]]] = None,
+        consumer_blocks: list[SBlockRV | str] | None = None,
     ) -> SBlockRV:
         """Create a block that reads a buffer region into a read cache. It requires:
 
@@ -1368,10 +1363,10 @@ def cache_read(
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The consumer block of the target buffer.
 
-        buffer: Union[int, str, Buffer]
+        buffer: int | str | Buffer
             The index of the buffer in block's read region, the unique
             name of a read buffer in the block, or a Buffer object
             that is within the blocks read region.
@@ -1379,7 +1374,7 @@ def cache_read(
         storage_scope: str
             The target storage scope.
 
-        consumer_blocks: Optional[List[Union[SBlockRV, str]]]
+        consumer_blocks: Optional[List[SBlockRV | str]]
             An optional list of consumers that should read from the cache. If not specified,
             all consumers will use the cache.
 
@@ -1449,10 +1444,10 @@ def after_cache_read(a: T.handle, b: T.handle) -> None:
     @type_checked
     def cache_write(
         self,
-        block: Union[SBlockRV, str],
-        write_buffer_index: Union[int, str, Buffer],
+        block: SBlockRV | str,
+        write_buffer_index: int | str | Buffer,
         storage_scope: str,
-        consumer_blocks: Optional[List[Union[SBlockRV, str]]] = None,
+        consumer_blocks: list[SBlockRV | str] | None = None,
     ) -> SBlockRV:
         """Create a block that reads a buffer region into a write cache. It requires:
 
@@ -1462,7 +1457,7 @@ def cache_write(
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The producer block of the target buffer.
 
         write_buffer_index: int
@@ -1473,7 +1468,7 @@ def cache_write(
         storage_scope: str
             The target storage scope.
 
-        consumer_blocks: Optional[List[Union[SBlockRV, str]]]
+        consumer_blocks: Optional[List[SBlockRV | str]]
             An optional list of consumers that should read directly from the cache.
             If not specified, all consumers will read from the original buffer.
 
@@ -1543,10 +1538,10 @@ def after_cache_write(a: T.handle, b: T.handle) -> None:
     @type_checked
     def reindex_cache_read(
         self,
-        block: Union[SBlockRV, str],
+        block: SBlockRV | str,
         read_buffer_index: int,
         storage_scope: str,
-        index_map: Union[IndexMap, Callable],
+        index_map: IndexMap | Callable,
     ) -> SBlockRV:
         """Create a block that reads a buffer region into a read cache using customized
         indices specified by index map. The read region of the buffer must be a single point.
@@ -1567,7 +1562,7 @@ def reindex_cache_read(
             The index of the buffer in block's read region.
         storage_scope: str
             The target storage scope.
-        index_map: Union[IndexMap, Callable]
+        index_map: IndexMap | Callable
             User defined indices to access allocated cache buffer, maps from block iter vars.
 
         Returns
@@ -1640,10 +1635,10 @@ def after_reindex_cache_read(a: T.handle, b: T.handle) -> None:
     @type_checked
     def reindex_cache_write(
         self,
-        block: Union[SBlockRV, str],
+        block: SBlockRV | str,
         write_buffer_index: int,
         storage_scope: str,
-        index_map: Union[Callable, IndexMap],
+        index_map: Callable | IndexMap,
     ) -> SBlockRV:
         r"""Create a block that reads a buffer region into a write cache using customized
         indices specified by index map. The write region of the buffer must be a single point.
@@ -1658,15 +1653,15 @@ def reindex_cache_write(
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The consumer block of the target buffer.
         write_buffer_index: int
             The index of the buffer in block's write region.
         storage_scope: str
             The target storage scope.
-        index_map: Union[Callable, IndexMap]
+        index_map: Callable | IndexMap
             User defined indices to access allocated cache buffer, maps from block iter vars.
-        consumer_blocks: Optional[List[Union[SBlockRV, str]]]
+        consumer_blocks: Optional[List[SBlockRV | str]]
             An optional list of consumers that should read directly from the cache.
             If not specified, all consumers will read from the original buffer.
 
@@ -1740,17 +1735,17 @@ def after_cache_write(a: T.handle, b: T.handle) -> None:
     @type_checked
     def cache_inplace(
         self,
-        block: Union[SBlockRV, str],
-        read_buffer_index: Union[int, str, Buffer],
+        block: SBlockRV | str,
+        read_buffer_index: int | str | Buffer,
         storage_scope: str,
-    ) -> List[SBlockRV]:
+    ) -> list[SBlockRV]:
         """Create blocks that reads & write a buffer region into a cache block.
         It requires the target block both read & write the target buffer.
         Mainly for inplace operation.
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The target block operates on the target buffer.
 
         read_buffer_index: int
@@ -1828,14 +1823,14 @@ def cache_inplace(data_io: T.Buffer(64, "int32")) -> None:
 
     @type_checked
     def cache_index(
-        self, block: Union[SBlockRV, str], storage_scope: str, cse_thresh: int = 0
-    ) -> List[SBlockRV]:
+        self, block: SBlockRV | str, storage_scope: str, cse_thresh: int = 0
+    ) -> list[SBlockRV]:
         """Create a block to cache precomputed index for later use.
         if there is no index computation, keep unchanged.
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The target block operates on the target buffer.
 
         storage_scope: str
@@ -1913,9 +1908,7 @@ def resize_cache_index(
         )
 
     @type_checked
-    def reindex(
-        self, block: Union[SBlockRV, str], buffer: Union[Tuple[str, int], str, Buffer]
-    ) -> SBlockRV:
+    def reindex(self, block: SBlockRV | str, buffer: tuple[str, int] | str | Buffer) -> SBlockRV:
         """Create a block that read/write a buffer region into a read/write cache with reindexing.
         The layout of the cache will be the same as by the iterators of the block that reads/writes
         the buffer. It requires:
@@ -1924,7 +1917,7 @@ def reindex(
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
 
             The block that accesses the target buffer.  If a string,
             this must uniquely identify a block.
@@ -2025,7 +2018,7 @@ def write_at(
     @type_checked
     def compute_at(
         self,
-        block: Union[SBlockRV, str],
+        block: SBlockRV | str,
         loop: LoopRV,
         preserve_unit_loops: bool = False,
         index: int = -1,
@@ -2049,7 +2042,7 @@ def compute_at(
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The block to be moved
 
         loop: LoopRV
@@ -2123,7 +2116,7 @@ def after_compute_at(a: T.handle, c: T.handle) -> None:
     @type_checked
     def reverse_compute_at(
         self,
-        block: Union[SBlockRV, str],
+        block: SBlockRV | str,
         loop: LoopRV,
         preserve_unit_loops: bool = False,
         index: int = -1,
@@ -2144,7 +2137,7 @@ def reverse_compute_at(
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The block to be moved
 
         loop: LoopRV
@@ -2216,7 +2209,7 @@ def after_reverse_compute_at(a: T.handle, c: T.handle) -> None:
         )
 
     @type_checked
-    def compute_inline(self, block: Union[SBlockRV, str]) -> None:
+    def compute_inline(self, block: SBlockRV | str) -> None:
         """Inline a block into its consumer(s). It requires:
 
         1) The block is a complete non-root block, which only produces one buffer
@@ -2231,7 +2224,7 @@ def compute_inline(self, block: Union[SBlockRV, str]) -> None:
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The block to be inlined to its consumer(s)
 
         Examples
@@ -2281,7 +2274,7 @@ def after_inline(a: T.handle, c: T.handle) -> None:
         _ffi_api.ScheduleComputeInline(self, block)  # type: ignore # pylint: disable=no-member
 
     @type_checked
-    def reverse_compute_inline(self, block: Union[SBlockRV, str]) -> None:
+    def reverse_compute_inline(self, block: SBlockRV | str) -> None:
         """Inline a block into its only producer. It requires:
 
         1) The block is a complete non-root block, which only produces and consumes one buffer
@@ -2299,7 +2292,7 @@ def reverse_compute_inline(self, block: Union[SBlockRV, str]) -> None:
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The block to be inlined to its producer
 
         Examples
@@ -2352,8 +2345,8 @@ def after_inline(a: T.handle, c: T.handle) -> None:
     @type_checked
     def fuse_reduction_epilogue(
         self,
-        reduction_block: Union[SBlockRV, str],
-        epilogue_block: Union[SBlockRV, str],
+        reduction_block: SBlockRV | str,
+        epilogue_block: SBlockRV | str,
     ) -> None:
         """Fuse an epilogue block into a reduction block.
 
@@ -2387,9 +2380,9 @@ def fuse_reduction_epilogue(
 
         Parameters
         ----------
-        reduction_block : Union[SBlockRV, str]
+        reduction_block : SBlockRV | str
             The reduction block (e.g., matmul)
-        epilogue_block : Union[SBlockRV, str]
+        epilogue_block : SBlockRV | str
             The epilogue block to be fused (e.g., bias add, ReLU, clipping)
 
         Examples
@@ -2404,7 +2397,7 @@ def fuse_reduction_epilogue(
     ########## Schedule: Reduction ##########
 
     @type_checked
-    def decompose_reduction(self, block: Union[SBlockRV, str], loop: LoopRV) -> SBlockRV:
+    def decompose_reduction(self, block: SBlockRV | str, loop: LoopRV) -> SBlockRV:
         """Decompose a reduction block into two separate blocks.
 
         a) The init block, which is translated from the init statement of the reduction block;
@@ -2423,7 +2416,7 @@ def decompose_reduction(self, block: Union[SBlockRV, str], loop: LoopRV) -> SBlo
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The reduction block to be decomposed
         loop : LoopRV
             The loop above which the init block is inserted before.
@@ -2636,7 +2629,7 @@ def after_rfactor(a: T.handle, b: T.handle) -> None:
 
     @type_checked
     def storage_align(  # pylint: disable=too-many-arguments
-        self, block: Union[SBlockRV, str], buffer_index: int, axis: int, factor: int, offset: int
+        self, block: SBlockRV | str, buffer_index: int, axis: int, factor: int, offset: int
     ) -> None:
         """Set alignment requirement for specific dimension such that
         stride[axis] == k * factor + offset for some k. This is useful to set memory layout for more
@@ -2645,7 +2638,7 @@ def storage_align(  # pylint: disable=too-many-arguments
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The producer block of the buffer.
         buffer_index : int
             The index of the buffer in block's write region.
@@ -2717,14 +2710,14 @@ def after_storage_align(a: T.handle, c: T.handle) -> None:
 
     @type_checked
     def set_scope(
-        self, block: Union[SBlockRV, str], buffer_index: Union[int, str, Buffer], storage_scope: str
+        self, block: SBlockRV | str, buffer_index: int | str | Buffer, storage_scope: str
     ) -> None:
         """Set the storage scope of a buffer, where the buffer is
         specified by the a block and a write-index.
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The producer block of the buffer
         buffer_index : int
             The index of the buffer in block's write region
@@ -2794,7 +2787,7 @@ def after_set_scope(
         )
 
     @type_checked
-    def unsafe_set_dtype(self, block: Union[SBlockRV, str], buffer_index: int, dtype: str) -> None:
+    def unsafe_set_dtype(self, block: SBlockRV | str, buffer_index: int, dtype: str) -> None:
         """Set the data type of a buffer, where the buffer is
         specified by the a block and write-index.
 
@@ -2803,7 +2796,7 @@ def unsafe_set_dtype(self, block: Union[SBlockRV, str], buffer_index: int, dtype
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The producer block of the buffer
         buffer_index : int
             The index of the buffer in block's write region
@@ -2873,7 +2866,7 @@ def after_set_dtype(
 
     @type_checked
     def blockize(
-        self, target: Union[LoopRV, List[SBlockRV]], preserve_unit_iters: bool = True
+        self, target: LoopRV | list[SBlockRV], preserve_unit_iters: bool = True
     ) -> SBlockRV:
         """Convert multiple blocks or the subtree rooted at a specific loop into a block.
 
@@ -2953,7 +2946,7 @@ def after_blockize(
     @type_checked
     def tensorize(
         self,
-        block_or_loop: Union[SBlockRV, LoopRV],
+        block_or_loop: SBlockRV | LoopRV,
         tensor_intrin: str,
         preserve_unit_iters: bool = True,
     ) -> None:
@@ -2961,7 +2954,7 @@ def tensorize(
 
         Parameters
         ----------
-        block_or_loop : Union[SBlockRV, LoopRV]
+        block_or_loop : SBlockRV | LoopRV
             The loop to be tensorized.
         tensor_intrin : str
             The tensor intrin or the name of the tensor intrin.
@@ -3105,22 +3098,22 @@ def after_tensorize(
 
     ########## Schedule: Annotation ##########
 
-    PrimAnnotationValueT = Union[str, int, float, ExprRV]
-    AnnotationValueT = Union[
-        PrimAnnotationValueT,
-        List[PrimAnnotationValueT],
-        Dict[str, Union[PrimAnnotationValueT, List[PrimAnnotationValueT]]],
-    ]
+    PrimAnnotationValueT = str | int | float | ExprRV
+    AnnotationValueT = (
+        PrimAnnotationValueT
+        | list[PrimAnnotationValueT]
+        | dict[str, PrimAnnotationValueT | list[PrimAnnotationValueT]]
+    )
 
     @type_checked
     def annotate(
-        self, block_or_loop: Union[SBlockRV, LoopRV], ann_key: str, ann_val: AnnotationValueT
+        self, block_or_loop: SBlockRV | LoopRV, ann_key: str, ann_val: AnnotationValueT
     ) -> None:
         """Annotate a block/loop with a key value pair
 
         Parameters
         ----------
-        block_or_loop: Union[SBlockRV, LoopRV]
+        block_or_loop: SBlockRV | LoopRV
             The block/loop to be annotated
         ann_key : str
             The annotation key
@@ -3171,12 +3164,12 @@ def after_annotate(a: T.handle, b: T.handle) -> None:
         )
 
     @type_checked
-    def unannotate(self, block_or_loop: Union[SBlockRV, LoopRV], ann_key: str) -> None:
+    def unannotate(self, block_or_loop: SBlockRV | LoopRV, ann_key: str) -> None:
         """Unannotate a block/loop's annotation with key ann_key
 
         Parameters
         ----------
-        block_or_loop: Union[SBlockRV, LoopRV]
+        block_or_loop: SBlockRV | LoopRV
             The block/loop to be unannotated
         ann_key : str
             The annotation key
@@ -3226,7 +3219,7 @@ def after_unannotate(a: T.handle, b: T.handle) -> None:
 
     ########## Schedule: Layout transformation ##########
 
-    def _normalize_block_arg(self, block: Union[SBlockRV, str]) -> SBlockRV:
+    def _normalize_block_arg(self, block: SBlockRV | str) -> SBlockRV:
         if isinstance(block, str):
             return self.get_sblock(block)
 
@@ -3235,9 +3228,9 @@ def _normalize_block_arg(self, block: Union[SBlockRV, str]) -> SBlockRV:
     def _normalize_buffer_arg(
         self,
         block: SBlockRV,
-        buffer: Union[Tuple[str, int], int, str, Buffer],
+        buffer: tuple[str, int] | int | str | Buffer,
         required_buffer_type=None,
-    ) -> Tuple[str, int, Buffer]:
+    ) -> tuple[str, int, Buffer]:
         block_obj: SBlock = self.get(block)
         block_name = block_obj.name_hint
 
@@ -3303,10 +3296,10 @@ def iter_buffers():
     @type_checked
     def transform_layout(
         self,
-        block: Union[SBlockRV, str],
-        buffer: Union[Tuple[str, int], str, Buffer],
-        index_map: Union[IndexMap, Callable],
-        pad_value: Optional[Union[int, float, PrimExpr, IndexMap, Callable]] = None,
+        block: SBlockRV | str,
+        buffer: tuple[str, int] | str | Buffer,
+        index_map: IndexMap | Callable,
+        pad_value: int | float | PrimExpr | IndexMap | Callable | None = None,
         *,
         assume_injective_transform: bool = False,
     ) -> None:
@@ -3314,7 +3307,7 @@ def transform_layout(
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
 
             The block that accesses the target buffer.  If a string,
             this must uniquely identify a block.
@@ -3336,7 +3329,7 @@ def transform_layout(
             If `buffer` is a Buffer object, it must exist within the
             reads/writes of the block.
 
-        index_map : Union[IndexMap, Callable]
+        index_map : IndexMap | Callable
 
             The transformation to apply.
 
@@ -3345,7 +3338,7 @@ def transform_layout(
             primitive will be called in addition to the
             TransformLayout primitive.
 
-        pad_value: Optional[Union[int, float, PrimExpr, IndexMap, Callable]]
+        pad_value: Optional[int | float | PrimExpr | IndexMap | Callable]
 
             The value to be used for any padding introduced by the
             transformation.  If the schedule contains a producer block
@@ -3482,17 +3475,15 @@ def two_elementwise_transformed_intermediate_buffer(a: T.handle, c: T.handle) ->
             )
 
     @type_checked
-    def transform_block_layout(
-        self, block: Union[SBlockRV, str], index_map: Union[IndexMap, Callable]
-    ) -> None:
+    def transform_block_layout(self, block: SBlockRV | str, index_map: IndexMap | Callable) -> None:
         """Apply a transformation represented by IndexMap to block
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The block to be transformed
 
-        index_map : Union[IndexMap, Callable]
+        index_map : IndexMap | Callable
             The transformation to apply.
 
         Examples
@@ -3546,16 +3537,16 @@ def after_transform_block_layout(
 
     def set_axis_separator(
         self,
-        block: Union[SBlockRV, str],
-        buffer: Union[Tuple[str, int], str, Buffer],
-        axis_separators: Optional[List[int]],
+        block: SBlockRV | str,
+        buffer: tuple[str, int] | str | Buffer,
+        axis_separators: list[int] | None,
     ) -> None:
         """Set the axis separator of a buffer, where the buffer is specified by a block and a read
         or write index.
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
 
             The block that accesses the target buffer.  If a string,
             this must uniquely identify a block.
@@ -3643,7 +3634,7 @@ def after_set_axis_separators(
 
     ########## Schedule: Padding decomposition #########
     @type_checked
-    def decompose_padding(self, block: Union[SBlockRV, str], loop: LoopRV) -> SBlockRV:
+    def decompose_padding(self, block: SBlockRV | str, loop: LoopRV) -> SBlockRV:
         """Decompose a block of padding computation pattern into two separate blocks.
 
         a) The block which fill const pad values into full write region;
@@ -3662,7 +3653,7 @@ def decompose_padding(self, block: Union[SBlockRV, str], loop: LoopRV) -> SBlock
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The padding block to be decomposed.
         loop : LoopRV
             The loop above which the pad value filling block is inserted before.
@@ -3715,13 +3706,13 @@ def after_decompose(x: T.Buffer(128, "int32"), y: T.Buffer(140, "int32")):
         )
 
     @type_checked
-    def can_decompose_padding(self, block: Union[SBlockRV, str], loop: LoopRV) -> bool:
+    def can_decompose_padding(self, block: SBlockRV | str, loop: LoopRV) -> bool:
         """Check whether the block match padding pattern and can be decomposed."""
         # pylint: disable-next=no-member
         return _ffi_api.CanDecomposePadding(self, block, loop)  # type: ignore
 
     @type_checked
-    def pad_einsum(self, block: Union[SBlockRV, str], padding: List[int]) -> None:
+    def pad_einsum(self, block: SBlockRV | str, padding: list[int]) -> None:
         """Pad the computation of Einsum.
 
         On a block with trivial binding, this primitive pads the iteration domain of the block by
@@ -3734,7 +3725,7 @@ def pad_einsum(self, block: Union[SBlockRV, str], padding: List[int]) -> None:
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The block that matches the Einsum pattern.
 
         padding : List[int]
@@ -3819,7 +3810,7 @@ def main(
     ######## Schedule: Buffer transformation ########
 
     @type_checked
-    def rolling_buffer(self, block: Union[SBlockRV, str], write_buffer_index: int) -> None:
+    def rolling_buffer(self, block: SBlockRV | str, write_buffer_index: int) -> None:
         """Compute the target buffer via rolling buffering, select the outermost rollable
         axis with a positive bound overlap that appears in the block's ancestor loops
         as `rolling axis`, fold and circularize the buffer along the rolling dimension,
@@ -3837,7 +3828,7 @@ def rolling_buffer(self, block: Union[SBlockRV, str], write_buffer_index: int) -
 
         Parameters
         ----------
-        block : Union[SBlockRV, str]
+        block : SBlockRV | str
             The producer block of the buffer.
         write_buffer_index : int
             The index of the buffer in block's write region.
@@ -3930,7 +3921,7 @@ def enter_postproc(self) -> None:
 
     @type_checked
     def unsafe_hide_buffer_access(
-        self, block: SBlockRV, buf_type: str, buf_index_array: List[int]
+        self, block: SBlockRV, buf_type: str, buf_index_array: list[int]
     ) -> None:
         """Hide some buffer access in a given block. This is an unsafe schedule primitive.
 
@@ -4066,7 +4057,7 @@ def after_annotate_buffer_access(
 
         result = []
         for rng in new_ranges_spec:
-            if isinstance(rng, (tuple, list)):
+            if isinstance(rng, tuple | list):
                 if len(rng) != 2:
                     raise ValueError(
                         "Tuple must have exactly 2 elements to represent (begin, end)."
diff --git a/python/tvm/s_tir/schedule/state.py b/python/tvm/s_tir/schedule/state.py
index 74d5c14fe120..19d27c06788e 100644
--- a/python/tvm/s_tir/schedule/state.py
+++ b/python/tvm/s_tir/schedule/state.py
@@ -19,7 +19,6 @@
 
 from collections import namedtuple
 from enum import IntEnum
-from typing import Dict, Optional, Union
 
 from tvm_ffi import register_object
 
@@ -52,7 +51,7 @@ class ScheduleDebugMask(IntEnum):
     VERIFY_CACHED_FLAGS = 2
 
 
-def _parse_mod(mod: Union[PrimFunc, IRModule]) -> IRModule:
+def _parse_mod(mod: PrimFunc | IRModule) -> IRModule:
     if isinstance(mod, PrimFunc):
         mod = IRModule({"main": mod})
     if not isinstance(mod, IRModule):
@@ -60,7 +59,7 @@ def _parse_mod(mod: Union[PrimFunc, IRModule]) -> IRModule:
     return mod
 
 
-def _parse_debug_mask(debug_mask: Union[str, int]) -> int:
+def _parse_debug_mask(debug_mask: str | int) -> int:
     if isinstance(debug_mask, str):
         if debug_mask == "all":
             debug_mask = ScheduleDebugMask.VERIFY_SREF_TREE | ScheduleDebugMask.VERIFY_CACHED_FLAGS
@@ -109,9 +108,9 @@ class ScheduleState(Object):
 
     def __init__(
         self,
-        mod: Union[PrimFunc, IRModule],
+        mod: PrimFunc | IRModule,
         *,
-        debug_mask: Union[str, int] = "none",
+        debug_mask: str | int = "none",
         enable_check: bool = True,
     ) -> None:
         """Construct a schedule state from an IRModule or a PrimFunc
@@ -135,7 +134,7 @@ def __init__(
             _parse_enable_checks(enable_check),
         )
 
-    def get_sref(self, stmt: Union[SBlock, For]) -> Optional[StmtSRef]:
+    def get_sref(self, stmt: SBlock | For) -> StmtSRef | None:
         """Return the corresponding sref that points to the stmt
 
         Parameters
@@ -200,8 +199,8 @@ def _get_cached_flags(self, block_sref: StmtSRef) -> CachedFlags:
     def replace(
         self,
         src_sref: StmtSRef,
-        tgt_stmt: Union[SBlock, For, SBlockRealize],
-        block_sref_reuse: Optional[Dict[SBlock, SBlock]] = None,
+        tgt_stmt: SBlock | For | SBlockRealize,
+        block_sref_reuse: dict[SBlock, SBlock] | None = None,
     ) -> None:
         """
         Replace the part of the AST, as being pointed to by `src_sref`,
diff --git a/python/tvm/s_tir/schedule/testing.py b/python/tvm/s_tir/schedule/testing.py
index 5fca562fd8b0..7fe0d55087ac 100644
--- a/python/tvm/s_tir/schedule/testing.py
+++ b/python/tvm/s_tir/schedule/testing.py
@@ -17,7 +17,8 @@
 # pylint: disable=dangerous-default-value
 """Testing utilities for the TensorIR schedule API"""
 
-from typing import Any, Sequence, Union
+from collections.abc import Sequence
+from typing import Any
 
 import tvm
 from tvm.ir import IRModule, assert_structural_equal
@@ -46,10 +47,10 @@ def assert_structural_equal_ignore_global_symbol(
 
 def verify_trace_roundtrip(
     sch: Schedule,
-    mod: Union[PrimFunc, IRModule],
+    mod: PrimFunc | IRModule,
     *,
-    debug_mask: Union[str, int] = "all",
-    text_format: Union[str, Sequence[str]] = ["python", "json"],
+    debug_mask: str | int = "all",
+    text_format: str | Sequence[str] = ["python", "json"],
 ) -> Schedule:
     """Serialize a traced schedule to JSON, then replay the JSON trace by applying to
     a fresh new schedule, verifying the reproducibility of scheduling.
diff --git a/python/tvm/s_tir/schedule/trace.py b/python/tvm/s_tir/schedule/trace.py
index b7e0447a0c89..f2602eb281d2 100644
--- a/python/tvm/s_tir/schedule/trace.py
+++ b/python/tvm/s_tir/schedule/trace.py
@@ -17,7 +17,8 @@
 """An execution trace of a scheduling program"""
 
 import os
-from typing import TYPE_CHECKING, Any, Callable, Dict, List, Optional
+from collections.abc import Callable
+from typing import TYPE_CHECKING, Any
 
 from tvm_ffi import register_object as _register_object
 
@@ -40,7 +41,7 @@
 def _json_from_tvm(obj):
     if obj is None:
         return None
-    elif isinstance(obj, (bool, int, float, str)):
+    elif isinstance(obj, bool | int | float | str):
         return obj
     elif isinstance(obj, Array):
         return [_json_from_tvm(i) for i in obj]
@@ -48,7 +49,7 @@ def _json_from_tvm(obj):
         return {_json_from_tvm(k): _json_from_tvm(v) for k, v in obj.items()}
     elif isinstance(obj, str):
         return str(obj)
-    elif isinstance(obj, (IntImm, FloatImm)):
+    elif isinstance(obj, IntImm | FloatImm):
         return obj
     elif isinstance(obj, IndexMap):
         return save_json(obj)
@@ -80,13 +81,13 @@ class Trace(Object):
         The random decisions made upon those instructions
     """
 
-    insts: List[Instruction]
-    decisions: Dict[Instruction, DECISION_TYPE]
+    insts: list[Instruction]
+    decisions: dict[Instruction, DECISION_TYPE]
 
     def __init__(
         self,
-        insts: List[Instruction],
-        decisions: Dict[Instruction, DECISION_TYPE],
+        insts: list[Instruction],
+        decisions: dict[Instruction, DECISION_TYPE],
     ) -> None:
         """Constructor
 
@@ -103,7 +104,7 @@ def __init__(
             decisions,
         )
 
-    def get_decision(self, inst: Instruction) -> Optional[DECISION_TYPE]:
+    def get_decision(self, inst: Instruction) -> DECISION_TYPE | None:
         """Retrieve the decision made on a specific instruction
 
         Parameters
@@ -121,7 +122,7 @@ def get_decision(self, inst: Instruction) -> Optional[DECISION_TYPE]:
     def append(
         self,
         inst: Instruction,
-        decision: Optional[DECISION_TYPE] = None,
+        decision: DECISION_TYPE | None = None,
     ) -> None:
         """Append a new instruction to the trace
 
@@ -134,7 +135,7 @@ def append(
         """
         _ffi_api.TraceAppend(self, inst, decision)  # type: ignore # pylint: disable=no-member
 
-    def pop(self) -> Optional[Instruction]:
+    def pop(self) -> Instruction | None:
         """Remove the last instruction, along with the decision made on that instruction, if any
 
         Returns
@@ -148,11 +149,12 @@ def apply_to_schedule(
         self,
         sch: "Schedule",
         remove_postproc: bool,
-        decision_provider: Optional[
+        decision_provider: (
             Callable[
-                [Instruction, List[INPUT_RV_TYPE], List[ATTR_TYPE], DECISION_TYPE], DECISION_TYPE
+                [Instruction, list[INPUT_RV_TYPE], list[ATTR_TYPE], DECISION_TYPE], DECISION_TYPE
             ]
-        ] = None,
+            | None
+        ) = None,
     ) -> None:
         """Apply the trace to a TensorIR schedule
 
@@ -194,7 +196,7 @@ def as_json(self, remove_postproc: bool = False) -> JSON_TYPE:
         obj = _ffi_api.TraceAsJSON(self, remove_postproc)  # type: ignore # pylint: disable=no-member
         return _json_from_tvm(obj)
 
-    def as_python(self, remove_postproc: bool = False) -> List[str]:
+    def as_python(self, remove_postproc: bool = False) -> list[str]:
         """Serialize the trace as a sequence of python statements
 
         Parameters
@@ -267,7 +269,7 @@ def apply_json_to_schedule(json_obj: JSON_TYPE, sch: "Schedule") -> None:
         """
         _ffi_api.TraceApplyJSONToSchedule(json_obj, sch)  # type: ignore # pylint: disable=no-member
 
-    def show(self, style: Optional[str] = None, black_format: bool = False) -> None:
+    def show(self, style: str | None = None, black_format: bool = False) -> None:
         """A sugar for print highlighted TVM script.
 
         Parameters
diff --git a/python/tvm/s_tir/schedule/transform.py b/python/tvm/s_tir/schedule/transform.py
index 8b2646983b26..cbefa548351d 100644
--- a/python/tvm/s_tir/schedule/transform.py
+++ b/python/tvm/s_tir/schedule/transform.py
@@ -16,8 +16,6 @@
 # under the License.
 """Transformation on TIR schedule."""
 
-from typing import Optional
-
 from tvm.s_tir.schedule import LoopRV, SBlockRV, Schedule
 
 from . import _ffi_api
@@ -25,7 +23,7 @@
 
 def tile_with_tensor_intrin(
     sch: Schedule, block: SBlockRV, intrin_name: str, allow_padding: bool = False
-) -> Optional[LoopRV]:
+) -> LoopRV | None:
     """Tile a subset of loops in the block according to the given tensor intrinsic.
 
     Parameters
diff --git a/python/tvm/s_tir/tensor_intrin/cuda.py b/python/tvm/s_tir/tensor_intrin/cuda.py
index ddbabcc4ce82..3e0a877f5aa0 100644
--- a/python/tvm/s_tir/tensor_intrin/cuda.py
+++ b/python/tvm/s_tir/tensor_intrin/cuda.py
@@ -18,7 +18,7 @@
 # ruff: noqa: E731
 """Intrinsics for tensorization on NVIDIA GPU."""
 
-from typing import Dict, Literal, Optional, Tuple
+from typing import Literal
 
 from tvm_ffi import register_global_func
 
@@ -713,8 +713,8 @@ def get_mma_intrin_group(
     trans_a: bool,
     trans_b: bool,
     not_use_mma_store_intrinic: bool = True,
-    store_to_smem_dtype: Optional[Literal["float16", "float32", "int32"]] = None,
-) -> Dict[str, str]:
+    store_to_smem_dtype: Literal["float16", "float32", "int32"] | None = None,
+) -> dict[str, str]:
     """Get a group of intrinsics for mma tensor core with the given configurations
 
     Parameters
@@ -822,7 +822,7 @@ def get_wmma_load_intrin(
     shared_scope: str,
     is_b: bool,
     is_col_major: bool,
-) -> Tuple[PrimFunc, PrimFunc]:
+) -> tuple[PrimFunc, PrimFunc]:
     """Generator of wmma_load intrins"""
     wmma_fragment_scope = f"wmma.matrix_{'b' if is_b else 'a'}"
     layout = "col_major" if is_col_major else "row_major"
@@ -899,7 +899,7 @@ def wmma_load_impl(a: T.handle, c: T.handle) -> None:
 
 def get_wmma_fill_intrin(
     m_dim: int, n_dim: int, k_dim: int, dtype: str
-) -> Tuple[PrimFunc, PrimFunc]:
+) -> tuple[PrimFunc, PrimFunc]:
     """Generator of wmma_fill intrins"""
     zero = IntImm("int32", 0).astype(dtype)
     offset_factor = n_dim
@@ -955,7 +955,7 @@ def wmma_fill_impl(c: T.handle) -> None:
 
 def get_wmma_store_intrin(
     m_dim: int, n_dim: int, k_dim: int, dtype: str, scope: str
-) -> Tuple[PrimFunc, PrimFunc]:
+) -> tuple[PrimFunc, PrimFunc]:
     """Generator of wmma_store intrins"""
     offset_factor = n_dim
 
@@ -1026,7 +1026,7 @@ def wmma_store_impl(a: T.handle, c: T.handle) -> None:
 
 def get_wmma_sync_intrin(
     m_dim: int, n_dim: int, k_dim: int, in_dtype: str, out_dtype: str, b_transposed: bool
-) -> Tuple[PrimFunc, PrimFunc]:
+) -> tuple[PrimFunc, PrimFunc]:
     """Generator of wmma_sync intrins"""
 
     def maybe_cast(v):
@@ -1380,7 +1380,7 @@ def get_wmma_intrin_group(
     in_dtype: str,
     out_dtype: str,
     trans_b: bool,
-) -> Dict[str, str]:
+) -> dict[str, str]:
     """Get a group of intrinsics for wmma tensor core with the given configurations
 
     Parameters
@@ -1473,7 +1473,7 @@ def get_index_C(elem_offset, stride):
 
 def get_mma_init_intrin(
     m_dim: int, n_dim: int, k_dim: int, dtype: str
-) -> Tuple[PrimFunc, PrimFunc]:
+) -> tuple[PrimFunc, PrimFunc]:
     """Generator of mma init intrins"""
     del k_dim  # unused
     zero = IntImm("int32", 0).astype(dtype)
@@ -1520,7 +1520,7 @@ def get_mma_load_intrin(
     shared_scope: str,
     is_b: bool,
     is_col_major: bool,
-) -> Tuple[PrimFunc, PrimFunc]:
+) -> tuple[PrimFunc, PrimFunc]:
     """Generator of mma ldmatrix intrins"""
     mma_fragment_scope = f"m16n8k8.matrix{'B' if is_b else 'A'}"
     frag_m, frag_n = (k_dim, n_dim) if is_b else (m_dim, k_dim)
@@ -1597,7 +1597,7 @@ def mma_load_impl(a: T.handle, c: T.handle) -> None:
 
 def get_mma_sync_intrin(
     m_dim: int, n_dim: int, k_dim: int, in_dtype: str, out_dtype: str, b_transposed: bool
-) -> Tuple[PrimFunc, PrimFunc]:
+) -> tuple[PrimFunc, PrimFunc]:
     """Generator of mma sync intrins"""
 
     def maybe_cast(v):
@@ -1698,7 +1698,7 @@ def mma_sync_impl(a: T.handle, b: T.handle, c: T.handle) -> None:
 
 def get_mma_store_dummy_intrin(
     m_dim: int, n_dim: int, k_dim: int, dtype: str
-) -> Tuple[PrimFunc, PrimFunc]:
+) -> tuple[PrimFunc, PrimFunc]:
     """Disable mma store intrin for now."""
     del k_dim  # unused
 
diff --git a/python/tvm/s_tir/tensor_intrin/metal.py b/python/tvm/s_tir/tensor_intrin/metal.py
index 1eb107357c2b..45c482b39e8e 100644
--- a/python/tvm/s_tir/tensor_intrin/metal.py
+++ b/python/tvm/s_tir/tensor_intrin/metal.py
@@ -17,7 +17,7 @@
 # pylint: disable=invalid-name,missing-function-docstring,unused-variable
 """Intrinsics for tensorization on Apple GPU."""
 
-from typing import Dict, Literal, Tuple
+from typing import Literal
 
 from tvm.script import tir as T
 from tvm.tir import Buffer, PrimExpr, PrimFunc, TensorIntrin
@@ -39,7 +39,7 @@ def get_simdgroup_index(buffer: Buffer, stride: PrimExpr, col: int, row: int):
 
 def get_make_filled_simdgroup_matrix_intrin(
     dtype: str, col: int = 8, row: int = 8
-) -> Tuple[PrimFunc, PrimFunc]:
+) -> tuple[PrimFunc, PrimFunc]:
     @T.prim_func
     def desc(a: T.handle) -> None:
         A = T.match_buffer(a, (col, row), dtype, scope="metal.simdgroup", offset_factor=1)
@@ -77,7 +77,7 @@ def get_simdgroup_load_intrin(
     col: int = 8,
     row: int = 8,
     transpose_matrix: bool = False,
-) -> Tuple[PrimFunc, PrimFunc]:
+) -> tuple[PrimFunc, PrimFunc]:
     align = col * row
 
     @T.prim_func
@@ -141,7 +141,7 @@ def get_simdgroup_store_intrin(
     col: int = 8,
     row: int = 8,
     transpose_matrix: bool = False,
-) -> Tuple[PrimFunc, PrimFunc]:
+) -> tuple[PrimFunc, PrimFunc]:
     align = col * row
 
     @T.prim_func
@@ -194,7 +194,7 @@ def impl(a: T.handle, c: T.handle) -> None:
 
 def get_simdgroup_multiply_accumulate_intrin(
     m_dim: int, n_dim: int, k_dim: int, dtype: str
-) -> Tuple[PrimFunc, PrimFunc]:
+) -> tuple[PrimFunc, PrimFunc]:
     @T.prim_func
     def desc(a: T.handle, b: T.handle, c: T.handle) -> None:
         A = T.match_buffer(a, (m_dim, k_dim), dtype, scope="metal.simdgroup", offset_factor=1)
@@ -297,7 +297,7 @@ def get_simdgroup_intrin_group(
     dtype: str,
     trans_a: bool = False,
     trans_b: bool = False,
-) -> Dict[str, str]:
+) -> dict[str, str]:
     """Get a group of intrinsics for tensorization on Apple GPU.
 
     Parameters
diff --git a/python/tvm/script/highlight.py b/python/tvm/script/highlight.py
index 37731ca88917..19db958b9823 100644
--- a/python/tvm/script/highlight.py
+++ b/python/tvm/script/highlight.py
@@ -23,12 +23,12 @@
 import subprocess
 import sys
 import warnings
-from typing import Any, Optional, Union
+from typing import Any, Union
 
 
 def cprint(
-    printable: Union[Any, str],
-    style: Optional[str] = None,
+    printable: Any | str,
+    style: str | None = None,
     black_format: bool = False,
 ) -> None:
     """Print TVMScript string with Pygments highlight and Black auto-formatting.
@@ -96,7 +96,7 @@ def cprint(
 
 
 @functools.lru_cache
-def _get_formatter(formatter: Optional[str] = None):
+def _get_formatter(formatter: str | None = None):
     def get_ruff_formatter():
         if shutil.which("ruff") is None:
             return None
@@ -161,7 +161,7 @@ def formatter(code_str):
     return get_fallback_formatter()
 
 
-def _format(code_str: str, formatter: Optional[str] = None) -> str:
+def _format(code_str: str, formatter: str | None = None) -> str:
     """Format a code string using Black.
 
     Parameters
@@ -186,8 +186,8 @@ def _format(code_str: str, formatter: Optional[str] = None) -> str:
 
 
 def _get_pygments_style(
-    style: Optional[str], is_in_notebook: bool
-) -> Optional[Union["pygments.style.Style", str]]:
+    style: str | None, is_in_notebook: bool
+) -> Union["pygments.style.Style", str] | None:
     """Select a pygments style to use
 
     Parameters
diff --git a/python/tvm/script/ir_builder/base.py b/python/tvm/script/ir_builder/base.py
index c1de092fbd8e..73000136877c 100644
--- a/python/tvm/script/ir_builder/base.py
+++ b/python/tvm/script/ir_builder/base.py
@@ -16,7 +16,8 @@
 # under the License.
 """A generic IRBuilder across the TVM stack"""
 
-from typing import Any, Callable, List
+from collections.abc import Callable
+from typing import Any
 
 from tvm_ffi import register_object as _register_object
 
@@ -180,9 +181,9 @@ def name(s: str, v: Any) -> Any:
 
     @staticmethod
     def name_many(  # pylint: disable=invalid-name
-        s: List[str],
-        vs: List[Any],
-    ) -> List[Any]:
+        s: list[str],
+        vs: list[Any],
+    ) -> list[Any]:
         """Set the name of a list of objects.
 
         Parameters
diff --git a/python/tvm/script/ir_builder/ir/ir.py b/python/tvm/script/ir_builder/ir/ir.py
index 995201971b8a..12e31038fc04 100644
--- a/python/tvm/script/ir_builder/ir/ir.py
+++ b/python/tvm/script/ir_builder/ir/ir.py
@@ -16,8 +16,6 @@
 # under the License.
 """Package tvm.script.ir_builder.ir.ir"""
 
-from typing import Dict, List, Optional
-
 from tvm.ir import BaseFunc, DummyGlobalInfo, GlobalInfo, GlobalVar, VDevice
 from tvm.runtime import Object as tvm_Object
 
@@ -76,7 +74,7 @@ def def_function(func_name: str, func: BaseFunc) -> None:
     return _ffi_api.DefFunction(func_name, func)  # type: ignore[attr-defined] # pylint: disable=no-member
 
 
-def module_attrs(attrs: Dict[str, tvm_Object], allow_overwrite=False) -> None:
+def module_attrs(attrs: dict[str, tvm_Object], allow_overwrite=False) -> None:
     """Specify the attrs of the ir_module frame.
     Parameters
     ----------
@@ -98,7 +96,7 @@ def current_ir_module() -> IRModuleFrame:
     return _ffi_api.CurrentIRModule()  # type: ignore[attr-defined] # pylint: disable=no-member
 
 
-def module_get_attrs() -> Dict[str, tvm_Object]:
+def module_get_attrs() -> dict[str, tvm_Object]:
     """Get the attrs of the ir_module frame.
     Returns
     -------
@@ -108,7 +106,7 @@ def module_get_attrs() -> Dict[str, tvm_Object]:
     return _ffi_api.ModuleGetAttrs()  # type: ignore[attr-defined] # pylint: disable=no-member
 
 
-def module_get_attr(attr_key: str) -> Optional[tvm_Object]:
+def module_get_attr(attr_key: str) -> tvm_Object | None:
     """Get the specified attr of the ir_module frame.
     Parameters
     ----------
@@ -123,7 +121,7 @@ def module_get_attr(attr_key: str) -> Optional[tvm_Object]:
 
 
 def module_set_attr(
-    attr_key: str, attr_value: Optional[tvm_Object], allow_overwrite: bool = False
+    attr_key: str, attr_value: tvm_Object | None, allow_overwrite: bool = False
 ) -> None:
     """Set the specified attr of the ir_module frame.
     Parameters
@@ -138,7 +136,7 @@ def module_set_attr(
     return _ffi_api.ModuleSetAttr(attr_key, attr_value, allow_overwrite)  # type: ignore[attr-defined] # pylint: disable=no-member
 
 
-def module_global_infos(global_infos: Dict[str, List[GlobalInfo]]) -> None:
+def module_global_infos(global_infos: dict[str, list[GlobalInfo]]) -> None:
     """Specify the global infos of the ir_module frame.
     Parameters
     ----------
@@ -180,7 +178,7 @@ def vdevice(target=None, vdevice_id: int = 0, memory_scope: str = "global") -> V
     return VDevice(target, vdevice_id, memory_scope)  # type: ignore[attr-defined] # pylint: disable=no-member
 
 
-def lookup_vdevice(target_kind: Optional[str] = None, device_index: int = -1) -> VDevice:
+def lookup_vdevice(target_kind: str | None = None, device_index: int = -1) -> VDevice:
     """Retrieve a virtual device from the globalinfo vdevice list.
     Parameters
     ----------
diff --git a/python/tvm/script/ir_builder/relax/distributed/ir.py b/python/tvm/script/ir_builder/relax/distributed/ir.py
index b7d1a015707d..49bfb6139a97 100644
--- a/python/tvm/script/ir_builder/relax/distributed/ir.py
+++ b/python/tvm/script/ir_builder/relax/distributed/ir.py
@@ -20,7 +20,7 @@
 """IRBuilder for distributed Relax dialect"""
 
 from numbers import Number
-from typing import List, Optional, Tuple, Union
+from typing import Optional, Union
 
 import numpy as _np  # type: ignore
 
@@ -51,10 +51,10 @@
 
 @args_converter.auto
 def call_tir(
-    func: Union[str, Expr],
+    func: str | Expr,
     args: Expr,
-    out_sinfo: Union[DTensorStructInfo, List[DTensorStructInfo]],
-    tir_vars: Optional[Union[ShapeExpr, Tuple[PrimExpr], List[PrimExpr]]] = None,
+    out_sinfo: DTensorStructInfo | list[DTensorStructInfo],
+    tir_vars: ShapeExpr | tuple[PrimExpr] | list[PrimExpr] | None = None,
 ) -> Call:
     """Distributed version of call_tir
 
@@ -88,14 +88,14 @@ def call_tir(
     if not isinstance(out_sinfo, list):
         out_sinfo = [out_sinfo]
 
-    if isinstance(tir_vars, (list, tuple)):
+    if isinstance(tir_vars, list | tuple):
         tir_vars = ShapeExpr(tir_vars)
 
     return _ffi_api.call_tir_dist(func, args, out_sinfo, tir_vars)  # type: ignore
 
 
 def const(
-    value: Union[bool, int, float, _np.ndarray, tvm.runtime.Tensor],
+    value: bool | int | float | _np.ndarray | tvm.runtime.Tensor,
     struct_info: DTensorStructInfo,
 ) -> Constant:
     """Create a constant value.
@@ -121,10 +121,10 @@ def const(
     if not isinstance(struct_info, DTensorStructInfo):
         raise TypeError("struct_info needs to be an instance of DTensorStructInfo. ")
     dtype = str(struct_info.tensor_sinfo.dtype)
-    if isinstance(value, (Number, (bool, list))):
+    if isinstance(value, Number | (bool | list)):
         value = _np.array(value, dtype=dtype)
 
-    if isinstance(value, (_np.ndarray, _np.generic)):
+    if isinstance(value, _np.ndarray | _np.generic):
         if dtype is not None:
             value = value.astype(dtype)
         value = _tensor.tensor(value)
@@ -150,7 +150,7 @@ def _lookup_device_mesh(device_mesh_str: py_str) -> DeviceMesh:
 
 
 def annotate_sharding(
-    value: Expr, device_mesh: Union[py_str, DeviceMesh], placement: Union[py_str, Placement]
+    value: Expr, device_mesh: py_str | DeviceMesh, placement: py_str | Placement
 ) -> Expr:
     if isinstance(device_mesh, py_str):
         device_mesh = _lookup_device_mesh(device_mesh)
@@ -160,7 +160,7 @@ def annotate_sharding(
 
 
 def redistribute(
-    value: Expr, device_mesh: Union[py_str, DeviceMesh], placement: Union[py_str, Placement]
+    value: Expr, device_mesh: py_str | DeviceMesh, placement: py_str | Placement
 ) -> Expr:
     if isinstance(device_mesh, py_str):
         device_mesh = _lookup_device_mesh(device_mesh)
diff --git a/python/tvm/script/ir_builder/relax/ir.py b/python/tvm/script/ir_builder/relax/ir.py
index 6bff4ef76a26..a7428c32a30b 100644
--- a/python/tvm/script/ir_builder/relax/ir.py
+++ b/python/tvm/script/ir_builder/relax/ir.py
@@ -20,7 +20,8 @@
 import builtins
 import functools
 import inspect
-from typing import Any, Callable, Dict, List, Optional, Tuple, Type, Union
+from collections.abc import Callable
+from typing import Any
 
 import tvm
 from tvm import DataType, relax
@@ -227,7 +228,7 @@
 ################################ Device ################################
 
 
-def to_vdevice(data: Expr, dst_vdevice: Union[py_str, VDevice]) -> Expr:
+def to_vdevice(data: Expr, dst_vdevice: py_str | VDevice) -> Expr:
     """Copy data to the destination device.
 
     Parameters
@@ -304,7 +305,7 @@ def func_name(name: py_str) -> None:
     return _ffi_api.FuncName(name)  # type: ignore[attr-defined] # pylint: disable=no-member
 
 
-def func_attr(attrs: Dict[py_str, tvm_Object]) -> None:
+def func_attr(attrs: dict[py_str, tvm_Object]) -> None:
     """Specify the attrs of the last function frame.
     Parameters
     ----------
@@ -334,7 +335,7 @@ def func_ret_value(value: Expr) -> None:
     return _ffi_api.FuncRetValue(value)  # type: ignore[attr-defined] # pylint: disable=no-member
 
 
-def rewriter(rewriter_mod: Union[IRModule, Type]) -> PatternMatchingRewriter:
+def rewriter(rewriter_mod: IRModule | type) -> PatternMatchingRewriter:
     """Define a pattern-rewrite rule
 
     The IRModule must have two publicly-exposed functions, `pattern`
@@ -389,7 +390,7 @@ def dataflow() -> frame.BindingBlockFrame:
     return _ffi_api.Dataflow()  # type: ignore[attr-defined] # pylint: disable=no-member
 
 
-def output(*vars: Tuple[Var]) -> None:
+def output(*vars: tuple[Var]) -> None:
     """Expose the dataflow block output variables as global ones.
     Parameters
     ----------
@@ -406,7 +407,7 @@ def output(*vars: Tuple[Var]) -> None:
 def call_packed(
     func: py_str,
     *args: Expr,
-    sinfo_args: Optional[Union[StructInfo, List[StructInfo]]] = None,
+    sinfo_args: StructInfo | list[StructInfo] | None = None,
     **kwargs: Any,
 ) -> Call:
     """Create a relax Call, which calls a packed function.
@@ -463,7 +464,7 @@ def call_packed(
 def call_py_func(
     py_func_name: py_str,
     *args: Expr,
-    out_sinfo: Union[StructInfo, List[StructInfo]],
+    out_sinfo: StructInfo | list[StructInfo],
 ) -> Call:
     """Create a relax Call, which calls a Python function.
 
@@ -514,7 +515,7 @@ def _sinfo_arg_wrapper(func):
     """A wrapper to convert StructInfoProxies to StructInfo for builtin operators with sinfo_args"""
 
     def _convert_tensor_type(args):
-        if isinstance(args, (list, py_tuple)):  # type: ignore
+        if isinstance(args, list | py_tuple):  # type: ignore
             new_args = [_convert_tensor_type(x) for x in args]
             return type(args)(new_args)
         if isinstance(args, dict):
@@ -540,7 +541,7 @@ def wrapped(*args, **kwargs):
 ############################### Emits ###############################
 
 
-def emit(value: Expr, annotate_struct_info: Optional[StructInfo] = None) -> Var:
+def emit(value: Expr, annotate_struct_info: StructInfo | None = None) -> Var:
     """Emit a binding to the last binding block frame.
     Parameters
     ----------
@@ -627,7 +628,7 @@ def emit_var_binding(value: VarBinding) -> Var:
 def emit_with_sinfo(
     op: str,
     args: Expr,
-    sinfo_args: Optional[Union[StructInfo, List[StructInfo]]] = None,
+    sinfo_args: StructInfo | list[StructInfo] | None = None,
 ) -> Call:
     """Create a relax Call with sinfo_args.
     Parameters
@@ -664,7 +665,7 @@ def SeqExpr() -> frame.SeqExprFrame:  # pylint: disable=invalid-name
 ############################# If Then Else #############################
 
 
-def If(condition: Union[Expr, PrimExpr]) -> frame.IfFrame:  # pylint: disable=invalid-name
+def If(condition: Expr | PrimExpr) -> frame.IfFrame:  # pylint: disable=invalid-name
     """Create an if frame.
 
     Parameters
@@ -729,7 +730,7 @@ def tuple(*fields: Expr) -> Expr:
 ############################### R.shape ################################
 
 
-def shape(value: List[PrimExpr]) -> Expr:
+def shape(value: list[PrimExpr]) -> Expr:
     """Create a ShapeExpr.
     Parameters
     ----------
@@ -774,7 +775,7 @@ def str(value: py_str) -> Expr:
     return relax.StringImm(value)  # type: ignore[attr-defined] # pylint: disable=no-member
 
 
-def dtype(value: Union[py_str, DataType]) -> Expr:
+def dtype(value: py_str | DataType) -> Expr:
     """Create a dtype imm expression.
     Parameters
     ----------
diff --git a/python/tvm/script/ir_builder/tir/external_kernel.py b/python/tvm/script/ir_builder/tir/external_kernel.py
index b5fe0b6537a4..63d30f809e73 100644
--- a/python/tvm/script/ir_builder/tir/external_kernel.py
+++ b/python/tvm/script/ir_builder/tir/external_kernel.py
@@ -22,7 +22,7 @@
 import os
 import tempfile
 from pathlib import Path
-from typing import Any, Dict, List, Tuple, Union
+from typing import Any
 
 from tvm import __version__ as tvm_version
 from tvm import tir
@@ -35,7 +35,7 @@ class BaseKernel:  # pylint: disable=too-few-public-methods
 
     def compile_to_device_module(
         self, launch_args, *args, **kwargs
-    ) -> Tuple[str, Module, List[Any]]:
+    ) -> tuple[str, Module, list[Any]]:
         """Compile the kernel to a device module."""
         raise NotImplementedError()
 
@@ -113,10 +113,10 @@ def __init__(self, source_code: str):
 
     def compile_to_device_module(  # pylint: disable=arguments-differ
         self,
-        grid: List[List[Union[int, tir.PrimExpr]]],
-        *args: List[Any],
-        **kwargs: Dict[str, Any],
-    ) -> Tuple[str, Module, List[Any]]:
+        grid: list[list[int | tir.PrimExpr]],
+        *args: list[Any],
+        **kwargs: dict[str, Any],
+    ) -> tuple[str, Module, list[Any]]:
         """Compile the kernel to a device module."""
         from tvm.relax.frontend.nn import (  # pylint: disable=import-outside-toplevel
             SourceModule,
@@ -128,7 +128,7 @@ def compile_to_device_module(  # pylint: disable=arguments-differ
             "['blockIdx.x', 'blockIdx.y', 'blockIdx.z'] and "
             "['threadIdx.x', 'threadIdx.y', 'threadIdx.z']"
         )
-        assert isinstance(grid[0], (list, tuple)) and isinstance(grid[1], (list, tuple))
+        assert isinstance(grid[0], list | tuple) and isinstance(grid[1], list | tuple)
         launch_param_tags = ["blockIdx.x", "blockIdx.y", "blockIdx.z"][: len(grid[0])] + [
             "threadIdx.x",
             "threadIdx.y",
@@ -182,9 +182,9 @@ def compile_to_device_module(  # pylint: disable=arguments-differ
 
 def call_kernel(
     kernel,
-    launch_args: List[Union[int, tir.PrimExpr, List[Union[int, tir.PrimExpr]]]],
-    *args: List[Any],
-    **kwargs: Dict[str, Any],
+    launch_args: list[int | tir.PrimExpr | list[int | tir.PrimExpr]],
+    *args: list[Any],
+    **kwargs: dict[str, Any],
 ):
     """
     Call an external kernel.
@@ -225,7 +225,7 @@ def call_kernel(
     )
 
     # Attach the kernel module to the current IRModule
-    external_mods: List[Module] = module_get_attr("external_mods") or []
+    external_mods: list[Module] = module_get_attr("external_mods") or []
     kernel_exists = any([mod.implements_function(kernel_name) for mod in external_mods])
     if kernel_exists:
         logging.debug("Kernel %s already exists in the IRModule", kernel_name)
diff --git a/python/tvm/script/ir_builder/tir/frame.py b/python/tvm/script/ir_builder/tir/frame.py
index b3af71137df9..a31c72e271d4 100644
--- a/python/tvm/script/ir_builder/tir/frame.py
+++ b/python/tvm/script/ir_builder/tir/frame.py
@@ -16,8 +16,6 @@
 # under the License.
 """IRBuilder for TIR"""
 
-from typing import List, Union
-
 from tvm_ffi import register_object as _register_object
 
 from tvm.tir import Buffer, Var
@@ -43,7 +41,7 @@ class BlockInitFrame(TIRFrame): ...
 
 @_register_object("script.ir_builder.tir.ForFrame")
 class ForFrame(TIRFrame):
-    def __enter__(self) -> Union[Var, List[Var]]:  # type: ignore[override]
+    def __enter__(self) -> Var | list[Var]:  # type: ignore[override]
         super().__enter__()
         return self.vars if len(self.vars) > 1 else self.vars[0]
 
diff --git a/python/tvm/script/ir_builder/tir/ir.py b/python/tvm/script/ir_builder/tir/ir.py
index a3b521f955fa..1d259e192a8f 100644
--- a/python/tvm/script/ir_builder/tir/ir.py
+++ b/python/tvm/script/ir_builder/tir/ir.py
@@ -20,13 +20,13 @@
 import contextlib
 import functools
 import inspect
-import sys
 import threading
+from collections.abc import Callable
 from numbers import Integral
-from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+from typing import Any, ParamSpec, TypeVar
 
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 # isort: on
 
@@ -117,16 +117,16 @@ def block_name_suffix_context(block_suffix: str):
 
 
 def buffer(
-    shape: Union[List[PrimExpr], Tuple[PrimExpr], PrimExpr, Integral],
+    shape: list[PrimExpr] | tuple[PrimExpr] | PrimExpr | Integral,
     dtype: str = "float32",
     data: Var = None,
-    strides: Optional[List[PrimExpr]] = None,
+    strides: list[PrimExpr] | None = None,
     elem_offset: PrimExpr = None,
     scope: str = "global",
     align: int = 0,
     offset_factor: int = 0,
     buffer_type: str = "",
-    axis_separators: Optional[List[int]] = None,
+    axis_separators: list[int] | None = None,
 ) -> Buffer:
     """The buffer declaration function.
 
@@ -167,7 +167,7 @@ def buffer(
     res : Buffer
         The declared buffer.
     """
-    shape = (shape,) if isinstance(shape, (PrimExpr, Integral)) else shape
+    shape = (shape,) if isinstance(shape, PrimExpr | Integral) else shape
     if strides is not None:
         strides = [Var(s, "int32") if isinstance(s, str) else s for s in strides]
     else:
@@ -210,7 +210,7 @@ def prim_func(is_private: bool = False) -> frame.PrimFuncFrame:
     return _ffi_api.PrimFunc(is_private)  # type: ignore[attr-defined] # pylint: disable=no-member
 
 
-def arg(name: str, obj: Union[Var, Buffer]) -> Union[Var, Buffer]:
+def arg(name: str, obj: Var | Buffer) -> Var | Buffer:
     """The PrimFunc arguments adding function.
 
     Parameters
@@ -240,7 +240,7 @@ def func_name(name: str) -> None:
     _ffi_api.FuncName(name)  # type: ignore[attr-defined] # pylint: disable=no-member
 
 
-def func_attr(attrs: Dict[str, Any]) -> None:
+def func_attr(attrs: dict[str, Any]) -> None:
     """The PrimFunc annotation statement.
 
     Parameters
@@ -268,17 +268,17 @@ def func_ret(ret_type: Type) -> Type:
 
 
 def match_buffer(
-    param: Union[Var, BufferLoad, BufferRegion],
-    shape: Union[List[PrimExpr], Tuple[PrimExpr], PrimExpr, Integral] = None,
+    param: Var | BufferLoad | BufferRegion,
+    shape: list[PrimExpr] | tuple[PrimExpr] | PrimExpr | Integral = None,
     dtype: str = "float32",
     data: Var = None,
-    strides: Optional[List[PrimExpr]] = None,
+    strides: list[PrimExpr] | None = None,
     elem_offset: PrimExpr = None,
     scope: str = "global",
     align: int = -1,
     offset_factor: int = 0,
     buffer_type: str = "default",
-    axis_separators: Optional[List[int]] = None,
+    axis_separators: list[int] | None = None,
 ) -> Buffer:
     """The buffer match function.
 
@@ -344,7 +344,7 @@ def match_buffer(
             shape = [region.extent for region in param.region]
         else:
             raise ValueError("Shape must be specified when binding input param")
-    shape = (shape,) if isinstance(shape, (PrimExpr, Integral)) else shape
+    shape = (shape,) if isinstance(shape, PrimExpr | Integral) else shape
     if strides is not None:
         idx_dtype = shape[0].dtype if isinstance(shape[0], PrimExpr) else "int32"
         strides = [Var(s, idx_dtype) if isinstance(s, str) else s for s in strides]
@@ -398,7 +398,7 @@ def init() -> frame.BlockInitFrame:
     return _ffi_api.Init()  # type: ignore[attr-defined] # pylint: disable=no-member
 
 
-def where(predicate: Union[PrimExpr, int]) -> None:
+def where(predicate: PrimExpr | int) -> None:
     """The block predicate statement.
 
     Parameters
@@ -416,7 +416,7 @@ def where(predicate: Union[PrimExpr, int]) -> None:
     _ffi_api.Where(predicate)  # type: ignore[attr-defined] # pylint: disable=no-member
 
 
-def reads(*buffer_slices: List[Union[BufferRegion, BufferLoad]]) -> None:
+def reads(*buffer_slices: list[BufferRegion | BufferLoad]) -> None:
     """The block buffer region reading statement.
 
     Parameters
@@ -436,7 +436,7 @@ def reads(*buffer_slices: List[Union[BufferRegion, BufferLoad]]) -> None:
     _ffi_api.Reads(buffer_slices)  # type: ignore[attr-defined] # pylint: disable=no-member
 
 
-def writes(*buffer_slices: List[Union[BufferRegion, BufferLoad]]) -> None:
+def writes(*buffer_slices: list[BufferRegion | BufferLoad]) -> None:
     """The block buffer region writing statement.
 
     Parameters
@@ -456,7 +456,7 @@ def writes(*buffer_slices: List[Union[BufferRegion, BufferLoad]]) -> None:
     _ffi_api.Writes(buffer_slices)  # type: ignore[attr-defined] # pylint: disable=no-member
 
 
-def sblock_attr(attrs: Dict[str, Any]) -> None:
+def sblock_attr(attrs: dict[str, Any]) -> None:
     """The block annotation statement.
 
     Parameters
@@ -468,16 +468,16 @@ def sblock_attr(attrs: Dict[str, Any]) -> None:
 
 
 def alloc_buffer(
-    shape: Union[List[PrimExpr], Tuple[PrimExpr], PrimExpr, Integral],
+    shape: list[PrimExpr] | tuple[PrimExpr] | PrimExpr | Integral,
     dtype: str = "float32",
     data: Var = None,
-    strides: Optional[List[PrimExpr]] = None,
+    strides: list[PrimExpr] | None = None,
     elem_offset: PrimExpr = None,
     scope: str = "global",
     align: int = -1,
     offset_factor: int = 0,
     buffer_type: str = "default",
-    axis_separators: Optional[List[int]] = None,
+    axis_separators: list[int] | None = None,
 ) -> Buffer:
     """The buffer alllocation function.
 
@@ -518,7 +518,7 @@ def alloc_buffer(
     res : Buffer
         The allocated buffer.
     """
-    shape = (shape,) if isinstance(shape, (PrimExpr, Integral)) else shape
+    shape = (shape,) if isinstance(shape, PrimExpr | Integral) else shape
     if strides is not None:
         strides = [Var(s, "int32") if isinstance(s, str) else s for s in strides]
     else:
@@ -537,7 +537,7 @@ def alloc_buffer(
     )
 
 
-def _as_range(dom: Union[ir.Range, List[PrimExpr]]) -> ir.Range:
+def _as_range(dom: ir.Range | list[PrimExpr]) -> ir.Range:
     """The range constructor.
 
     Parameters
@@ -552,7 +552,7 @@ def _as_range(dom: Union[ir.Range, List[PrimExpr]]) -> ir.Range:
     """
     if isinstance(dom, ir.Range):
         return dom
-    if isinstance(dom, (list, tuple)):
+    if isinstance(dom, list | tuple):
         from tvm.arith import Analyzer  # pylint: disable=import-outside-toplevel
 
         extent = Analyzer().simplify(dom[1] - dom[0])
@@ -569,7 +569,7 @@ class axis:  # pylint: disable=invalid-name
 
     @staticmethod
     def spatial(
-        dom: Union[ir.Range, List[PrimExpr], Tuple[PrimExpr]],
+        dom: ir.Range | list[PrimExpr] | tuple[PrimExpr],
         binding: PrimExpr,
         dtype: str = "int32",
     ) -> Var:
@@ -597,7 +597,7 @@ def spatial(
 
     @staticmethod
     def reduce(
-        dom: Union[ir.Range, List[PrimExpr], Tuple[PrimExpr]],
+        dom: ir.Range | list[PrimExpr] | tuple[PrimExpr],
         binding: PrimExpr,
         dtype: str = "int32",
     ) -> Var:
@@ -625,7 +625,7 @@ def reduce(
 
     @staticmethod
     def scan(
-        dom: Union[ir.Range, List[PrimExpr], Tuple[PrimExpr]],
+        dom: ir.Range | list[PrimExpr] | tuple[PrimExpr],
         binding: PrimExpr,
         dtype: str = "int32",
     ) -> Var:
@@ -653,7 +653,7 @@ def scan(
 
     @staticmethod
     def opaque(
-        dom: Union[ir.Range, List[PrimExpr], Tuple[PrimExpr]],
+        dom: ir.Range | list[PrimExpr] | tuple[PrimExpr],
         binding: PrimExpr,
         dtype: str = "int32",
     ) -> Var:
@@ -680,7 +680,7 @@ def opaque(
         )
 
     @staticmethod
-    def remap(kinds: str, bindings: List[PrimExpr], dtype: str = "int32") -> Union[List[Var], Var]:
+    def remap(kinds: str, bindings: list[PrimExpr], dtype: str = "int32") -> list[Var] | Var:
         """The block axis remapping function.
 
         Parameters
@@ -712,8 +712,8 @@ def serial(
     start: PrimExpr,
     stop: PrimExpr = None,
     *,
-    annotations: Optional[Dict[str, Any]] = None,
-    step: Optional[PrimExpr] = None,
+    annotations: dict[str, Any] | None = None,
+    step: PrimExpr | None = None,
 ) -> frame.ForFrame:
     """The serial For statement.
 
@@ -749,8 +749,8 @@ def parallel(
     start: PrimExpr,
     stop: PrimExpr = None,
     *,
-    annotations: Optional[Dict[str, Any]] = None,
-    step: Optional[PrimExpr] = None,
+    annotations: dict[str, Any] | None = None,
+    step: PrimExpr | None = None,
 ) -> frame.ForFrame:
     """The parallel For statement.
 
@@ -786,8 +786,8 @@ def vectorized(
     start: PrimExpr,
     stop: PrimExpr = None,
     *,
-    annotations: Optional[Dict[str, Any]] = None,
-    step: Optional[PrimExpr] = None,
+    annotations: dict[str, Any] | None = None,
+    step: PrimExpr | None = None,
 ) -> frame.ForFrame:
     """The vectorized For statement.
 
@@ -823,8 +823,8 @@ def unroll(
     start: PrimExpr,
     stop: PrimExpr = None,
     *,
-    annotations: Optional[Dict[str, Any]] = None,
-    step: Optional[PrimExpr] = None,
+    annotations: dict[str, Any] | None = None,
+    step: PrimExpr | None = None,
 ) -> frame.ForFrame:
     """The unrolled For statement.
 
@@ -859,9 +859,9 @@ def unroll(
 def thread_binding(
     start: PrimExpr,
     stop: PrimExpr = None,
-    thread: Optional[str] = None,
+    thread: str | None = None,
     *,
-    annotations: Optional[Dict[str, Any]] = None,
+    annotations: dict[str, Any] | None = None,
 ) -> frame.ForFrame:
     """The thread-binding For statement.
 
@@ -943,9 +943,9 @@ def Assert(condition: PrimExpr, message: str) -> frame.AssertFrame:  # pylint: d
 
 def LetStmt(  # pylint: disable=invalid-name
     value: PrimExpr,
-    type_annotation: Optional[Type] = None,  # pylint: disable=redefined-outer-name
+    type_annotation: Type | None = None,  # pylint: disable=redefined-outer-name
     *,
-    var: Optional[Var] = None,  # pylint: disable=redefined-outer-name
+    var: Var | None = None,  # pylint: disable=redefined-outer-name
 ) -> frame.LetFrame:
     """Create a LetStmt binding
 
@@ -974,7 +974,7 @@ def LetStmt(  # pylint: disable=invalid-name
 
 def Let(  # pylint: disable=invalid-name
     expr: PrimExpr,
-    where: Dict[Var, PrimExpr],  # pylint: disable=redefined-outer-name
+    where: dict[Var, PrimExpr],  # pylint: disable=redefined-outer-name
 ) -> PrimExpr:
     """Create a Let expression binding"""
     assert len(where) == 1, "T.Let only allows `where` to have exactly one element"
@@ -1021,7 +1021,7 @@ def let_stmt(v: Var, value: PrimExpr) -> frame.LetFrame:
 
 
 def allocate(
-    extents: List[PrimExpr],
+    extents: list[PrimExpr],
     dtype: str,
     scope: str = "global",
     condition: PrimExpr = None,
@@ -1053,7 +1053,7 @@ def allocate(
     )
 
 
-def attr(node: Any, attr_key: str, value: Union[PrimExpr, str]) -> frame.AttrFrame:
+def attr(node: Any, attr_key: str, value: PrimExpr | str) -> frame.AttrFrame:
     """Create an attribute node.
 
     Parameters
@@ -1187,7 +1187,7 @@ def decl_buffer(
     res : frame.DeclBufferFrame
         The result DeclBufferFrame.
     """
-    shape = (shape,) if isinstance(shape, (PrimExpr, Integral)) else shape
+    shape = (shape,) if isinstance(shape, PrimExpr | Integral) else shape
     if strides is not None:
         strides = [Var(s, "int32") if isinstance(s, str) else s for s in strides]
     else:
@@ -1208,7 +1208,7 @@ def decl_buffer(
 
 
 def launch_thread(
-    thread: Union[IterVar, str],  # pylint: disable=redefined-outer-name
+    thread: IterVar | str,  # pylint: disable=redefined-outer-name
     extent: PrimExpr,
 ) -> frame.LaunchThreadFrame:
     """Launch a thread.
@@ -1265,8 +1265,8 @@ def env_thread(thread_tag: str, dtype: str = "int32") -> IterVar:
 def buffer_store(
     buffer: Buffer,  # pylint: disable=redefined-outer-name
     value: PrimExpr,
-    indices: List[Union[PrimExpr, slice]],
-    predicate: Optional[PrimExpr] = None,
+    indices: list[PrimExpr | slice],
+    predicate: PrimExpr | None = None,
 ) -> None:
     """Buffer store node.
 
@@ -1288,7 +1288,7 @@ def buffer_store(
     """
     from tvm.arith import Analyzer  # pylint: disable=import-outside-toplevel
 
-    if not isinstance(indices, (list, tuple, ir.Array)):
+    if not isinstance(indices, list | tuple | ir.Array):
         indices = [indices]
 
     expr_indices = []
@@ -1336,13 +1336,7 @@ def func_gen(name: str):
     """
 
     def func(
-        expr: Union[
-            None,
-            PrimExpr,
-            Literal["inf", "-inf", "nan"],
-            int,
-            float,
-        ] = None,
+        expr: None | PrimExpr | Literal["inf", "-inf", "nan"] | int | float = None,
         *,
         is_size_var: bool = False,
     ) -> PrimExpr:
@@ -1521,7 +1515,7 @@ def func(
 # pylint: enable=invalid-name
 
 
-def boolean(expr: Optional[PrimExpr] = None, is_size_var: bool = False) -> PrimExpr:
+def boolean(expr: PrimExpr | None = None, is_size_var: bool = False) -> PrimExpr:
     """Construct a new tir.Var with type boolean or cast expression to type boolean.
 
     Parameters
@@ -1541,7 +1535,7 @@ def boolean(expr: Optional[PrimExpr] = None, is_size_var: bool = False) -> PrimE
 
 
 def handle(
-    dtype: Optional[str] = None, storage_scope: str = "global", *, is_size_var: bool = False
+    dtype: str | None = None, storage_scope: str = "global", *, is_size_var: bool = False
 ) -> Var:
     """Create a TIR var that represents a pointer.
 
@@ -1574,7 +1568,7 @@ def handle(
     )
 
 
-def void(expr: Optional[PrimExpr] = None, *, is_size_var: bool = False) -> PrimExpr:
+def void(expr: PrimExpr | None = None, *, is_size_var: bool = False) -> PrimExpr:
     """Construct a new tir.Var with type void or cast expression to type void.
 
     Parameters
@@ -1690,7 +1684,7 @@ def max(a: PrimExpr, b: PrimExpr) -> PrimExpr:  # pylint: disable=redefined-buil
     return _ffi_api.max(a, b)  # type: ignore[attr-defined] # pylint: disable=no-member
 
 
-def iter_var(v: Union[Var, str], dom: ir.Range, iter_type: str, thread_tag: str) -> IterVar:
+def iter_var(v: Var | str, dom: ir.Range, iter_type: str, thread_tag: str) -> IterVar:
     """The iteration variable.
 
     Parameters
@@ -1716,7 +1710,7 @@ def iter_var(v: Union[Var, str], dom: ir.Range, iter_type: str, thread_tag: str)
     return IterVar(dom, v, iter_type, thread_tag)
 
 
-def comm_reducer(combiner: Callable, identity: List[PrimExpr]) -> CommReducer:
+def comm_reducer(combiner: Callable, identity: list[PrimExpr]) -> CommReducer:
     """
     Create a CommReducer from lambda inputs/outputs and the identities
 
@@ -1750,7 +1744,7 @@ def comm_reducer(combiner: Callable, identity: List[PrimExpr]) -> CommReducer:
 def index_map(
     mapping: Callable,
     *,
-    inverse_index_map: Optional[Callable] = None,
+    inverse_index_map: Callable | None = None,
     index_dtype: str = "int64",
 ) -> IndexMap:
     """Create a TIR Index mapping"""
@@ -1758,8 +1752,8 @@ def index_map(
 
 
 def target(
-    target_config: Union[Dict, str],
-    host: Optional[Union[Dict, str, Target]] = None,
+    target_config: dict | str,
+    host: dict | str | Target | None = None,
 ) -> Target:
     """
     Create a target
@@ -1777,11 +1771,11 @@ def target(
     res : Target
         The target.
     """
-    if not isinstance(target_config, (str, dict)):
+    if not isinstance(target_config, str | dict):
         raise ValueError(
             f"T.target expected a config dict or string, but got {type(target_config)}"
         )
-    if host is not None and not isinstance(host, (str, dict, Target)):
+    if host is not None and not isinstance(host, str | dict | Target):
         raise ValueError(
             "T.target expected the host to be "
             "a config dict, string, or T.target, "
@@ -1835,31 +1829,18 @@ def f():
 # pylint: disable=invalid-name
 
 
-if sys.version_info >= (3, 10):
-    from typing import ParamSpec, TypeVar  # pylint: disable=import-error
+T = TypeVar("T")
+P = ParamSpec("P")
 
-    T = TypeVar("T")
-    P = ParamSpec("P")
 
-    def _op_wrapper(func: Callable[P, T]) -> Callable[P, T]:
-        @functools.wraps(func)
-        def wrapped(*args, **kwargs) -> T:
-            if "dtype" in kwargs:
-                kwargs.pop("dtype")
-            return func(*args, **kwargs)
-
-        return wrapped
-
-else:
-
-    def _op_wrapper(func):
-        @functools.wraps(func)
-        def wrapped(*args, **kwargs):
-            if "dtype" in kwargs:
-                kwargs.pop("dtype")
-            return func(*args, **kwargs)
+def _op_wrapper(func: Callable[P, T]) -> Callable[P, T]:
+    @functools.wraps(func)
+    def wrapped(*args, **kwargs) -> T:
+        if "dtype" in kwargs:
+            kwargs.pop("dtype")
+        return func(*args, **kwargs)
 
-        return wrapped
+    return wrapped
 
 
 abs = _op_wrapper(_tir_op.abs)  # pylint: disable=redefined-builtin
diff --git a/python/tvm/script/ir_builder/tir/triton.py b/python/tvm/script/ir_builder/tir/triton.py
index 386f5fe90e50..75184d291645 100644
--- a/python/tvm/script/ir_builder/tir/triton.py
+++ b/python/tvm/script/ir_builder/tir/triton.py
@@ -17,7 +17,7 @@
 # ruff: noqa: RUF005
 """Triton kernel integration with TIR"""
 
-from typing import Any, Dict, List, Tuple, Union
+from typing import Any
 
 import triton
 from packaging import version
@@ -45,10 +45,10 @@ def __init__(self, func):
 
     def compile_to_device_module(
         self,
-        launch_args: List[Union[int, tir.PrimExpr]],
-        *args: List[Any],
-        **kwargs: Dict[str, Any],
-    ) -> Tuple[str, Module, List[Any]]:
+        launch_args: list[int | tir.PrimExpr],
+        *args: list[Any],
+        **kwargs: dict[str, Any],
+    ) -> tuple[str, Module, list[Any]]:
         """Compile the kernel to a device module.
 
         Parameters
@@ -93,7 +93,7 @@ def compile_to_device_module(
 
     def _generate_triton_kernel(
         self, func, *args, **kwargs
-    ) -> Tuple["triton.compiler.CompiledKernel", List[tir.PrimExpr]]:
+    ) -> tuple["triton.compiler.CompiledKernel", list[tir.PrimExpr]]:
         """Deduce the kernel signature and generate the Triton kernel"""
 
         kernel_params = func.params
diff --git a/python/tvm/script/ir_builder/tir/utils.py b/python/tvm/script/ir_builder/tir/utils.py
index 7fcb0f0ee3d8..84fab649c9c6 100644
--- a/python/tvm/script/ir_builder/tir/utils.py
+++ b/python/tvm/script/ir_builder/tir/utils.py
@@ -17,7 +17,6 @@
 """Utility helpers for TIR IRBuilder."""
 
 import contextlib
-from typing import List
 
 from tvm import tir
 from tvm.tir import Buffer
@@ -59,7 +58,7 @@ class _FrameScope:
     """
 
     def __init__(self, frames):
-        self.frames = frames if isinstance(frames, (list, tuple)) else [frames]
+        self.frames = frames if isinstance(frames, list | tuple) else [frames]
         self._stack = None
 
     def __enter__(self):
@@ -72,7 +71,7 @@ def __exit__(self, *args):
         return self._stack.__exit__(*args)
 
 
-def frame_scope(frames: List[frame.TIRFrame]) -> _FrameScope:
+def frame_scope(frames: list[frame.TIRFrame]) -> _FrameScope:
     """Enter multiple IRBuilder frames without deep nesting.
 
     This function provides a way to enter multiple frames in a single `with`
diff --git a/python/tvm/script/parser/core/diagnostics.py b/python/tvm/script/parser/core/diagnostics.py
index 7d2b5cb6688f..4e5ca08aae91 100644
--- a/python/tvm/script/parser/core/diagnostics.py
+++ b/python/tvm/script/parser/core/diagnostics.py
@@ -19,7 +19,6 @@
 
 import inspect
 import sys
-from typing import Union
 
 from tvm.ir import IRModule, SourceName, Span, diagnostics
 
@@ -55,7 +54,7 @@ class Source:
     source: str
     full_source: str
 
-    def __init__(self, program: Union[str, doc.AST]):
+    def __init__(self, program: str | doc.AST):
         if isinstance(program, str):
             self.source_name = "<str>"
             self.start_line = 1
diff --git a/python/tvm/script/parser/core/dispatch.py b/python/tvm/script/parser/core/dispatch.py
index f803be05de92..a8dcf0fe7dbc 100644
--- a/python/tvm/script/parser/core/dispatch.py
+++ b/python/tvm/script/parser/core/dispatch.py
@@ -16,7 +16,8 @@
 # under the License.
 """Parser dispatching infrastructure"""
 
-from typing import TYPE_CHECKING, Any, Callable, Dict, Optional, Tuple, Type
+from collections.abc import Callable
+from typing import TYPE_CHECKING, Any
 
 from .doc import AST
 
@@ -25,10 +26,10 @@
 
 
 ParseMethod = Callable[["Parser", AST], None]
-ParseVTable: Dict[Tuple[str, str], ParseMethod] = {}
+ParseVTable: dict[tuple[str, str], ParseMethod] = {}
 
 OpMethod = Callable[..., Any]
-OpVTable: Dict[Tuple[Type, AST, int], OpMethod] = {}
+OpVTable: dict[tuple[type, AST, int], OpMethod] = {}
 
 
 def register(token: str, type_name: str):
@@ -65,8 +66,8 @@ def func(method: ParseMethod):
 def get(
     token: str,
     type_name: str,
-    default: Optional[ParseMethod] = None,
-) -> Optional[ParseMethod]:
+    default: ParseMethod | None = None,
+) -> ParseMethod | None:
     """Get a registered method for a dispatch token and type name,
     or return a default method if no registered methods with this dispatch token and type name.
 
@@ -89,7 +90,7 @@ def get(
     return ParseVTable.get((token, type_name), default)
 
 
-def register_op(operand_type: Type, op_node_type: AST, operand_index: int):
+def register_op(operand_type: type, op_node_type: AST, operand_index: int):
     """Register a method for a operand type, AST operator node and operand index.
 
     Parameters
@@ -124,11 +125,11 @@ def func(method: OpMethod):
 
 
 def get_op(
-    operand_type: Type,
-    op_node_type: Type,
+    operand_type: type,
+    op_node_type: type,
     operand_index: int,
-    default: Optional[OpMethod] = None,
-) -> Optional[OpMethod]:
+    default: OpMethod | None = None,
+) -> OpMethod | None:
     """Register a method for a operand type, AST operator node and operand index.
 
     Parameters
diff --git a/python/tvm/script/parser/core/doc.py b/python/tvm/script/parser/core/doc.py
index 9a0680395c2b..8c94d445c288 100644
--- a/python/tvm/script/parser/core/doc.py
+++ b/python/tvm/script/parser/core/doc.py
@@ -41,8 +41,8 @@ class Entry:
         The callable methods for converting doc AST to python AST node.
     """
 
-    to_doc: typing.Optional[FnToDoc]
-    from_doc: typing.Optional[FnFromDoc]
+    to_doc: FnToDoc | None
+    from_doc: FnFromDoc | None
 
     def __init__(self):
         self.to_doc = None
@@ -64,7 +64,7 @@ class Registry:
     """
 
     _inst: typing.Optional["Registry"] = None
-    table: typing.Dict[str, Entry]
+    table: dict[str, Entry]
 
     def __init__(self):
         self.table = defaultdict(Entry)
@@ -118,14 +118,7 @@ def _is_atomic_type(node):
         or node in [..., True, False]
         or isinstance(
             node,
-            (
-                int,
-                float,
-                str,
-                bool,
-                bytes,
-                complex,
-            ),
+            int | float | str | bool | bytes | complex,
         )
     )
 
@@ -236,7 +229,7 @@ class NodeVisitor:
     """Node visitor for doc AST"""
 
     def visit(self, node: doc.AST) -> None:
-        if isinstance(node, (list, tuple)):
+        if isinstance(node, list | tuple):
             for item in node:
                 self.visit(item)
             return
@@ -253,7 +246,7 @@ def generic_visit(self, node: doc.AST) -> None:
             value = getattr(node, field, None)
             if value is None:
                 pass
-            elif isinstance(value, (doc.AST, list, tuple)):
+            elif isinstance(value, doc.AST | list | tuple):
                 self.visit(value)
 
 
@@ -274,12 +267,12 @@ def visit(self, node: doc.AST) -> doc.AST:
         )(node)
 
     def generic_visit(self, node: doc.AST) -> doc.AST:
-        kv: typing.Dict[str, typing.Any] = {}
+        kv: dict[str, typing.Any] = {}
         for field in node.__class__._FIELDS:  # pylint: disable=protected-access
             value = getattr(node, field, None)
             if value is None:
                 pass
-            elif isinstance(value, (doc.AST, list, tuple)):
+            elif isinstance(value, doc.AST | list | tuple):
                 value = self.visit(value)
             kv[field] = value
         return node.__class__(**kv)
diff --git a/python/tvm/script/parser/core/entry.py b/python/tvm/script/parser/core/entry.py
index c25db1e5c1e6..e1d9260c805d 100644
--- a/python/tvm/script/parser/core/entry.py
+++ b/python/tvm/script/parser/core/entry.py
@@ -17,7 +17,7 @@
 """The entry point of TVM parser."""
 
 import inspect
-from typing import Any, Dict, Optional, Union
+from typing import Any
 
 import tvm
 
@@ -36,7 +36,7 @@
 )
 
 
-def _default_globals() -> Dict[str, Any]:
+def _default_globals() -> dict[str, Any]:
     # lazy import here to avoid circular deps
     from tvm.script.parser import (
         ir,  # pylint: disable=import-outside-toplevel
@@ -56,7 +56,7 @@ def _default_globals() -> Dict[str, Any]:
     return extra_vars
 
 
-def scan_macro(program: Union[Any, str], extra_vars: Optional[Dict[str, Any]] = None) -> Any:
+def scan_macro(program: Any | str, extra_vars: dict[str, Any] | None = None) -> Any:
     """Generate the AST, and the source code for __repr__."""
     # The AST will be converted into TIR at the time of expansion.
     source = Source(program)
@@ -65,8 +65,8 @@ def scan_macro(program: Union[Any, str], extra_vars: Optional[Dict[str, Any]] =
 
 
 def parse(
-    program: Union[doc.AST, Any, str],
-    extra_vars: Optional[Dict[str, Any]] = None,
+    program: doc.AST | Any | str,
+    extra_vars: dict[str, Any] | None = None,
     check_well_formed: bool = True,
 ) -> Any:
     """Register a method for a operand type, AST operator node and operand index.
@@ -120,7 +120,7 @@ def parse(
 
         source_ast = source.as_ast()
 
-        if isinstance(ret, (IRModule, tvm.relax.Function)) and not tvm.relax.analysis.well_formed(
+        if isinstance(ret, IRModule | tvm.relax.Function) and not tvm.relax.analysis.well_formed(
             ret
         ):
             parser.report_error(source_ast, err=WELL_FORMED_ERROR_MESSAGE)
diff --git a/python/tvm/script/parser/core/evaluator.py b/python/tvm/script/parser/core/evaluator.py
index c69d2a328b9d..a8ce4ce6c1a1 100644
--- a/python/tvm/script/parser/core/evaluator.py
+++ b/python/tvm/script/parser/core/evaluator.py
@@ -17,7 +17,8 @@
 """AST Evaluation"""
 
 import ast
-from typing import TYPE_CHECKING, Any, Callable, Dict, List, Optional, Type, Union
+from collections.abc import Callable
+from typing import TYPE_CHECKING, Any
 
 import tvm
 
@@ -27,7 +28,7 @@
 if TYPE_CHECKING:
     from .parser import Parser
 
-DEFAULT_OP: Dict[Type, Callable[..., Any]] = {
+DEFAULT_OP: dict[type, Callable[..., Any]] = {
     doc.Add: lambda a, b: a + b,
     doc.Sub: lambda a, b: a - b,
     doc.Mult: lambda a, b: a * b,
@@ -87,17 +88,17 @@ class ExprEvaluator:
     """
 
     parser: "Parser"
-    value_table: Dict[str, Any]
+    value_table: dict[str, Any]
     new_value_count: int
 
-    def __init__(self, parser: "Parser", value_table: Dict[str, Any]) -> None:
+    def __init__(self, parser: "Parser", value_table: dict[str, Any]) -> None:
         super().__init__()
         self.parser = parser
         self.value_table = value_table
         self.new_value_count = 0
 
     @staticmethod
-    def eval(parser: "Parser", value_table: Dict[str, Any], node: doc.AST) -> Any:
+    def eval(parser: "Parser", value_table: dict[str, Any], node: doc.AST) -> Any:
         """Expression evaluation for TVMScript parser.
 
         Parameters
@@ -175,7 +176,7 @@ def _visit(self, node: doc.AST) -> Any:
             isinstance(node, doc.Call)
             and hasattr(node.func, "attr")
             and node.func.attr not in ["reads", "writes", "match_buffer"]
-        ) or isinstance(node, (doc.BinOp, doc.UnaryOp, doc.Compare, doc.BoolOp, doc.IfExp)):
+        ) or isinstance(node, doc.BinOp | doc.UnaryOp | doc.Compare | doc.BoolOp | doc.IfExp):
             if isinstance(node, doc.BinOp):
                 args = [node.left, node.right]
             elif isinstance(node, doc.UnaryOp):
@@ -189,7 +190,7 @@ def _visit(self, node: doc.AST) -> Any:
             elif isinstance(node, doc.BoolOp):
                 args = node.values
         for arg in args:
-            if isinstance(arg, doc.Subscript) and isinstance(arg.slice, (doc.Slice, doc.Tuple)):
+            if isinstance(arg, doc.Subscript) and isinstance(arg.slice, doc.Slice | doc.Tuple):
                 if isinstance(arg.slice, doc.Slice):
                     check_slices = [arg.slice]
                 else:
@@ -218,17 +219,10 @@ def _visit(self, node: doc.AST) -> Any:
             return node
         if isinstance(
             node,
-            (
-                doc.Constant,
-                doc.expr_context,
-                doc.operator,
-                doc.boolop,
-                doc.unaryop,
-                doc.cmpop,
-            ),
+            doc.Constant | doc.expr_context | doc.operator | doc.boolop | doc.unaryop | doc.cmpop,
         ):
             return node
-        if not isinstance(node, (doc.expr, doc.Slice)):
+        if not isinstance(node, doc.expr | doc.Slice):
             return node
         if isinstance(node, doc.Lambda):
             return self._eval_lambda(node)
@@ -246,7 +240,7 @@ def _visit(self, node: doc.AST) -> Any:
         fields = {}
         for field in node.__class__._FIELDS:  # pylint: disable=protected-access
             attr = getattr(node, field)
-            if isinstance(attr, (doc.AST, tuple, list)):
+            if isinstance(attr, doc.AST | tuple | list):
                 fields[field] = self._visit(attr)
             else:
                 fields[field] = attr
@@ -288,7 +282,7 @@ def _eval_lambda(self, node: doc.Lambda) -> Any:
             self.parser.report_error(node, err)
         return self._add_intermediate_result(value)
 
-    def _eval_bool_op(self, fields: Dict[str, Any]) -> Any:
+    def _eval_bool_op(self, fields: dict[str, Any]) -> Any:
         """The doc AST boolean operator node evaluating method.
 
         Parameters
@@ -303,14 +297,14 @@ def _eval_bool_op(self, fields: Dict[str, Any]) -> Any:
             The evaluation result.
         """
         op = fields["op"]
-        if not isinstance(op, (doc.And, doc.Or)):
+        if not isinstance(op, doc.And | doc.Or):
             raise TypeError(f"Unexpected operator: {op}")
         value = self._eval_expr(fields["values"][0])
         for rhs in fields["values"][1:]:
             value = _eval_op(op, values=[value, self._eval_expr(rhs)])
         return value
 
-    def _eval_compare(self, fields: Dict[str, Any]) -> Any:
+    def _eval_compare(self, fields: dict[str, Any]) -> Any:
         """The doc AST comparison operation node evaluating method.
 
         Parameters
@@ -337,7 +331,7 @@ def _eval_compare(self, fields: Dict[str, Any]) -> Any:
                 result = _eval_op(doc.And(), values=[result, sub_result])
         return result
 
-    def _eval_unary_op(self, fields: Dict[str, Any]) -> Any:
+    def _eval_unary_op(self, fields: dict[str, Any]) -> Any:
         """The doc AST unary operation node evaluating method.
 
         Parameters
@@ -355,7 +349,7 @@ def _eval_unary_op(self, fields: Dict[str, Any]) -> Any:
         value = _eval_op(fields["op"], values=[value])
         return value
 
-    def _eval_bin_op(self, fields: Dict[str, Any]) -> Any:
+    def _eval_bin_op(self, fields: dict[str, Any]) -> Any:
         """The doc AST binary operation node evaluating method.
 
         Parameters
@@ -377,7 +371,7 @@ def _eval_bin_op(self, fields: Dict[str, Any]) -> Any:
             ],
         )
 
-    def _eval_if_exp(self, fields: Dict[str, Any]) -> Any:
+    def _eval_if_exp(self, fields: dict[str, Any]) -> Any:
         """The doc AST if-else expression node evaluating method.
 
         Parameters
@@ -401,7 +395,7 @@ def _eval_if_exp(self, fields: Dict[str, Any]) -> Any:
         else:
             raise TypeError(f"Expected Python bool or TIR bool, but got {type(test)}")
 
-    def _eval_slice(self, fields: Dict[str, Any]) -> slice:
+    def _eval_slice(self, fields: dict[str, Any]) -> slice:
         """The doc AST slice node evaluating method.
 
         Parameters
@@ -441,8 +435,8 @@ def _eval_expr(self, v: Any) -> Any:
 
 def eval_expr(
     parser: "Parser",
-    node: Union[doc.expr, doc.Expression],
-    dict_globals: Optional[Dict[str, Any]],
+    node: doc.expr | doc.Expression,
+    dict_globals: dict[str, Any] | None,
 ) -> Any:
     """Expression evaluation for TVMScript parser.
 
@@ -472,7 +466,7 @@ def eval_assign(
     parser: "Parser",
     target: doc.expr,
     source: Any,
-) -> Dict[str, Any]:
+) -> dict[str, Any]:
     """Expression assignment evaluation for TVMScript parser.
 
     Parameters
@@ -499,8 +493,8 @@ def eval_assign(
 
 
 def _eval_expr(
-    node: Union[doc.expr, doc.Expression],
-    dict_globals: Optional[Dict[str, Any]],
+    node: doc.expr | doc.Expression,
+    dict_globals: dict[str, Any] | None,
 ) -> Any:
     """Expression evaluation implementation for TVMScript parser.
 
@@ -530,7 +524,7 @@ def _eval_expr(
 
 def _eval_op(
     op: doc.AST,
-    values: List[Any],
+    values: list[Any],
 ):
     """Operation expression evaluation implementation for TVMScript parser.
 
@@ -563,7 +557,7 @@ def _eval_op(
 def _eval_assign(
     target: doc.expr,
     source: Any,
-) -> Dict[str, Any]:
+) -> dict[str, Any]:
     """Expression assignment evaluation implementation for TVMScript parser.
 
     Parameters
diff --git a/python/tvm/script/parser/core/parser.py b/python/tvm/script/parser/core/parser.py
index 289c88ef222b..eb617c85c9bd 100644
--- a/python/tvm/script/parser/core/parser.py
+++ b/python/tvm/script/parser/core/parser.py
@@ -19,8 +19,9 @@
 import abc
 import inspect
 from collections import defaultdict
+from collections.abc import Callable
 from contextlib import contextmanager
-from typing import Any, Callable, Dict, List, Optional, Set, Union
+from typing import Any
 
 import numpy as np
 
@@ -95,7 +96,7 @@ class ScriptMacro(abc.ABC):
     def __init__(
         self,
         source: Source,
-        closure_vars: Dict[str, Any],
+        closure_vars: dict[str, Any],
         func: Callable,
         hygienic: bool,
     ) -> None:
@@ -180,7 +181,7 @@ class VarTableFrame:
         The set of variable names in the variable table frame.
     """
 
-    vars: Set[str]
+    vars: set[str]
 
     def __init__(self):
         self.vars = set()
@@ -223,8 +224,8 @@ class VarTable:
         The dictionary for variable table name-based query.
     """
 
-    frames: List[VarTableFrame]
-    name2value: Dict[str, List[Any]]
+    frames: list[VarTableFrame]
+    name2value: dict[str, list[Any]]
 
     def __init__(self):
         self.frames = []
@@ -273,7 +274,7 @@ def add(self, var: str, value: Any, allow_shadowing: bool = False):
             self.frames[-1].add(var)
             self.name2value[var].append(value)
 
-    def get(self) -> Dict[str, Any]:
+    def get(self) -> dict[str, Any]:
         """Get a variable dictionary of latest variables.
 
         Returns
@@ -339,17 +340,17 @@ class Parser(doc.NodeVisitor):
     """
 
     diag: Diagnostics
-    dispatch_tokens: List[str]
-    function_annotations: Optional[Dict[str, Dict[str, Any]]]
+    dispatch_tokens: list[str]
+    function_annotations: dict[str, dict[str, Any]] | None
     var_table: VarTable
     inside_function: bool  # whether we are within a function
-    current_class: Optional[str] = None  # current class being parsed
+    current_class: str | None = None  # current class being parsed
     base_py_module_context: bool = False  # whether current class inherits from BasePyModule
 
     def __init__(
         self,
         source: Source,
-        function_annotations: Dict[str, Dict[str, Any]],
+        function_annotations: dict[str, dict[str, Any]],
     ) -> None:
         self.diag = Diagnostics(source)
         self.dispatch_tokens = ["default"]
@@ -357,7 +358,7 @@ def __init__(
         self.var_table = VarTable()
         self.inside_function = False
 
-    def parse(self, extra_vars: Optional[Dict[str, Any]] = None) -> Any:
+    def parse(self, extra_vars: dict[str, Any] | None = None) -> Any:
         """The main parse method for parser.
 
         Parameters
@@ -429,7 +430,7 @@ def set_class_context(self, class_name: str, is_base_py_module: bool = False):
         self.current_class = class_name
         self.base_py_module_context = is_base_py_module
 
-    def _get_current_class_context(self) -> Optional[str]:
+    def _get_current_class_context(self) -> str | None:
         """Get the current class context.
 
         Returns
@@ -473,8 +474,8 @@ def pop_source():
 
     def eval_expr(
         self,
-        node: Union[doc.Expression, doc.expr],
-        extra_vars: Optional[Dict[str, Any]] = None,
+        node: doc.Expression | doc.expr,
+        extra_vars: dict[str, Any] | None = None,
     ) -> Any:
         """Expression evaluation when parsing.
 
@@ -498,7 +499,7 @@ def eval_expr(
         var_values[ScriptMacro.parser_object_name] = self
         return eval_expr(self, node, var_values)
 
-    def _duplicate_lhs_check(self, target: doc.expr) -> Union[bool, Set[str]]:
+    def _duplicate_lhs_check(self, target: doc.expr) -> bool | set[str]:
         """Check whether duplicate lhs exists in assignment.
 
         Parameters
@@ -512,8 +513,8 @@ def _duplicate_lhs_check(self, target: doc.expr) -> Union[bool, Set[str]]:
             The result of true if duplicate lhs exists,
             or the set of lhs names if no duplicate lhs exists.
         """
-        if isinstance(target, (doc.Tuple, doc.List)):
-            vars: Set[str] = set()  # pylint: disable=redefined-builtin
+        if isinstance(target, doc.Tuple | doc.List):
+            vars: set[str] = set()  # pylint: disable=redefined-builtin
             for i in target.elts:
                 res = self._duplicate_lhs_check(i)
                 if isinstance(res, bool) and res:
@@ -537,7 +538,7 @@ def eval_assign(
         source: Any,
         bind_value: Callable[["Parser", doc.expr, str, Any], Any],
         allow_shadowing: bool = False,
-    ) -> Dict[str, Any]:
+    ) -> dict[str, Any]:
         """Expression assignment evaluation when parsing.
 
         Parameters
@@ -567,7 +568,7 @@ def eval_assign(
             self.var_table.add(k, var, allow_shadowing)
         return var_values
 
-    def report_error(self, node: doc.AST, err: Union[Exception, str]) -> None:  # pylint: disable=no-self-use
+    def report_error(self, node: doc.AST, err: Exception | str) -> None:  # pylint: disable=no-self-use
         """The error reporting when parsing.
 
         Parameters
@@ -617,7 +618,7 @@ def visit(self, node: doc.AST) -> None:
         res : Any
             The visiting result.
         """
-        if isinstance(node, (list, tuple)):
+        if isinstance(node, list | tuple):
             for item in node:
                 self.visit(item)
             return
@@ -635,7 +636,7 @@ def visit(self, node: doc.AST) -> None:
         except Exception as err:  # pylint: disable=broad-except
             self.report_error(node, err)
 
-    def visit_body(self, node: List[doc.stmt]) -> Any:
+    def visit_body(self, node: list[doc.stmt]) -> Any:
         """The general body visiting method.
 
         Parameters
diff --git a/python/tvm/script/parser/core/utils.py b/python/tvm/script/parser/core/utils.py
index 0810ead3231f..85190b96d9ce 100644
--- a/python/tvm/script/parser/core/utils.py
+++ b/python/tvm/script/parser/core/utils.py
@@ -17,8 +17,9 @@
 """TVM Script Parser utils"""
 
 import inspect
+from collections.abc import Callable
 from types import FrameType
-from typing import Any, Callable, Dict, List
+from typing import Any
 
 from .diagnostics import findsource
 
@@ -47,7 +48,7 @@ def get_func_nonlocals(func):
     return nonlocal_vars
 
 
-def inspect_function_capture(func: Callable) -> Dict[str, Any]:
+def inspect_function_capture(func: Callable) -> dict[str, Any]:
     """Capture function non-locals and global variables.
 
     Parameters
@@ -67,7 +68,7 @@ def inspect_function_capture(func: Callable) -> Dict[str, Any]:
     return captured
 
 
-def inspect_class_capture(cls: type) -> Dict[str, Any]:
+def inspect_class_capture(cls: type) -> dict[str, Any]:
     """Capture class non-locals and global variables.
 
     Parameters
@@ -80,7 +81,7 @@ def inspect_class_capture(cls: type) -> Dict[str, Any]:
     res : Dict[str, Any]
         The class variables map with non-local or global variables.
     """
-    result: Dict[str, Any] = {}
+    result: dict[str, Any] = {}
     for _, v in cls.__dict__.items():
         if inspect.isfunction(v):
             func_vars = inspect_function_capture(v)
@@ -88,7 +89,7 @@ def inspect_class_capture(cls: type) -> Dict[str, Any]:
     return result
 
 
-def is_defined_in_class(frames: List[FrameType], obj: Any) -> bool:
+def is_defined_in_class(frames: list[FrameType], obj: Any) -> bool:
     """Check whether a object is defined in a class scope.
 
     Parameters
diff --git a/python/tvm/script/parser/ir/entry.py b/python/tvm/script/parser/ir/entry.py
index e4a891fc34ff..8f7a5be663cf 100644
--- a/python/tvm/script/parser/ir/entry.py
+++ b/python/tvm/script/parser/ir/entry.py
@@ -18,7 +18,7 @@
 """The entry point of TVM parser for ir module."""
 
 import inspect
-from typing import Callable, Optional, Type
+from collections.abc import Callable
 
 from tvm import cpu, ir
 from tvm.ir import GlobalVar, IRModule
@@ -29,7 +29,7 @@
 # this formulation allows us to support having @I.ir_module
 # appear as a decorator by itself or to have optional arguments
 # like @I.ir_module(check_well_formed=False)
-def ir_module(mod: Optional[Type] = None, check_well_formed: bool = True) -> IRModule:
+def ir_module(mod: type | None = None, check_well_formed: bool = True) -> IRModule:
     """The parsing method for ir module, by using `@ir_module` as decorator.
 
     Parameters
diff --git a/python/tvm/script/parser/relax/dist.py b/python/tvm/script/parser/relax/dist.py
index d4848eb80d1f..2aa0397b96a2 100644
--- a/python/tvm/script/parser/relax/dist.py
+++ b/python/tvm/script/parser/relax/dist.py
@@ -17,7 +17,7 @@
 # pylint: disable=redefined-builtin,missing-docstring, invalid-name, unused-import, redefined-outer-name
 # ruff: noqa: F401
 
-from typing import Any, Dict, List, Optional, Set, Union
+from typing import Any, Optional, Union
 
 from tvm.ir import Range
 from tvm.relax import TensorStructInfo
@@ -55,10 +55,10 @@ def __init__(
         self.tensor_sinfo_proxy = tensor_sinfo_proxy
         super().__init__()
 
-    def get_symbolic_vars(self) -> Set[str]:
+    def get_symbolic_vars(self) -> set[str]:
         return self.tensor_sinfo_proxy.get_symbolic_vars()
 
-    def as_struct_info(self, dict_globals: Optional[Dict[str, Any]] = None) -> TensorStructInfo:
+    def as_struct_info(self, dict_globals: dict[str, Any] | None = None) -> TensorStructInfo:
         return DTensorStructInfo(
             self.tensor_sinfo_proxy.as_struct_info(dict_globals),
             self.device_mesh,
@@ -67,10 +67,10 @@ def as_struct_info(self, dict_globals: Optional[Dict[str, Any]] = None) -> Tenso
 
 
 def DTensor(
-    shape: Optional[List[Union[PrimExpr, str]]] = None,
-    dtype: Optional[str] = None,
-    device_mesh: Union[DeviceMesh, str] = DeviceMesh([], Range(0, 1)),
-    placement: Union[Placement, str] = "",
+    shape: list[PrimExpr | str] | None = None,
+    dtype: str | None = None,
+    device_mesh: DeviceMesh | str = DeviceMesh([], Range(0, 1)),
+    placement: Placement | str = "",
     *,
     ndim: int = -1,
 ) -> DTensorProxy:
@@ -81,7 +81,7 @@ def DTensor(
         dtype = shape
         shape = None
 
-    if shape is not None and not isinstance(shape, (tuple, list)):
+    if shape is not None and not isinstance(shape, tuple | list):
         raise ValueError(f"shape must be a list or tuple, but got: {shape}")
     if isinstance(device_mesh, str):
         if not IRBuilder.is_in_scope():
diff --git a/python/tvm/script/parser/relax/entry.py b/python/tvm/script/parser/relax/entry.py
index 475e12261783..324a11daff45 100644
--- a/python/tvm/script/parser/relax/entry.py
+++ b/python/tvm/script/parser/relax/entry.py
@@ -16,8 +16,8 @@
 # under the License.
 # pylint: disable=missing-docstring, invalid-name
 import inspect
-from typing import Any, Dict, List, Optional, Set, TypeVar, Union
-from typing import Callable as _Callable
+from collections.abc import Callable as _Callable
+from typing import Any, TypeVar
 
 import tvm
 from tvm.relax import (
@@ -52,8 +52,8 @@
 # appear as a decorator by itself or to have optional arguments
 # like @R.function(pure=False)
 def function(
-    f: Optional[FType] = None, pure: bool = True, private: bool = False, check_well_formed=True
-) -> Union[Function, FType]:
+    f: FType | None = None, pure: bool = True, private: bool = False, check_well_formed=True
+) -> Function | FType:
     # pylint: disable=unused-argument
     # (pure and private aren't used here, but are used later in parsing)
 
@@ -147,10 +147,10 @@ def wrapper(*args, **kwargs):
 
 
 class StructInfoProxy(ObjectConvertible):
-    def as_struct_info(self, dict_globals: Optional[Dict[str, Any]] = None) -> StructInfo:
+    def as_struct_info(self, dict_globals: dict[str, Any] | None = None) -> StructInfo:
         raise NotImplementedError()
 
-    def get_symbolic_vars(self) -> Set[str]:
+    def get_symbolic_vars(self) -> set[str]:
         return {}
 
     def asobject(self):
@@ -175,10 +175,10 @@ class ObjectProxy(StructInfoProxy):
     def __init__(self) -> None:
         pass
 
-    def get_symbolic_vars(self) -> Set[str]:
+    def get_symbolic_vars(self) -> set[str]:
         return set()
 
-    def as_struct_info(self, dict_globals: Optional[Dict[str, Any]] = None) -> ShapeStructInfo:
+    def as_struct_info(self, dict_globals: dict[str, Any] | None = None) -> ShapeStructInfo:
         return ObjectStructInfo()
 
 
@@ -189,7 +189,7 @@ def Object() -> ObjectProxy:
 ############################### R.Tensor ###############################
 
 
-def _eval_shape(expr: Union[str, PrimExpr], dict_globals: Optional[Dict[str, Any]]) -> PrimExpr:
+def _eval_shape(expr: str | PrimExpr, dict_globals: dict[str, Any] | None) -> PrimExpr:
     if isinstance(expr, str):
         code = compile(expr, "<string>", "eval")
         return eval(code, dict_globals or {})  # pylint: disable=eval-used
@@ -198,20 +198,20 @@ def _eval_shape(expr: Union[str, PrimExpr], dict_globals: Optional[Dict[str, Any
 
 
 class TensorProxy(StructInfoProxy):
-    shape: Optional[List[Union[str, PrimExpr]]]
+    shape: list[str | PrimExpr] | None
     dtype: str
-    vdevice: Optional[str]
+    vdevice: str | None
     ndim: int
 
     def __init__(
         self,
-        shape: Optional[Union[List[Union[PrimExpr, str]], Expr]] = None,
-        dtype: Optional[str] = None,
-        vdevice: Optional[str] = None,
+        shape: list[PrimExpr | str] | Expr | None = None,
+        dtype: str | None = None,
+        vdevice: str | None = None,
         ndim: int = -1,
     ) -> None:
         if isinstance(shape, Expr):
-            if not isinstance(shape, (ShapeExpr, Var)):
+            if not isinstance(shape, ShapeExpr | Var):
                 raise ValueError(
                     "When the shape is an Expr, it must be a ShapeExpr or a Var with ShapeExpr "
                     f"value. But got: {shape} with type: {type(shape)}"
@@ -226,13 +226,13 @@ def __init__(
         self.vdevice = vdevice
         self.ndim = ndim
 
-    def get_symbolic_vars(self) -> Set[str]:
+    def get_symbolic_vars(self) -> set[str]:
         if self.shape is None or isinstance(self.shape, Expr):
             return {}
         else:
             return {s for s in self.shape if isinstance(s, str) and s.isidentifier()}
 
-    def as_struct_info(self, dict_globals: Optional[Dict[str, Any]] = None) -> TensorStructInfo:
+    def as_struct_info(self, dict_globals: dict[str, Any] | None = None) -> TensorStructInfo:
         vdev = self.vdevice
         if isinstance(self.vdevice, str):
             if ":" in self.vdevice:
@@ -243,7 +243,7 @@ def as_struct_info(self, dict_globals: Optional[Dict[str, Any]] = None) -> Tenso
 
         if self.shape is None:
             return TensorStructInfo(None, self.dtype, vdev, self.ndim)
-        elif isinstance(self.shape, (ShapeExpr, Var)):
+        elif isinstance(self.shape, ShapeExpr | Var):
             return TensorStructInfo(self.shape, self.dtype, vdev, self.ndim)
         else:
             if dict_globals is None and any([isinstance(s, str) for s in self.shape]):
@@ -256,9 +256,9 @@ def as_struct_info(self, dict_globals: Optional[Dict[str, Any]] = None) -> Tenso
 
 
 def Tensor(
-    shape: Optional[Union[List[Union[PrimExpr, str]], Expr]] = None,
-    dtype: Optional[str] = None,
-    vdevice: Optional[str] = None,
+    shape: list[PrimExpr | str] | Expr | None = None,
+    dtype: str | None = None,
+    vdevice: str | None = None,
     ndim: int = -1,
 ) -> TensorProxy:
     # scalar tensor case
@@ -268,7 +268,7 @@ def Tensor(
         dtype = shape
         shape = None
 
-    if shape is not None and not isinstance(shape, (tuple, list)) and not isinstance(shape, Expr):
+    if shape is not None and not isinstance(shape, tuple | list) and not isinstance(shape, Expr):
         raise ValueError(f"shape must be a list/tuple or an Expr, but got: {shape}")
     return TensorProxy(shape, dtype, vdevice, ndim)
 
@@ -277,10 +277,10 @@ def Tensor(
 
 
 class CallableProxy(StructInfoProxy):
-    params: List[StructInfoProxy]
+    params: list[StructInfoProxy]
     ret: StructInfoProxy
     purity: bool
-    derive_func: Optional[Union[str, tvm.ir.EnvFunc]]
+    derive_func: str | tvm.ir.EnvFunc | None
 
     """Function type.
 
@@ -311,15 +311,15 @@ class CallableProxy(StructInfoProxy):
 
     def __init__(
         self,
-        params: Optional[Union[StructInfoProxy, List[StructInfoProxy]]] = None,
-        ret: Optional[StructInfoProxy] = None,
-        purity: Optional[bool] = None,
-        derive_func: Optional[Union[str, tvm.ir.EnvFunc]] = None,
+        params: StructInfoProxy | list[StructInfoProxy] | None = None,
+        ret: StructInfoProxy | None = None,
+        purity: bool | None = None,
+        derive_func: str | tvm.ir.EnvFunc | None = None,
     ) -> None:
         if params is None:
             self.params = params
         else:
-            if not isinstance(params, (list, tuple)):
+            if not isinstance(params, list | tuple):
                 params = [params]
             # convert `R.Callable` to `R.Callable()`
             self.params = [param() if callable(param) else param for param in params]
@@ -334,13 +334,13 @@ def __init__(
         self.purity = purity
         self.derive_func = derive_func
 
-    def get_symbolic_vars(self) -> Set[str]:
+    def get_symbolic_vars(self) -> set[str]:
         if self.params is None:
             return set()
         else:
             return set().union(*[p.get_symbolic_vars() for p in self.params])
 
-    def as_struct_info(self, dict_globals: Optional[Dict[str, Any]] = None) -> FuncStructInfo:
+    def as_struct_info(self, dict_globals: dict[str, Any] | None = None) -> FuncStructInfo:
         if self.ret is None:
             ret = None
         else:
@@ -360,10 +360,10 @@ def as_struct_info(self, dict_globals: Optional[Dict[str, Any]] = None) -> FuncS
 
 
 def Callable(
-    params: Optional[Union[StructInfoProxy, List[StructInfoProxy]]] = None,
-    ret: Optional[StructInfoProxy] = None,
-    purity: Optional[bool] = None,
-    derive_func: Optional[Union[str, tvm.ir.EnvFunc]] = None,
+    params: StructInfoProxy | list[StructInfoProxy] | None = None,
+    ret: StructInfoProxy | None = None,
+    purity: bool | None = None,
+    derive_func: str | tvm.ir.EnvFunc | None = None,
 ) -> CallableProxy:
     return CallableProxy(params, ret, purity=purity, derive_func=derive_func)
 
@@ -372,7 +372,7 @@ def Callable(
 
 
 class TupleProxy(StructInfoProxy):
-    fields: List[StructInfoProxy]
+    fields: list[StructInfoProxy]
     """The type of tuple values.
 
     Parameters
@@ -383,22 +383,22 @@ class TupleProxy(StructInfoProxy):
 
     def __init__(
         self,
-        *fields: List[StructInfoProxy],
+        *fields: list[StructInfoProxy],
     ) -> None:
-        if len(fields) == 1 and isinstance(fields[0], (tuple, list)):
+        if len(fields) == 1 and isinstance(fields[0], tuple | list):
             fields = fields[0]
         # convert `R.Tensor` to `R.Tensor()`
         self.fields = [field() if callable(field) else field for field in fields]
 
-    def get_symbolic_vars(self) -> Set[str]:
+    def get_symbolic_vars(self) -> set[str]:
         return set().union(*[f.get_symbolic_vars() for f in self.fields])
 
-    def as_struct_info(self, dict_globals: Optional[Dict[str, Any]] = None) -> TupleStructInfo:
+    def as_struct_info(self, dict_globals: dict[str, Any] | None = None) -> TupleStructInfo:
         fields = [field.as_struct_info(dict_globals) for field in self.fields]
         return TupleStructInfo(fields)
 
 
-def Tuple(*fields: List[StructInfoProxy]) -> TupleProxy:
+def Tuple(*fields: list[StructInfoProxy]) -> TupleProxy:
     return TupleProxy(*fields)
 
 
@@ -406,7 +406,7 @@ def Tuple(*fields: List[StructInfoProxy]) -> TupleProxy:
 
 
 class ShapeProxy(StructInfoProxy):
-    values: Optional[List[PrimExpr]]
+    values: list[PrimExpr] | None
     ndim: int
     """The type of shape values.
 
@@ -421,24 +421,24 @@ class ShapeProxy(StructInfoProxy):
 
     def __init__(
         self,
-        values: Optional[List[PrimExpr]] = None,
+        values: list[PrimExpr] | None = None,
         ndim: int = -1,
     ) -> None:
         self.values = values
         self.ndim = ndim
 
-    def get_symbolic_vars(self) -> Set[str]:
+    def get_symbolic_vars(self) -> set[str]:
         if self.values is None:
             return set()
         else:
             return {v for v in self.values if isinstance(v, str) and v.isidentifier()}
 
-    def as_struct_info(self, dict_globals: Optional[Dict[str, Any]] = None) -> ShapeStructInfo:
+    def as_struct_info(self, dict_globals: dict[str, Any] | None = None) -> ShapeStructInfo:
         values = [_eval_shape(v, dict_globals) for v in self.values] if self.values else None
         return ShapeStructInfo(values, self.ndim)
 
 
-def Shape(values: Optional[List[PrimExpr]] = None, ndim: int = -1) -> ShapeProxy:
+def Shape(values: list[PrimExpr] | None = None, ndim: int = -1) -> ShapeProxy:
     return ShapeProxy(values, ndim)
 
 
@@ -446,8 +446,8 @@ def Shape(values: Optional[List[PrimExpr]] = None, ndim: int = -1) -> ShapeProxy
 
 
 class PrimProxy(StructInfoProxy):
-    dtype: Optional[str]
-    value: Optional[Union[int, float, str, PrimExpr]]
+    dtype: str | None
+    value: int | float | str | PrimExpr | None
 
     """The type of TIR-representable values.
 
@@ -462,8 +462,8 @@ class PrimProxy(StructInfoProxy):
 
     def __init__(
         self,
-        dtype: Optional[str] = None,
-        value: Optional[Union[int, float, str, PrimExpr]] = None,
+        dtype: str | None = None,
+        value: int | float | str | PrimExpr | None = None,
     ) -> None:
         if dtype is None and value is None:
             raise TypeError(
@@ -473,13 +473,13 @@ def __init__(
         self.dtype = dtype
         self.value = value
 
-    def get_symbolic_vars(self) -> Set[str]:
+    def get_symbolic_vars(self) -> set[str]:
         if isinstance(self.value, str) and self.value.isidentifier():
             return {self.value}
         else:
             return set()
 
-    def as_struct_info(self, dict_globals: Optional[Dict[str, Any]] = None) -> ShapeStructInfo:
+    def as_struct_info(self, dict_globals: dict[str, Any] | None = None) -> ShapeStructInfo:
         if self.value is None:
             return PrimStructInfo(dtype=self.dtype)
         else:
@@ -488,8 +488,8 @@ def as_struct_info(self, dict_globals: Optional[Dict[str, Any]] = None) -> Shape
 
 
 def Prim(
-    dtype: Optional[str] = None,
-    value: Optional[Union[int, float, str, PrimExpr]] = None,
+    dtype: str | None = None,
+    value: int | float | str | PrimExpr | None = None,
 ) -> PrimProxy:
     return PrimProxy(dtype, value)
 
@@ -525,9 +525,7 @@ def _normalize_struct_info_proxy(annotation) -> StructInfoProxy:
         raise TypeError(f"Expected StructInfoProxy but got {type(annotation)}.")
 
 
-def _normalize_struct_info(
-    struct_info, dict_globals: Optional[Dict[str, Any]] = None
-) -> StructInfo:
+def _normalize_struct_info(struct_info, dict_globals: dict[str, Any] | None = None) -> StructInfo:
     if isinstance(struct_info, StructInfo):
         return struct_info
     else:
diff --git a/python/tvm/script/parser/relax/parser.py b/python/tvm/script/parser/relax/parser.py
index fd3c2f508f2b..bb1a867331bf 100644
--- a/python/tvm/script/parser/relax/parser.py
+++ b/python/tvm/script/parser/relax/parser.py
@@ -18,7 +18,7 @@
 
 import functools
 import numbers
-from typing import Any, Dict, Optional
+from typing import Any
 
 from tvm import relax, tir
 from tvm.ir import GlobalVar, structural_equal
@@ -43,7 +43,7 @@ def bind_assign_value(
     node: doc.expr,
     var_name: str,
     value: Any,
-    anno_sinfo: Optional[StructInfo] = None,
+    anno_sinfo: StructInfo | None = None,
 ) -> Any:
     var_table = self.var_table.get()
 
@@ -125,11 +125,11 @@ def is_called(node: Any, func_name: str) -> bool:
         # Recursive call was found
         if isinstance(node.func, doc.Name) and node.func.id == func_name:
             return True
-    elif isinstance(node, (list, tuple)):
+    elif isinstance(node, list | tuple):
         for stmt in node:
             if is_called(stmt, func_name):
                 return True
-    elif isinstance(node, (doc.AnnAssign, doc.Assign, doc.Return, doc.Expr)):
+    elif isinstance(node, doc.AnnAssign | doc.Assign | doc.Return | doc.Expr):
         return is_called(node.value, func_name)
     elif isinstance(node, doc.With):
         return is_called(node.body, func_name)
@@ -152,8 +152,8 @@ def is_recursive(node: doc.FunctionDef) -> bool:
 
 
 def collect_symbolic_var_from_prelude(
-    self: Parser, node: doc.FunctionDef, symbolic_vars: Dict[str, tir.Var]
-) -> Dict[str, tir.Var]:
+    self: Parser, node: doc.FunctionDef, symbolic_vars: dict[str, tir.Var]
+) -> dict[str, tir.Var]:
     prelude_vars = {}
     for stmt in node.body:
         if isinstance(stmt, doc.Assign) and all(
@@ -415,7 +415,7 @@ def visit_if(self: Parser, node: doc.If) -> None:
 
 
 @dispatch.register(token="relax", type_name="enter_token")
-def enter_token(self: Parser) -> Dict[str, Any]:
+def enter_token(self: Parser) -> dict[str, Any]:
     def relax_call(self, *args) -> Expr:
         args = [convert_to_expr(arg) if isinstance(arg, tuple) else arg for arg in args]
 
@@ -430,6 +430,6 @@ def relax_call(self, *args) -> Expr:
 
 
 @dispatch.register(token="relax", type_name="exit_token")
-def exit_token(self: Parser, context: Dict[str, Any]) -> None:
+def exit_token(self: Parser, context: dict[str, Any]) -> None:
     assert "GlobalVar.__call__" in context
     GlobalVar.__call__ = context.get("GlobalVar.__call__")
diff --git a/python/tvm/script/parser/tir/entry.py b/python/tvm/script/parser/tir/entry.py
index 80c58186fd72..da09851e6757 100644
--- a/python/tvm/script/parser/tir/entry.py
+++ b/python/tvm/script/parser/tir/entry.py
@@ -17,7 +17,7 @@
 """The entry point of TVM parser for tir."""
 
 import inspect
-from typing import Callable, Optional, Union
+from collections.abc import Callable
 
 from tvm.ir.base import deprecated
 from tvm.tir import Buffer, PrimFunc
@@ -28,8 +28,8 @@
 
 
 def prim_func(
-    func: Optional[Callable] = None, private: bool = False, check_well_formed=True
-) -> Union[PrimFunc, Callable]:
+    func: Callable | None = None, private: bool = False, check_well_formed=True
+) -> PrimFunc | Callable:
     """The parsing method for tir prim func, by using `@prim_func` as decorator.
 
     Parameters
@@ -173,6 +173,14 @@ def wrapper(*args, **kwargs):
 class BufferProxy:
     """Buffer proxy class for constructing tir buffer."""
 
+    def __or__(self, other):
+        """Support ``T.Buffer | None`` union syntax in annotations."""
+        return self
+
+    def __ror__(self, other):
+        """Support ``None | T.Buffer`` union syntax in annotations."""
+        return self
+
     def __call__(
         self,
         shape,
@@ -211,6 +219,14 @@ def __getitem__(self, keys) -> Buffer:
 class PtrProxy:
     """Ptr proxy class for constructing tir pointer."""
 
+    def __or__(self, other):
+        """Support union syntax in annotations."""
+        return self
+
+    def __ror__(self, other):
+        """Support union syntax in annotations."""
+        return self
+
     @deprecated("T.Ptr(...)", "T.handle(...)")
     def __call__(self, dtype, storage_scope="global"):
         if callable(dtype):
diff --git a/python/tvm/script/parser/tir/operation.py b/python/tvm/script/parser/tir/operation.py
index b22b0a7335db..2bb1e44da72b 100644
--- a/python/tvm/script/parser/tir/operation.py
+++ b/python/tvm/script/parser/tir/operation.py
@@ -16,8 +16,6 @@
 # under the License.
 """The tir expression operation registration"""
 
-from typing import Type
-
 from tvm import tir
 from tvm.runtime import DataType, DataTypeCode
 from tvm.tir import IntImm
@@ -26,7 +24,7 @@
 from .._core import OpMethod, doc, register_op
 
 
-def _register_expr_op(ty: Type):  # pylint: disable=invalid-name
+def _register_expr_op(ty: type):  # pylint: disable=invalid-name
     ty._dispatch_type = ty  # pylint: disable=protected-access
 
     def _and(a, b):
@@ -118,7 +116,7 @@ def _gt(a, b):
     def _ge(a, b):
         return _auto_broadcast(a, b, tir.GE)
 
-    def r(op: Type, i: int, m: OpMethod):  # pylint: disable=invalid-name
+    def r(op: type, i: int, m: OpMethod):  # pylint: disable=invalid-name
         register_op(ty, op, i)(m)
 
     for i in [0, 1]:
diff --git a/python/tvm/script/parser/tir/parser.py b/python/tvm/script/parser/tir/parser.py
index 2d9811194961..5cad41e216b2 100644
--- a/python/tvm/script/parser/tir/parser.py
+++ b/python/tvm/script/parser/tir/parser.py
@@ -18,7 +18,7 @@
 
 import contextlib
 from functools import partial
-from typing import Any, Dict, Optional
+from typing import Any
 
 import tvm
 from tvm.ir import GlobalVar, PrimType
@@ -55,11 +55,11 @@ def bind_with_value(self: Parser, node: doc.expr, var_name: str, value: Any) ->
     res : Any
         The bound value.
     """
-    if isinstance(value, (list, tuple)):
+    if isinstance(value, list | tuple):
         for i, v in enumerate(value):
             bind_with_value(self, node, f"{var_name}_{i}", v)
         return value
-    elif isinstance(value, (Buffer, Var)):
+    elif isinstance(value, Buffer | Var):
         IRBuilder.name(var_name, value)
         return value
     else:
@@ -91,7 +91,7 @@ def bind_for_value(self: Parser, node: doc.expr, var_name: str, value: Any) -> A
     res : Any
         The bound value.
     """
-    if isinstance(value, (list, tuple, tvm.ir.Array)):
+    if isinstance(value, list | tuple | tvm.ir.Array):
         for i, v in enumerate(value):
             bind_for_value(self, node, f"{var_name}_{i}", v)
         return value
@@ -129,7 +129,7 @@ def bind_assign_value(self: Parser, node: doc.expr, var_name: str, value: Any) -
     """
     if isinstance(value, T.meta_var):
         return value.value
-    elif isinstance(value, (list, tuple)):
+    elif isinstance(value, list | tuple):
         for i, v in enumerate(value):
             bind_assign_value(self, node, f"{var_name}_{i}", v)
         return value
@@ -138,7 +138,7 @@ def bind_assign_value(self: Parser, node: doc.expr, var_name: str, value: Any) -
         res = value.__enter__()
         IRBuilder.name(var_name, res)
         return res
-    elif isinstance(value, (Buffer, IterVar)) or (
+    elif isinstance(value, Buffer | IterVar) or (
         isinstance(value, Var) and not self.var_table.exist(value)
     ):
         IRBuilder.name(var_name, value)
@@ -171,9 +171,9 @@ def find_decorator_annotation(node: doc.FunctionDef, annotation: str, default: b
 def range_sugar(
     start: PrimExpr,
     stop: PrimExpr = None,
-    step: Optional[PrimExpr] = None,
+    step: PrimExpr | None = None,
     *,
-    annotations: Optional[Dict[str, Any]] = None,
+    annotations: dict[str, Any] | None = None,
 ) -> T.frame.ForFrame:
     """The sugar for python range builtin."""
 
@@ -473,7 +473,7 @@ def visit_expr_stmt(self: Parser, node: doc.Expr) -> None:
         res.__enter__()
     elif isinstance(res, PrimExpr):
         T.evaluate(res)
-    elif isinstance(res, (int, bool)):
+    elif isinstance(res, int | bool):
         T.evaluate(tvm.tir.const(res))
     elif isinstance(res, tvm.relax.Call) and not res.args:
         # Using GlobalVar.__call__ with no arguments is ambiguous, as
@@ -503,7 +503,7 @@ def visit_if(self: Parser, node: doc.If) -> None:
     """
     with self.var_table.with_frame():
         predicate = self.eval_expr(node.test)
-        if isinstance(predicate, (PrimExpr, tvm.tir.expr.ExprOp)):
+        if isinstance(predicate, PrimExpr | tvm.tir.expr.ExprOp):
             with T.If(self.eval_expr(node.test)):
                 with T.Then():
                     with self.var_table.with_frame():
diff --git a/python/tvm/script/printer/doc.py b/python/tvm/script/printer/doc.py
index 73309f542511..a5f3a08c210b 100644
--- a/python/tvm/script/printer/doc.py
+++ b/python/tvm/script/printer/doc.py
@@ -16,8 +16,9 @@
 # under the License.
 """Doc types for TVMScript Unified Printer"""
 
+from collections.abc import Sequence
 from enum import IntEnum, unique
-from typing import Dict, List, Optional, Sequence, Tuple, Union
+from typing import Union
 
 from tvm_ffi import register_object
 from tvm_ffi.access_path import AccessPath
@@ -52,7 +53,7 @@ def attr(self, name: str) -> "AttrAccessDoc":
         """
         return _ffi_api.ExprDocAttr(self, name)  # type: ignore # pylint: disable=no-member
 
-    def call(self, *args: Tuple["ExprDoc"], **kwargs: Dict[str, "ExprDoc"]) -> "CallDoc":
+    def call(self, *args: tuple["ExprDoc"], **kwargs: dict[str, "ExprDoc"]) -> "CallDoc":
         """
         Create a doc that represents function call, with self as callee.
 
@@ -73,7 +74,7 @@ def call(self, *args: Tuple["ExprDoc"], **kwargs: Dict[str, "ExprDoc"]) -> "Call
 
     _IndexType = Union["ExprDoc", "SliceDoc"]
 
-    def __getitem__(self, indices: Union[Tuple[_IndexType], _IndexType]) -> "IndexDoc":
+    def __getitem__(self, indices: tuple[_IndexType] | _IndexType) -> "IndexDoc":
         """
         Create a doc that represents index access on self.
 
@@ -118,7 +119,7 @@ class StmtBlockDoc(Doc):
 
     stmts: Sequence[StmtDoc]
 
-    def __init__(self, stmts: List[StmtDoc]):
+    def __init__(self, stmts: list[StmtDoc]):
         self.__init_handle_by_constructor__(_ffi_api.StmtBlockDoc, stmts)  # type: ignore # pylint: disable=no-member
 
 
@@ -126,12 +127,12 @@ def __init__(self, stmts: List[StmtDoc]):
 class LiteralDoc(ExprDoc):
     """Doc that represents literal value"""
 
-    value: Union[str, IntImm, FloatImm, None]
+    value: str | IntImm | FloatImm | None
 
     def __init__(
         self,
-        value: Union[str, float, bool, int, None],
-        path: Optional[AccessPath] = None,
+        value: str | float | bool | int | None,
+        path: AccessPath | None = None,
     ):
         if value is None:
             self.__init_handle_by_constructor__(_ffi_api.LiteralDocNone, path)  # type: ignore # pylint: disable=no-member
@@ -191,7 +192,7 @@ class IndexDoc(ExprDoc):
     value: ExprDoc
     indices: Sequence[Union[ExprDoc, "SliceDoc"]]
 
-    def __init__(self, value: ExprDoc, indices: List[Union[ExprDoc, "SliceDoc"]]):
+    def __init__(self, value: ExprDoc, indices: list[Union[ExprDoc, "SliceDoc"]]):
         self.__init_handle_by_constructor__(_ffi_api.IndexDoc, value, indices)  # type: ignore # pylint: disable=no-member
 
 
@@ -204,7 +205,7 @@ class CallDoc(ExprDoc):
     kwargs_keys: Sequence[str]
     kwargs_values: Sequence[ExprDoc]
 
-    def __init__(self, callee: ExprDoc, *args: Tuple[ExprDoc], **kwargs: Dict[str, ExprDoc]):
+    def __init__(self, callee: ExprDoc, *args: tuple[ExprDoc], **kwargs: dict[str, ExprDoc]):
         kwargs_keys = list(kwargs.keys())
         kwargs_values = list(kwargs.values())
         self.__init_handle_by_constructor__(
@@ -275,7 +276,7 @@ class OperationDoc(ExprDoc):
     kind: OperationKind
     operands: Sequence[ExprDoc]
 
-    def __init__(self, kind: OperationKind, operands: List[ExprDoc]):
+    def __init__(self, kind: OperationKind, operands: list[ExprDoc]):
         self.__init_handle_by_constructor__(_ffi_api.OperationDoc, kind, operands)  # type: ignore # pylint: disable=no-member
 
 
@@ -286,7 +287,7 @@ class LambdaDoc(ExprDoc):
     args: Sequence[IdDoc]
     body: ExprDoc
 
-    def __init__(self, args: List[IdDoc], body: ExprDoc):
+    def __init__(self, args: list[IdDoc], body: ExprDoc):
         self.__init_handle_by_constructor__(_ffi_api.LambdaDoc, args, body)  # type: ignore # pylint: disable=no-member
 
 
@@ -296,7 +297,7 @@ class TupleDoc(ExprDoc):
 
     elements: Sequence[ExprDoc]
 
-    def __init__(self, elements: List[ExprDoc]):
+    def __init__(self, elements: list[ExprDoc]):
         self.__init_handle_by_constructor__(_ffi_api.TupleDoc, elements)  # type: ignore # pylint: disable=no-member
 
 
@@ -306,7 +307,7 @@ class ListDoc(ExprDoc):
 
     elements: Sequence[ExprDoc]
 
-    def __init__(self, elements: List[ExprDoc]):
+    def __init__(self, elements: list[ExprDoc]):
         self.__init_handle_by_constructor__(_ffi_api.ListDoc, elements)  # type: ignore # pylint: disable=no-member
 
 
@@ -317,7 +318,7 @@ class DictDoc(ExprDoc):
     keys: Sequence[ExprDoc]
     values: Sequence[ExprDoc]
 
-    def __init__(self, content: Dict[ExprDoc, ExprDoc]):
+    def __init__(self, content: dict[ExprDoc, ExprDoc]):
         keys = list(content.keys())
         values = list(content.values())
         self.__init_handle_by_constructor__(_ffi_api.DictDoc, keys, values)  # type: ignore # pylint: disable=no-member
@@ -331,15 +332,15 @@ class SliceDoc(ExprDoc):
     This doc can only appear in `IndexDoc.indices`.
     """
 
-    start: Optional[ExprDoc]
-    stop: Optional[ExprDoc]
-    step: Optional[ExprDoc]
+    start: ExprDoc | None
+    stop: ExprDoc | None
+    step: ExprDoc | None
 
     def __init__(
         self,
-        start: Optional[ExprDoc] = None,
-        stop: Optional[ExprDoc] = None,
-        step: Optional[ExprDoc] = None,
+        start: ExprDoc | None = None,
+        stop: ExprDoc | None = None,
+        step: ExprDoc | None = None,
     ):
         self.__init_handle_by_constructor__(_ffi_api.SliceDoc, start, stop, step)  # type: ignore # pylint: disable=no-member
 
@@ -349,10 +350,10 @@ class AssignDoc(StmtDoc):
     """Doc that represents assign statement."""
 
     lhs: ExprDoc
-    rhs: Optional[ExprDoc]
-    annotation: Optional[ExprDoc]
+    rhs: ExprDoc | None
+    annotation: ExprDoc | None
 
-    def __init__(self, lhs: ExprDoc, rhs: Optional[ExprDoc], annotation: Optional[ExprDoc] = None):
+    def __init__(self, lhs: ExprDoc, rhs: ExprDoc | None, annotation: ExprDoc | None = None):
         self.__init_handle_by_constructor__(
             _ffi_api.AssignDoc,  # type: ignore # pylint: disable=no-member
             lhs,
@@ -369,7 +370,7 @@ class IfDoc(StmtDoc):
     then_branch: Sequence[StmtDoc]
     else_branch: Sequence[StmtDoc]
 
-    def __init__(self, predicate: ExprDoc, then_branch: List[StmtDoc], else_branch: List[StmtDoc]):
+    def __init__(self, predicate: ExprDoc, then_branch: list[StmtDoc], else_branch: list[StmtDoc]):
         self.__init_handle_by_constructor__(
             _ffi_api.IfDoc,  # type: ignore # pylint: disable=no-member
             predicate,
@@ -385,7 +386,7 @@ class WhileDoc(StmtDoc):
     predicate: ExprDoc
     body: Sequence[StmtDoc]
 
-    def __init__(self, predicate: ExprDoc, body: List[StmtDoc]):
+    def __init__(self, predicate: ExprDoc, body: list[StmtDoc]):
         self.__init_handle_by_constructor__(_ffi_api.WhileDoc, predicate, body)  # type: ignore # pylint: disable=no-member
 
 
@@ -397,7 +398,7 @@ class ForDoc(StmtDoc):
     rhs: ExprDoc
     body: Sequence[StmtDoc]
 
-    def __init__(self, lhs: ExprDoc, rhs: ExprDoc, body: List[StmtDoc]):
+    def __init__(self, lhs: ExprDoc, rhs: ExprDoc, body: list[StmtDoc]):
         self.__init_handle_by_constructor__(_ffi_api.ForDoc, lhs, rhs, body)  # type: ignore # pylint: disable=no-member
 
 
@@ -412,11 +413,11 @@ class ScopeDoc(StmtDoc):
         <body...>
     """
 
-    lhs: Optional[ExprDoc]
+    lhs: ExprDoc | None
     rhs: ExprDoc
     body: Sequence[StmtDoc]
 
-    def __init__(self, lhs: Optional[ExprDoc], rhs: ExprDoc, body: List[StmtDoc]):
+    def __init__(self, lhs: ExprDoc | None, rhs: ExprDoc, body: list[StmtDoc]):
         self.__init_handle_by_constructor__(_ffi_api.ScopeDoc, lhs, rhs, body)  # type: ignore # pylint: disable=no-member
 
 
@@ -435,9 +436,9 @@ class AssertDoc(StmtDoc):
     """Doc that represents assert statement."""
 
     test: ExprDoc
-    msg: Optional[ExprDoc]
+    msg: ExprDoc | None
 
-    def __init__(self, test: ExprDoc, msg: Optional[ExprDoc] = None):
+    def __init__(self, test: ExprDoc, msg: ExprDoc | None = None):
         self.__init_handle_by_constructor__(_ffi_api.AssertDoc, test, msg)  # type: ignore # pylint: disable=no-member
 
 
@@ -458,16 +459,16 @@ class FunctionDoc(StmtDoc):
     name: IdDoc
     args: Sequence[AssignDoc]
     decorators: Sequence[ExprDoc]
-    return_type: Optional[ExprDoc]
+    return_type: ExprDoc | None
     body: Sequence[StmtDoc]
 
     def __init__(
         self,
         name: IdDoc,
-        args: List[AssignDoc],
-        decorators: List[ExprDoc],
-        return_type: Optional[ExprDoc],
-        body: List[StmtDoc],
+        args: list[AssignDoc],
+        decorators: list[ExprDoc],
+        return_type: ExprDoc | None,
+        body: list[StmtDoc],
     ):
         self.__init_handle_by_constructor__(
             _ffi_api.FunctionDoc,  # type: ignore # pylint: disable=no-member
@@ -487,7 +488,7 @@ class ClassDoc(StmtDoc):
     decorators: Sequence[ExprDoc]
     body: Sequence[StmtDoc]
 
-    def __init__(self, name: IdDoc, decorators: List[ExprDoc], body: List[StmtDoc]):
+    def __init__(self, name: IdDoc, decorators: list[ExprDoc], body: list[StmtDoc]):
         self.__init_handle_by_constructor__(
             _ffi_api.ClassDoc,  # type: ignore # pylint: disable=no-member
             name,
diff --git a/python/tvm/script/printer/doc_printer.py b/python/tvm/script/printer/doc_printer.py
index dc92d4261f83..42216a2c1843 100644
--- a/python/tvm/script/printer/doc_printer.py
+++ b/python/tvm/script/printer/doc_printer.py
@@ -16,8 +16,6 @@
 # under the License.
 """Functions to print doc into text format"""
 
-from typing import List, Optional
-
 from tvm_ffi.access_path import AccessPath
 
 from tvm.runtime.script_printer import PrinterConfig
@@ -30,8 +28,8 @@ def to_python_script(
     doc: Doc,
     indent_spaces: int = 4,
     print_line_numbers: bool = False,
-    num_context_lines: Optional[int] = None,
-    path_to_underline: Optional[List[AccessPath]] = None,
+    num_context_lines: int | None = None,
+    path_to_underline: list[AccessPath] | None = None,
 ) -> str:
     """Convert Doc into Python script.
 
diff --git a/python/tvm/target/codegen.py b/python/tvm/target/codegen.py
index 32952af4b66e..ace0e8f7184f 100644
--- a/python/tvm/target/codegen.py
+++ b/python/tvm/target/codegen.py
@@ -57,7 +57,7 @@ def target_has_features(cpu_features, target=None):
         True if target has the feature(s).
     """
     assert isinstance(target, Target) or target is None
-    assert isinstance(cpu_features, (Array, list, tuple, str))
+    assert isinstance(cpu_features, Array | list | tuple | str)
     has_feats = True
     cpu_features = [cpu_features] if isinstance(cpu_features, str) else cpu_features
     for feat in cpu_features:
@@ -204,7 +204,7 @@ def llvm_cpu_has_features(cpu_features, target=None):
         True if target CPU has the feature(s).
     """
     assert isinstance(target, Target) or target is None
-    assert isinstance(cpu_features, (Array, list, tuple, str))
+    assert isinstance(cpu_features, Array | list | tuple | str)
     has_feats = True
     cpu_features = [cpu_features] if isinstance(cpu_features, str) else cpu_features
     for feat in cpu_features:
diff --git a/python/tvm/target/detect_target.py b/python/tvm/target/detect_target.py
index 37bd35c3fb94..81accfed1287 100644
--- a/python/tvm/target/detect_target.py
+++ b/python/tvm/target/detect_target.py
@@ -16,8 +16,6 @@
 # under the License.
 """Detect target."""
 
-from typing import Union
-
 from tvm_ffi import get_global_func
 
 from ..runtime import Device, device
@@ -108,7 +106,7 @@ def _detect_cpu(dev: Device) -> Target:  # pylint: disable=unused-argument
     )
 
 
-def detect_target_from_device(dev: Union[str, Device]) -> Target:
+def detect_target_from_device(dev: str | Device) -> Target:
     """Detects Target associated with the given device. If the device does not exist,
     there will be an Error.
 
diff --git a/python/tvm/target/tag_registry/registry.py b/python/tvm/target/tag_registry/registry.py
index 32582b35d011..957a624b15cf 100644
--- a/python/tvm/target/tag_registry/registry.py
+++ b/python/tvm/target/tag_registry/registry.py
@@ -16,13 +16,13 @@
 # under the License.
 """Target tag registry functions."""
 
-from typing import Any, Dict, Optional
+from typing import Any
 
 from .. import _ffi_api
 from ..target import Target
 
 
-def list_tags() -> Optional[Dict[str, Target]]:
+def list_tags() -> dict[str, Target] | None:
     """Returns a dict of tags, which maps each tag name to its corresponding target.
 
     Returns
@@ -36,7 +36,7 @@ def list_tags() -> Optional[Dict[str, Target]]:
     return None
 
 
-def register_tag(name: str, config: Dict[str, Any], override: bool = False) -> Optional[Target]:
+def register_tag(name: str, config: dict[str, Any], override: bool = False) -> Target | None:
     """Add a user-defined tag into the target tag registry.
 
     Parameters
diff --git a/python/tvm/target/target.py b/python/tvm/target/target.py
index 92c4015bbacb..b008df498664 100644
--- a/python/tvm/target/target.py
+++ b/python/tvm/target/target.py
@@ -16,8 +16,6 @@
 # under the License.
 """Target data structure."""
 
-from typing import Union
-
 import tvm_ffi
 
 from tvm.ir.container import Map
@@ -133,14 +131,14 @@ def __init__(self, target, host=None):
             When using a dictionary or json string to configure target, the possible values are
             same as target.
         """
-        if isinstance(target, (dict, str)):
+        if isinstance(target, dict | str):
             target = convert(target)
-        if isinstance(host, (dict, str)):
+        if isinstance(host, dict | str):
             host = convert(host)
-        if target is None or not isinstance(target, (Map, String, Target, str)):
+        if target is None or not isinstance(target, Map | String | Target | str):
             raise ValueError(f"target has to be a string or dictionary. instead get {type(target)}")
         if host is not None:
-            if not isinstance(host, (Map, String, Target, str)):
+            if not isinstance(host, Map | String | Target | str):
                 raise ValueError("target host has to be a string or dictionary.")
             self.__init_handle_by_constructor__(_ffi_api.Target, Target(target), Target(host))
         else:
@@ -160,7 +158,7 @@ def with_host(self, host=None):
         return _ffi_api.WithHost(self, Target(host))
 
     @staticmethod
-    def from_device(device: Union[str, Device]) -> "Target":
+    def from_device(device: str | Device) -> "Target":
         """Detects Target associated with the given device. If the device does not exist,
         there will be an Error.
 
diff --git a/python/tvm/te/operation.py b/python/tvm/te/operation.py
index e1823c3176e3..e01ace8da572 100644
--- a/python/tvm/te/operation.py
+++ b/python/tvm/te/operation.py
@@ -20,7 +20,6 @@
 
 # pylint: disable=invalid-name
 from numbers import Integral as _Integral
-from typing import List, Optional, Union
 
 import tvm.arith._ffi_api
 import tvm.tir
@@ -129,7 +128,7 @@ def compute(shape, fcompute, name="compute", tag="", attrs=None, varargs_names=N
     dim_var = [tvm.tir.IterVar((0, s), x, 0) for x, s in zip(arg_names, shape[:out_ndim])]
     body = fcompute(*[v.var for v in dim_var])
 
-    if not isinstance(body, (list, tuple)):
+    if not isinstance(body, list | tuple):
         body = [body]
     body = convert(body)
     op_node = _ffi_api.ComputeOp(name, tag, attrs, dim_var, body)
@@ -284,8 +283,8 @@ def extern(
         if tag != "":
             raise ValueError("nested tag is not allowed for now")
         tag = _tag.TagScope.get_current().tag
-    shape = (shape,) if isinstance(shape, (tvm.tir.PrimExpr, _Integral)) else shape
-    if shape == () or isinstance(shape[0], (tvm.tir.PrimExpr, _Integral)):
+    shape = (shape,) if isinstance(shape, tvm.tir.PrimExpr | _Integral) else shape
+    if shape == () or isinstance(shape[0], tvm.tir.PrimExpr | _Integral):
         shape = [shape]
     if in_buffers is not None:
         in_buffers = [in_buffers] if not isinstance(in_buffers, list) else in_buffers
@@ -340,7 +339,7 @@ def extern(
     return res[0] if len(res) == 1 else res
 
 
-def extern_primfunc(input_tensors: List[_tensor.Tensor], primfunc: tvm.tir.PrimFunc, **kwargs):
+def extern_primfunc(input_tensors: list[_tensor.Tensor], primfunc: tvm.tir.PrimFunc, **kwargs):
     """Compute tensors via a schedulable TIR PrimFunc
 
     Parameters
@@ -547,7 +546,7 @@ def reduce_axis(dom, name="rv", thread_tag="", span=None):
 
 
 def create_prim_func(
-    ops: List[Union[_tensor.Tensor, tvm.tir.Var]], index_dtype_override: Optional[str] = None
+    ops: list[_tensor.Tensor | tvm.tir.Var], index_dtype_override: str | None = None
 ) -> tvm.tir.PrimFunc:
     """Create a TensorIR PrimFunc from tensor expression
 
@@ -598,7 +597,7 @@ def tir_matmul(a: T.handle, b: T.handle, c: T.handle) -> None:
     func : tir.PrimFunc
         The created function.
     """
-    if not isinstance(ops, (list, tuple, Array)):
+    if not isinstance(ops, list | tuple | Array):
         ops = [ops]
     return _ffi_api.CreatePrimFunc(ops, index_dtype_override)
 
diff --git a/python/tvm/testing/plugin.py b/python/tvm/testing/plugin.py
index 72230f410c64..d7e096472f4d 100644
--- a/python/tvm/testing/plugin.py
+++ b/python/tvm/testing/plugin.py
@@ -291,7 +291,7 @@ def sort_key(item):
 
 
 def _target_to_requirement(target):
-    if isinstance(target, (str, dict)):
+    if isinstance(target, str | dict):
         target = tvm.target.Target(target)
 
     # mapping from target to decorator
diff --git a/python/tvm/testing/runner.py b/python/tvm/testing/runner.py
index 732d96d5502f..366a4b5155cf 100644
--- a/python/tvm/testing/runner.py
+++ b/python/tvm/testing/runner.py
@@ -17,9 +17,8 @@
 # pylint: disable=invalid-name, missing-function-docstring
 """A utility method to run a TVM module on a remote device."""
 
-from typing import TYPE_CHECKING, Callable, List, Optional, Tuple, Union
-
-from typing_extensions import Literal
+from collections.abc import Callable
+from typing import TYPE_CHECKING, Literal, Optional, Union
 
 if TYPE_CHECKING:
     import numpy as np
@@ -37,9 +36,9 @@ def _args_to_device(args, device):
 
     uploaded_args = []
     for arg in args:
-        if isinstance(arg, (np.ndarray, Tensor)):
+        if isinstance(arg, np.ndarray | Tensor):
             uploaded_args.append(empty(arg.shape, dtype=arg.dtype, device=device).copyfrom(arg))
-        elif isinstance(arg, (int, float)):
+        elif isinstance(arg, int | float):
             uploaded_args.append(arg)
         else:
             raise ValueError(f"Unsupported input type: {type(arg)}")
@@ -58,7 +57,7 @@ def _args_to_numpy(args):
     return downloaded_args
 
 
-def _normalize_export_func(export_func, output_format) -> Tuple[Callable, str]:
+def _normalize_export_func(export_func, output_format) -> tuple[Callable, str]:
     from tvm.contrib import ndk, tar
 
     def export_with(func):
@@ -81,10 +80,10 @@ def export_with(func):
 def local_run(  # pylint: disable=too-many-arguments,too-many-locals
     mod: "Module",
     device_type: str,
-    args: List[Union["np.ndarray", "Tensor", int, float]],
+    args: list[Union["np.ndarray", "Tensor", int, float]],
     evaluator_config: Optional["EvaluatorConfig"] = None,
-    export_func: Union[Callable[["Module", str], None], Literal["tar", "ndk"]] = "tar",
-    output_format: Optional[str] = None,
+    export_func: Callable[["Module", str], None] | Literal["tar", "ndk"] = "tar",
+    output_format: str | None = None,
 ):
     """Run a TVM module on a local device.
 
@@ -153,11 +152,11 @@ def local_run(  # pylint: disable=too-many-arguments,too-many-locals
 def rpc_run(  # pylint: disable=too-many-arguments,too-many-locals
     mod: "Module",
     device_type: str,
-    args: List[Union["np.ndarray", "Tensor", int, float]],
+    args: list[Union["np.ndarray", "Tensor", int, float]],
     evaluator_config: Optional["EvaluatorConfig"] = None,
     rpc_config: Optional["RPCConfig"] = None,
-    export_func: Union[Callable[["Module", str], None], Literal["tar", "ndk"]] = "tar",
-    output_format: Optional[str] = None,
+    export_func: Callable[["Module", str], None] | Literal["tar", "ndk"] = "tar",
+    output_format: str | None = None,
 ):
     """Run a TVM module on a remote device.
 
diff --git a/python/tvm/testing/utils.py b/python/tvm/testing/utils.py
index 8b89b2a2ca6e..a6744a3b9e01 100644
--- a/python/tvm/testing/utils.py
+++ b/python/tvm/testing/utils.py
@@ -78,8 +78,8 @@ def test_something():
 import shutil
 import sys
 import time
+from collections.abc import Callable
 from pathlib import Path
-from typing import Callable, List, Optional, Tuple, Union
 
 import numpy as np
 import pytest
@@ -602,13 +602,13 @@ class Feature:
     def __init__(
         self,
         name: str,
-        long_name: Optional[str] = None,
-        cmake_flag: Optional[str] = None,
-        target_kind_enabled: Optional[str] = None,
-        compile_time_check: Optional[Callable[[], Union[bool, str]]] = None,
-        target_kind_hardware: Optional[str] = None,
-        run_time_check: Optional[Callable[[], Union[bool, str]]] = None,
-        parent_features: Optional[Union[str, List[str]]] = None,
+        long_name: str | None = None,
+        cmake_flag: str | None = None,
+        target_kind_enabled: str | None = None,
+        compile_time_check: Callable[[], bool | str] | None = None,
+        target_kind_hardware: str | None = None,
+        run_time_check: Callable[[], bool | str] | None = None,
+        parent_features: str | list[str] | None = None,
     ):
         self.name = name
         self.long_name = long_name or name
@@ -1667,7 +1667,7 @@ def wraps(func):
     return wraps(func)
 
 
-def get_dtype_range(dtype: str) -> Tuple[int, int]:
+def get_dtype_range(dtype: str) -> tuple[int, int]:
     """
     Produces the min,max for a give data type.
 
diff --git a/python/tvm/tir/analysis/analysis.py b/python/tvm/tir/analysis/analysis.py
index 96c8ea390def..ea78cd87c1b1 100644
--- a/python/tvm/tir/analysis/analysis.py
+++ b/python/tvm/tir/analysis/analysis.py
@@ -17,7 +17,6 @@
 """Wrapping existing analysis utils."""
 
 # pylint: disable=invalid-name
-from typing import List, Optional, Union
 
 from tvm.ir import IRModule
 from tvm.tir.expr import Var
@@ -97,7 +96,7 @@ def verify_memory(func: PrimFunc) -> bool:
     return _ffi_api.verify_memory(func)  # type: ignore
 
 
-def undefined_vars(node: Union[Stmt, PrimExpr], defs: Optional[List[Var]] = None) -> List[Var]:
+def undefined_vars(node: Stmt | PrimExpr, defs: list[Var] | None = None) -> list[Var]:
     """Find undefined vars in a TIR statement or expression.
 
     Parameters
@@ -117,7 +116,7 @@ def undefined_vars(node: Union[Stmt, PrimExpr], defs: Optional[List[Var]] = None
     return _ffi_api.UndefinedVars(node, defs)  # type: ignore # pylint: disable=no-member
 
 
-def verify_well_formed(obj: Union[PrimFunc, IRModule], assert_mode: bool = True) -> bool:
+def verify_well_formed(obj: PrimFunc | IRModule, assert_mode: bool = True) -> bool:
     """Verify if the given TIR is well-formed. The verification includes:
         - Check if expressions not contain vars that is defined outside the block.
 
diff --git a/python/tvm/tir/buffer.py b/python/tvm/tir/buffer.py
index 1b9a5d464a9a..bb00cfe90bb4 100644
--- a/python/tvm/tir/buffer.py
+++ b/python/tvm/tir/buffer.py
@@ -129,7 +129,7 @@ def vload(self, begin, dtype=None, predicate=None):
         load : Expr
             The corresponding load expression.
         """
-        begin = (begin,) if isinstance(begin, (int, PrimExpr)) else begin
+        begin = (begin,) if isinstance(begin, int | PrimExpr) else begin
         dtype = dtype if dtype else self.dtype
         return _ffi_api.BufferVLoad(self, begin, dtype, predicate)  # type: ignore
 
@@ -154,7 +154,7 @@ def vstore(self, begin, value, predicate=None):
         store : Stmt
             The corresponding store stmt.
         """
-        begin = (begin,) if isinstance(begin, (int, PrimExpr)) else begin
+        begin = (begin,) if isinstance(begin, int | PrimExpr) else begin
         return _ffi_api.BufferVStore(self, begin, value, predicate)  # type: ignore
 
     def scope(self):
@@ -198,7 +198,7 @@ def __getitem__(self, indices):
         from .expr import BufferLoad, Ramp, const  # pylint: disable=import-outside-toplevel
         from .stmt import BufferRegion  # pylint: disable=import-outside-toplevel
 
-        if not isinstance(indices, (tuple, list)):
+        if not isinstance(indices, tuple | list):
             indices = [indices]
         has_slice = any(isinstance(i, slice) for i in indices)
         has_step = any(isinstance(i, slice) and i.step is not None for i in indices)
@@ -327,7 +327,7 @@ def decl_buffer(
     # pylint: disable=import-outside-toplevel
     from .expr import Var
 
-    shape = (shape,) if isinstance(shape, (PrimExpr, Integral)) else shape
+    shape = (shape,) if isinstance(shape, PrimExpr | Integral) else shape
     dtype = "float32" if dtype is None else dtype
     strides = () if strides is None else strides
 
diff --git a/python/tvm/tir/build.py b/python/tvm/tir/build.py
index 060dd1e34920..5f6bf8adbfb5 100644
--- a/python/tvm/tir/build.py
+++ b/python/tvm/tir/build.py
@@ -18,8 +18,6 @@
 # pylint: disable=invalid-name
 """The build utils in python."""
 
-from typing import Dict, Optional, Tuple, Union
-
 import tvm
 from tvm import ir
 from tvm.ir.module import IRModule
@@ -27,7 +25,7 @@
 from tvm.tir import PrimFunc
 
 
-def split_host_device_mods(mod: IRModule) -> Tuple[IRModule, Dict[Target, IRModule]]:
+def split_host_device_mods(mod: IRModule) -> tuple[IRModule, dict[Target, IRModule]]:
     """Split an IRModule into host and device modules.
 
     This function takes an IRModule containing functions with different target attributes
@@ -109,7 +107,7 @@ def is_host_func(f):
     # TODO(syfeng): Here we use str as key since target hash is not correct
     target_str2target = {}
     device_func_dict = {}
-    device_mod_dict: Dict[Target, IRModule] = {}
+    device_mod_dict: dict[Target, IRModule] = {}
     for gv, func in device_mod.functions.items():
         target = func.attrs.get("target", None)
         target_str = str(target) if target is not None else ""
@@ -133,7 +131,7 @@ def codegen_build(mod: IRModule, target: Target) -> tvm.runtime.Module:
 
 
 def tir_to_runtime(
-    host_mod: IRModule, device_mod_dict: Dict[Target, IRModule], target_host: Target
+    host_mod: IRModule, device_mod_dict: dict[Target, IRModule], target_host: Target
 ):
     """Convert a collection of TIR IRModules (keyed by Target) into a single runtime Module."""
 
@@ -155,9 +153,9 @@ def tir_to_runtime(
 
 
 def build(
-    mod: Union[PrimFunc, IRModule],
-    target: Optional[Union[str, Target]] = None,
-    pipeline: Union[None, str, tvm.transform.Pass] = "default",
+    mod: PrimFunc | IRModule,
+    target: str | Target | None = None,
+    pipeline: None | str | tvm.transform.Pass = "default",
 ):
     """Build a function with a signature, generating code for devices
     coupled with target information.
diff --git a/python/tvm/tir/expr.py b/python/tvm/tir/expr.py
index 58ece1e7f995..0aa686e3a846 100644
--- a/python/tvm/tir/expr.py
+++ b/python/tvm/tir/expr.py
@@ -28,8 +28,6 @@
   assert(y.a == x)
 """
 
-from typing import List, Optional, Union
-
 import tvm_ffi
 
 import tvm.ir._ffi_api
@@ -188,7 +186,7 @@ def __nonzero__(self):
     def __bool__(self) -> bool:
         return self.__nonzero__()
 
-    def equal(self, other: PrimExpr, span: Optional[Span] = None) -> bool:
+    def equal(self, other: PrimExpr, span: Span | None = None) -> bool:
         """Build an equal check expression with other expr.
 
         Parameters
@@ -206,7 +204,7 @@ def equal(self, other: PrimExpr, span: Optional[Span] = None) -> bool:
         """
         return _ffi_api._OpEQ(self, other, span)  # type: ignore
 
-    def astype(self, dtype: str, span: Optional[Span] = None) -> PrimExpr:
+    def astype(self, dtype: str, span: Span | None = None) -> PrimExpr:
         """Cast the expression to other type.
 
         Parameters
@@ -246,7 +244,7 @@ class EqualOp(ObjectConvertible, ExprOp):
     # This class is not manipulated by C++. So use python's identity check function is sufficient
     same_as = object.__eq__
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None):
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None):
         self.a = a
         self.b = b
         self.span = span
@@ -283,7 +281,7 @@ class NotEqualOp(ObjectConvertible, ExprOp):
     # This class is not manipulated by C++. So use python's identity check function is sufficient
     same_as = object.__eq__
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.a = a
         self.b = b
         self.span = span
@@ -312,7 +310,7 @@ class IntImmEnum(ObjectConvertible):
         The location of the cast in the source.
     """
 
-    def __init__(self, value: int, span: Optional[Span] = None) -> None:
+    def __init__(self, value: int, span: Span | None = None) -> None:
         self.value = value
         self.span = span
 
@@ -366,7 +364,7 @@ class Var(PrimExprWithOp):
     name_hint: str
     type_annotation: ir.Type
 
-    def __init__(self, name: str, dtype: Union[str, ir.Type], span: Optional[Span] = None) -> None:
+    def __init__(self, name: str, dtype: str | ir.Type, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.Var, name, dtype, span)  # type: ignore
 
 
@@ -388,7 +386,7 @@ class SizeVar(Var):
     """
 
     # pylint: disable=super-init-not-called
-    def __init__(self, name: str, dtype: Union[str, ir.Type], span: Optional[Span] = None) -> None:
+    def __init__(self, name: str, dtype: str | ir.Type, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.SizeVar, name, dtype, span)  # type: ignore
 
 
@@ -439,13 +437,13 @@ class IterVar(ExprOp, Object, Scriptable):
     def __init__(
         self,
         dom: ir.Range,
-        var: Union[Var, str],
+        var: Var | str,
         iter_type: int,
         thread_tag: str = "",
-        span: Optional[Span] = None,
+        span: Span | None = None,
     ) -> None:
         if dom is not None:
-            if isinstance(dom, (list, tuple)):
+            if isinstance(dom, list | tuple):
                 if len(dom) != 2:
                     raise TypeError("need to be list of ranges")
                 dom = tvm.ir.Range(dom[0], dom[1])
@@ -492,18 +490,18 @@ class CommReducer(Object, Scriptable):
         The location of this expression in the source code.
     """
 
-    lhs: List[Var]
-    rhs: List[Var]
-    result: List[PrimExpr]
-    identity_element: List[PrimExpr]
+    lhs: list[Var]
+    rhs: list[Var]
+    result: list[PrimExpr]
+    identity_element: list[PrimExpr]
 
     def __init__(
         self,
-        lhs: List[Var],
-        rhs: List[Var],
-        result: List[PrimExpr],
-        identity_element: List[PrimExpr],
-        span: Optional[Span] = None,
+        lhs: list[Var],
+        rhs: list[Var],
+        result: list[PrimExpr],
+        identity_element: list[PrimExpr],
+        span: Span | None = None,
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.CommReducer,
@@ -544,21 +542,21 @@ class Reduce(PrimExprWithOp):
     """
 
     combiner: CommReducer
-    source: List[PrimExpr]
-    init: List[PrimExpr]
-    axis: List[IterVar]
+    source: list[PrimExpr]
+    init: list[PrimExpr]
+    axis: list[IterVar]
     condition: PrimExpr
     value_index: int
 
     def __init__(
         self,
         combiner: CommReducer,
-        src: List[PrimExpr],
-        rdom: List[IterVar],
+        src: list[PrimExpr],
+        rdom: list[IterVar],
         condition: PrimExpr,
         value_index: int,
-        init: Optional[List[PrimExpr]] = None,
-        span: Optional[Span] = None,
+        init: list[PrimExpr] | None = None,
+        span: Span | None = None,
     ) -> None:
         init = [] if init is None else init
         self.__init_handle_by_constructor__(
@@ -591,7 +589,7 @@ class FloatImm(ConstExpr):
 
     value: float
 
-    def __init__(self, dtype: str, value: float, span: Optional[Span] = None) -> None:
+    def __init__(self, dtype: str, value: float, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(
             tvm.ir._ffi_api.FloatImm,
             dtype,
@@ -621,7 +619,7 @@ class IntImm(ConstExpr):
 
     value: int
 
-    def __init__(self, dtype: str, value: int, span: Optional[Span] = None) -> None:
+    def __init__(self, dtype: str, value: int, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(
             tvm.ir._ffi_api.IntImm,
             dtype,
@@ -663,7 +661,7 @@ class StringImm(ConstExpr):
 
     value: str
 
-    def __init__(self, value: str, span: Optional[Span] = None) -> None:
+    def __init__(self, value: str, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.StringImm, value, span)  # type: ignore
 
     def __eq__(self, other: PrimExpr) -> bool:
@@ -698,7 +696,7 @@ class Cast(PrimExprWithOp):
 
     value: PrimExpr
 
-    def __init__(self, dtype, value, span: Optional[Span] = None) -> None:
+    def __init__(self, dtype, value, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.Cast, dtype, value, span)  # type: ignore
 
 
@@ -718,7 +716,7 @@ class Add(BinaryOpExpr):
         The location of this expression in the source code.
     """
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.Add, a, b, span)  # type: ignore
 
 
@@ -738,7 +736,7 @@ class Sub(BinaryOpExpr):
         The location of this expression in the source code.
     """
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.Sub, a, b, span)  # type: ignore
 
 
@@ -758,7 +756,7 @@ class Mul(BinaryOpExpr):
         The location of this expression in the source code.
     """
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.Mul, a, b, span)  # type: ignore
 
 
@@ -778,7 +776,7 @@ class Div(BinaryOpExpr):
         The location of this expression in the source code.
     """
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.Div, a, b, span)  # type: ignore
 
 
@@ -798,7 +796,7 @@ class Mod(BinaryOpExpr):
         The location of this expression in the source code.
     """
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.Mod, a, b, span)  # type: ignore
 
 
@@ -818,7 +816,7 @@ class FloorDiv(BinaryOpExpr):
         The location of this expression in the source code.
     """
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.FloorDiv, a, b, span)  # type: ignore
 
 
@@ -838,7 +836,7 @@ class FloorMod(BinaryOpExpr):
         The location of this expression in the source code.
     """
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.FloorMod, a, b, span)  # type: ignore
 
 
@@ -858,7 +856,7 @@ class Min(BinaryOpExpr):
         The location of this expression in the source code.
     """
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.Min, a, b, span)  # type: ignore
 
 
@@ -878,7 +876,7 @@ class Max(BinaryOpExpr):
         The location of this expression in the source code.
     """
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.Max, a, b, span)  # type: ignore
 
 
@@ -898,7 +896,7 @@ class EQ(CmpExpr):
         The location of this expression in the source code.
     """
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.EQ, a, b, span)  # type: ignore
 
 
@@ -918,7 +916,7 @@ class NE(CmpExpr):
         The location of this expression in the source code.
     """
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.NE, a, b, span)  # type: ignore
 
 
@@ -938,7 +936,7 @@ class LT(CmpExpr):
         The location of this expression in the source code.
     """
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.LT, a, b, span)  # type: ignore
 
 
@@ -958,7 +956,7 @@ class LE(CmpExpr):
         The location of this expression in the source code.
     """
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.LE, a, b, span)  # type: ignore
 
 
@@ -978,7 +976,7 @@ class GT(CmpExpr):
         The location of this expression in the source code.
     """
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.GT, a, b, span)  # type: ignore
 
 
@@ -998,7 +996,7 @@ class GE(CmpExpr):
         The location of this expression in the source code.
     """
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.GE, a, b, span)  # type: ignore
 
 
@@ -1018,7 +1016,7 @@ class And(LogicalExpr):
         The location of this expression in the source code.
     """
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.And, a, b, span)  # type: ignore
 
 
@@ -1041,7 +1039,7 @@ class Or(LogicalExpr):
     a: PrimExpr
     b: PrimExpr
 
-    def __init__(self, a: PrimExpr, b: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, b: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.Or, a, b, span)  # type: ignore
 
 
@@ -1060,7 +1058,7 @@ class Not(LogicalExpr):
 
     a: PrimExpr
 
-    def __init__(self, a: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, a: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.Not, a, span)  # type: ignore
 
 
@@ -1099,7 +1097,7 @@ def __init__(
         condition: PrimExpr,
         true_value: PrimExpr,
         false_value: PrimExpr,
-        span: Optional[Span] = None,
+        span: Span | None = None,
     ) -> None:
         if isinstance(condition, bool):
             condition = IntImm("bool", condition)
@@ -1133,14 +1131,14 @@ class BufferLoad(PrimExprWithOp):
     """
 
     buffer: Buffer
-    indices: List[PrimExpr]
+    indices: list[PrimExpr]
 
     def __init__(
         self,
         buffer: Buffer,
-        indices: List[PrimExpr],
-        predicate: Optional[PrimExpr] = None,
-        span: Optional[Span] = None,
+        indices: list[PrimExpr],
+        predicate: PrimExpr | None = None,
+        span: Span | None = None,
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.BufferLoad,
@@ -1168,10 +1166,10 @@ class ProducerLoad(PrimExprWithOp):
     """
 
     producer: DataProducer
-    indices: List[PrimExpr]
+    indices: list[PrimExpr]
 
     def __init__(
-        self, producer: DataProducer, indices: List[PrimExpr], span: Optional[Span] = None
+        self, producer: DataProducer, indices: list[PrimExpr], span: Span | None = None
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.ProducerLoad,
@@ -1205,7 +1203,7 @@ class Ramp(PrimExprWithOp):
     lanes: PrimExpr
 
     def __init__(
-        self, base: PrimExpr, stride: PrimExpr, lanes: PrimExpr, span: Optional[Span] = None
+        self, base: PrimExpr, stride: PrimExpr, lanes: PrimExpr, span: Span | None = None
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.Ramp,
@@ -1235,7 +1233,7 @@ class Broadcast(PrimExprWithOp):
     value: PrimExpr
     lanes: PrimExpr
 
-    def __init__(self, value: PrimExpr, lanes: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, value: PrimExpr, lanes: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.Broadcast, value, lanes, span)  # type: ignore
 
 
@@ -1255,11 +1253,11 @@ class Shuffle(PrimExprWithOp):
         The location of this expression in the source code.
     """
 
-    vectors: List[PrimExpr]
-    indices: List[PrimExpr]
+    vectors: list[PrimExpr]
+    indices: list[PrimExpr]
 
     def __init__(
-        self, vectors: List[PrimExpr], indices: List[PrimExpr], span: Optional[Span] = None
+        self, vectors: list[PrimExpr], indices: list[PrimExpr], span: Span | None = None
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.Shuffle,
@@ -1301,10 +1299,10 @@ class Call(PrimExprWithOp):
     """
 
     op: Op
-    args: List[PrimExpr]
+    args: list[PrimExpr]
 
     def __init__(
-        self, dtype: str, op: Union[Op, str], args: List[PrimExpr], span: Optional[Span] = None
+        self, dtype: str, op: Op | str, args: list[PrimExpr], span: Span | None = None
     ) -> None:
         if isinstance(op, str):
             if not op.startswith("tir."):
@@ -1343,7 +1341,5 @@ class Let(PrimExprWithOp):
     value: PrimExpr
     body: PrimExpr
 
-    def __init__(
-        self, var: Var, value: PrimExpr, body: PrimExpr, span: Optional[Span] = None
-    ) -> None:
+    def __init__(self, var: Var, value: PrimExpr, body: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.Let, var, value, body, span)  # type: ignore
diff --git a/python/tvm/tir/function.py b/python/tvm/tir/function.py
index 95db8aa082fb..f755fc1b0742 100644
--- a/python/tvm/tir/function.py
+++ b/python/tvm/tir/function.py
@@ -19,7 +19,8 @@
 
 import collections
 import inspect
-from typing import Callable, List, Mapping, Optional, Tuple, Union
+from collections.abc import Callable, Mapping
+from typing import Optional
 
 import tvm_ffi
 
@@ -119,7 +120,7 @@ def with_body(self, new_body, span=None):
             span,
         )
 
-    def specialize(self, param_map: Mapping[Var, Union[PrimExpr, Buffer]]):
+    def specialize(self, param_map: Mapping[Var, PrimExpr | Buffer]):
         """Specialize parameters of PrimFunc
 
         Parameters
@@ -247,8 +248,8 @@ class IndexMap(Object):
         index map.
     """
 
-    initial_indices: List[Var]
-    final_indices: List[PrimExpr]
+    initial_indices: list[Var]
+    final_indices: list[PrimExpr]
 
     # Sentinel value used to indicate which groups of pre-flattening axes
     # should be used to post-flattening axes axes.  See
@@ -265,8 +266,8 @@ def __init__(self, initial_indices, final_indices, inverse_index_map):
     @staticmethod
     def from_func(
         mapping_function: Callable,
-        ndim: Optional[int] = None,
-        inverse_index_map: Union[Callable, Optional["IndexMap"]] = None,
+        ndim: int | None = None,
+        inverse_index_map: Callable | Optional["IndexMap"] = None,
         *,
         index_dtype: str = "int64",
     ):
@@ -320,8 +321,8 @@ def from_func(
     @staticmethod
     def from_func_with_separators(
         mapping_function: Callable,
-        ndim: Optional[int] = None,
-        inverse_index_map: Union[Callable, Optional["IndexMap"]] = None,
+        ndim: int | None = None,
+        inverse_index_map: Callable | Optional["IndexMap"] = None,
         *,
         index_dtype: str = "int64",
     ):
@@ -458,7 +459,7 @@ def is_equivalent_to(self, other_map: "IndexMap") -> bool:
 
         return True
 
-    def map_indices(self, indices: List[PrimExpr]) -> List[PrimExpr]:
+    def map_indices(self, indices: list[PrimExpr]) -> list[PrimExpr]:
         """Apply the index map to a set of indices
 
         Parameters
@@ -473,7 +474,7 @@ def map_indices(self, indices: List[PrimExpr]) -> List[PrimExpr]:
         """
         return _ffi_api.IndexMapMapIndices(self, indices)
 
-    def map_shape(self, shape: List[PrimExpr]) -> List[PrimExpr]:
+    def map_shape(self, shape: list[PrimExpr]) -> list[PrimExpr]:
         """Apply the index map to a buffer shape
 
         Parameters
@@ -503,7 +504,7 @@ def map_tensor(self, arr_src: Tensor) -> Tensor:
         """
         return _ffi_api.IndexMapMapTensor(self, arr_src)
 
-    def inverse(self, shape: List[Union[Range, PrimExpr]]) -> "IndexMap":
+    def inverse(self, shape: list[Range | PrimExpr]) -> "IndexMap":
         """Return the inverse of the map
 
         Throws an error if the function is not bijective.
@@ -526,9 +527,7 @@ def inverse(self, shape: List[Union[Range, PrimExpr]]) -> "IndexMap":
         shape = [dim if isinstance(dim, Range) else Range(0, dim) for dim in shape]
         return _ffi_api.IndexMapInverse(self, shape)
 
-    def non_surjective_inverse(
-        self, shape: List[Union[Range, PrimExpr]]
-    ) -> Tuple["IndexMap", PrimExpr]:
+    def non_surjective_inverse(self, shape: list[Range | PrimExpr]) -> tuple["IndexMap", PrimExpr]:
         """Return the inverse of the map
 
         Can be applied to transformations that introduce padding.
diff --git a/python/tvm/tir/functor.py b/python/tvm/tir/functor.py
index a047fddb976a..846528cb3dd6 100644
--- a/python/tvm/tir/functor.py
+++ b/python/tvm/tir/functor.py
@@ -18,7 +18,7 @@
 # ruff: noqa: RUF012
 """The expression and statement functor of TIR."""
 
-from typing import Callable, Optional
+from collections.abc import Callable
 
 import tvm_ffi
 
@@ -157,56 +157,56 @@ class _PyStmtExprVisitor(tvm_ffi.core.Object):
 
     def __init__(
         self,
-        f_visit_stmt: Optional[Callable] = None,
-        f_visit_expr: Optional[Callable] = None,
+        f_visit_stmt: Callable | None = None,
+        f_visit_expr: Callable | None = None,
         # Stmt
-        f_visit_let_stmt: Optional[Callable] = None,
-        f_visit_attr_stmt: Optional[Callable] = None,
-        f_visit_if_then_else: Optional[Callable] = None,
-        f_visit_for: Optional[Callable] = None,
-        f_visit_while: Optional[Callable] = None,
-        f_visit_allocate: Optional[Callable] = None,
-        f_visit_decl_buffer: Optional[Callable] = None,
-        f_visit_buffer_store: Optional[Callable] = None,
-        f_visit_assert_stmt: Optional[Callable] = None,
-        f_visit_seq_stmt: Optional[Callable] = None,
-        f_visit_evaluate: Optional[Callable] = None,
-        f_visit_block: Optional[Callable] = None,
-        f_visit_sblock_realize: Optional[Callable] = None,
+        f_visit_let_stmt: Callable | None = None,
+        f_visit_attr_stmt: Callable | None = None,
+        f_visit_if_then_else: Callable | None = None,
+        f_visit_for: Callable | None = None,
+        f_visit_while: Callable | None = None,
+        f_visit_allocate: Callable | None = None,
+        f_visit_decl_buffer: Callable | None = None,
+        f_visit_buffer_store: Callable | None = None,
+        f_visit_assert_stmt: Callable | None = None,
+        f_visit_seq_stmt: Callable | None = None,
+        f_visit_evaluate: Callable | None = None,
+        f_visit_block: Callable | None = None,
+        f_visit_sblock_realize: Callable | None = None,
         # PrimExpr
-        f_visit_var: Optional[Callable] = None,
-        f_visit_size_var: Optional[Callable] = None,
-        f_visit_buffer_load: Optional[Callable] = None,
-        f_visit_producer_load: Optional[Callable] = None,
-        f_visit_let: Optional[Callable] = None,
-        f_visit_call: Optional[Callable] = None,
-        f_visit_add: Optional[Callable] = None,
-        f_visit_sub: Optional[Callable] = None,
-        f_visit_mul: Optional[Callable] = None,
-        f_visit_div: Optional[Callable] = None,
-        f_visit_mod: Optional[Callable] = None,
-        f_visit_floor_div: Optional[Callable] = None,
-        f_visit_floor_mod: Optional[Callable] = None,
-        f_visit_min: Optional[Callable] = None,
-        f_visit_max: Optional[Callable] = None,
-        f_visit_eq: Optional[Callable] = None,
-        f_visit_ne: Optional[Callable] = None,
-        f_visit_lt: Optional[Callable] = None,
-        f_visit_le: Optional[Callable] = None,
-        f_visit_gt: Optional[Callable] = None,
-        f_visit_ge: Optional[Callable] = None,
-        f_visit_and: Optional[Callable] = None,
-        f_visit_or: Optional[Callable] = None,
-        f_visit_reduce: Optional[Callable] = None,
-        f_visit_cast: Optional[Callable] = None,
-        f_visit_not: Optional[Callable] = None,
-        f_visit_select: Optional[Callable] = None,
-        f_visit_ramp: Optional[Callable] = None,
-        f_visit_broadcast: Optional[Callable] = None,
-        f_visit_shuffle: Optional[Callable] = None,
-        f_visit_int_imm: Optional[Callable] = None,
-        f_visit_float_imm: Optional[Callable] = None,
-        f_visit_string_imm: Optional[Callable] = None,
+        f_visit_var: Callable | None = None,
+        f_visit_size_var: Callable | None = None,
+        f_visit_buffer_load: Callable | None = None,
+        f_visit_producer_load: Callable | None = None,
+        f_visit_let: Callable | None = None,
+        f_visit_call: Callable | None = None,
+        f_visit_add: Callable | None = None,
+        f_visit_sub: Callable | None = None,
+        f_visit_mul: Callable | None = None,
+        f_visit_div: Callable | None = None,
+        f_visit_mod: Callable | None = None,
+        f_visit_floor_div: Callable | None = None,
+        f_visit_floor_mod: Callable | None = None,
+        f_visit_min: Callable | None = None,
+        f_visit_max: Callable | None = None,
+        f_visit_eq: Callable | None = None,
+        f_visit_ne: Callable | None = None,
+        f_visit_lt: Callable | None = None,
+        f_visit_le: Callable | None = None,
+        f_visit_gt: Callable | None = None,
+        f_visit_ge: Callable | None = None,
+        f_visit_and: Callable | None = None,
+        f_visit_or: Callable | None = None,
+        f_visit_reduce: Callable | None = None,
+        f_visit_cast: Callable | None = None,
+        f_visit_not: Callable | None = None,
+        f_visit_select: Callable | None = None,
+        f_visit_ramp: Callable | None = None,
+        f_visit_broadcast: Callable | None = None,
+        f_visit_shuffle: Callable | None = None,
+        f_visit_int_imm: Callable | None = None,
+        f_visit_float_imm: Callable | None = None,
+        f_visit_string_imm: Callable | None = None,
     ) -> None:
         """Constructor."""
         self.__init_handle_by_constructor__(
@@ -959,56 +959,56 @@ class _PyStmtExprMutator(tvm_ffi.core.Object):
 
     def __init__(
         self,
-        f_visit_stmt: Optional[Callable] = None,
-        f_visit_expr: Optional[Callable] = None,
+        f_visit_stmt: Callable | None = None,
+        f_visit_expr: Callable | None = None,
         # Stmt
-        f_visit_let_stmt: Optional[Callable] = None,
-        f_visit_attr_stmt: Optional[Callable] = None,
-        f_visit_if_then_else: Optional[Callable] = None,
-        f_visit_for: Optional[Callable] = None,
-        f_visit_while: Optional[Callable] = None,
-        f_visit_allocate: Optional[Callable] = None,
-        f_visit_decl_buffer: Optional[Callable] = None,
-        f_visit_buffer_store: Optional[Callable] = None,
-        f_visit_assert_stmt: Optional[Callable] = None,
-        f_visit_seq_stmt: Optional[Callable] = None,
-        f_visit_evaluate: Optional[Callable] = None,
-        f_visit_block: Optional[Callable] = None,
-        f_visit_sblock_realize: Optional[Callable] = None,
+        f_visit_let_stmt: Callable | None = None,
+        f_visit_attr_stmt: Callable | None = None,
+        f_visit_if_then_else: Callable | None = None,
+        f_visit_for: Callable | None = None,
+        f_visit_while: Callable | None = None,
+        f_visit_allocate: Callable | None = None,
+        f_visit_decl_buffer: Callable | None = None,
+        f_visit_buffer_store: Callable | None = None,
+        f_visit_assert_stmt: Callable | None = None,
+        f_visit_seq_stmt: Callable | None = None,
+        f_visit_evaluate: Callable | None = None,
+        f_visit_block: Callable | None = None,
+        f_visit_sblock_realize: Callable | None = None,
         # PrimExpr
-        f_visit_var: Optional[Callable] = None,
-        f_visit_size_var: Optional[Callable] = None,
-        f_visit_buffer_load: Optional[Callable] = None,
-        f_visit_producer_load: Optional[Callable] = None,
-        f_visit_let: Optional[Callable] = None,
-        f_visit_call: Optional[Callable] = None,
-        f_visit_add: Optional[Callable] = None,
-        f_visit_sub: Optional[Callable] = None,
-        f_visit_mul: Optional[Callable] = None,
-        f_visit_div: Optional[Callable] = None,
-        f_visit_mod: Optional[Callable] = None,
-        f_visit_floor_div: Optional[Callable] = None,
-        f_visit_floor_mod: Optional[Callable] = None,
-        f_visit_min: Optional[Callable] = None,
-        f_visit_max: Optional[Callable] = None,
-        f_visit_eq: Optional[Callable] = None,
-        f_visit_ne: Optional[Callable] = None,
-        f_visit_lt: Optional[Callable] = None,
-        f_visit_le: Optional[Callable] = None,
-        f_visit_gt: Optional[Callable] = None,
-        f_visit_ge: Optional[Callable] = None,
-        f_visit_and: Optional[Callable] = None,
-        f_visit_or: Optional[Callable] = None,
-        f_visit_reduce: Optional[Callable] = None,
-        f_visit_cast: Optional[Callable] = None,
-        f_visit_not: Optional[Callable] = None,
-        f_visit_select: Optional[Callable] = None,
-        f_visit_ramp: Optional[Callable] = None,
-        f_visit_broadcast: Optional[Callable] = None,
-        f_visit_shuffle: Optional[Callable] = None,
-        f_visit_int_imm: Optional[Callable] = None,
-        f_visit_float_imm: Optional[Callable] = None,
-        f_visit_string_imm: Optional[Callable] = None,
+        f_visit_var: Callable | None = None,
+        f_visit_size_var: Callable | None = None,
+        f_visit_buffer_load: Callable | None = None,
+        f_visit_producer_load: Callable | None = None,
+        f_visit_let: Callable | None = None,
+        f_visit_call: Callable | None = None,
+        f_visit_add: Callable | None = None,
+        f_visit_sub: Callable | None = None,
+        f_visit_mul: Callable | None = None,
+        f_visit_div: Callable | None = None,
+        f_visit_mod: Callable | None = None,
+        f_visit_floor_div: Callable | None = None,
+        f_visit_floor_mod: Callable | None = None,
+        f_visit_min: Callable | None = None,
+        f_visit_max: Callable | None = None,
+        f_visit_eq: Callable | None = None,
+        f_visit_ne: Callable | None = None,
+        f_visit_lt: Callable | None = None,
+        f_visit_le: Callable | None = None,
+        f_visit_gt: Callable | None = None,
+        f_visit_ge: Callable | None = None,
+        f_visit_and: Callable | None = None,
+        f_visit_or: Callable | None = None,
+        f_visit_reduce: Callable | None = None,
+        f_visit_cast: Callable | None = None,
+        f_visit_not: Callable | None = None,
+        f_visit_select: Callable | None = None,
+        f_visit_ramp: Callable | None = None,
+        f_visit_broadcast: Callable | None = None,
+        f_visit_shuffle: Callable | None = None,
+        f_visit_int_imm: Callable | None = None,
+        f_visit_float_imm: Callable | None = None,
+        f_visit_string_imm: Callable | None = None,
     ) -> None:
         """Constructor."""
         self.__init_handle_by_constructor__(
diff --git a/python/tvm/tir/op.py b/python/tvm/tir/op.py
index b0df5d40a591..616373c7658d 100644
--- a/python/tvm/tir/op.py
+++ b/python/tvm/tir/op.py
@@ -17,7 +17,7 @@
 # pylint: disable=redefined-builtin, invalid-name, too-many-arguments
 """Operators used in TIR expression."""
 
-from typing import Any, Optional, Union
+from typing import Any
 
 import tvm_ffi
 
@@ -556,7 +556,7 @@ def tvm_struct_set(arr, index, field, value):
     return call_intrin("int32", "tir.tvm_struct_set", arr, index, field, value)
 
 
-def address_of(obj: Union[Buffer, BufferLoad], span: Optional[Span] = None) -> PrimExpr:
+def address_of(obj: Buffer | BufferLoad, span: Span | None = None) -> PrimExpr:
     """Returns the address of an element in the buffer
 
     Parameters
@@ -2043,7 +2043,7 @@ def min_value(dtype, span=None):
     return _ffi_api.min_value(dtype, span)  # type: ignore
 
 
-def max_value(dtype: str, span: Optional[Span] = None) -> Any:
+def max_value(dtype: str, span: Span | None = None) -> Any:
     """maximum value of dtype
 
     Parameters
@@ -2062,7 +2062,7 @@ def max_value(dtype: str, span: Optional[Span] = None) -> Any:
     return _ffi_api.max_value(dtype, span)  # type: ignore
 
 
-def infinity(dtype: str, span: Optional[Span] = None) -> Any:
+def infinity(dtype: str, span: Span | None = None) -> Any:
     """infinity value of dtype
 
     Parameters
@@ -2081,7 +2081,7 @@ def infinity(dtype: str, span: Optional[Span] = None) -> Any:
     return _ffi_api.infinity(dtype, span)  # type: ignore
 
 
-def reinterpret(dtype, value, span: Optional[Span] = None) -> Any:
+def reinterpret(dtype, value, span: Span | None = None) -> Any:
     """infinity value of dtype
 
     Parameters
@@ -3449,7 +3449,7 @@ def _make_reduce(expr, axis, where=None, init=None):
             if init is not None:
                 init = [init]
         combiner = CommReducer(lhs, rhs, result, id_elem)
-        if not isinstance(axis, (list, tuple, tvm.ir.Array)):
+        if not isinstance(axis, list | tuple | tvm.ir.Array):
             axis = [axis]
         if where is None:
             where = tir.convert(True)
@@ -3463,7 +3463,7 @@ def _make_reduce(expr, axis, where=None, init=None):
 
     # pylint: disable=keyword-arg-before-vararg
     def reducer(expr, axis, where=None, init=None, *args):
-        if isinstance(axis, (tvm.tir.IterVar, list, tuple)):
+        if isinstance(axis, tvm.tir.IterVar | list | tuple):
             assert not args
             return _make_reduce(expr, axis, where, init)
 
@@ -3671,7 +3671,7 @@ def get_active_lane_mask(dtype, base, limit):
     return call_intrin(dtype, "tir.get_active_lane_mask", base, limit)
 
 
-def get_vscale_expr(dtype: Union[str, tvm_ffi.dtype], min_size: int = 128) -> PrimExpr:
+def get_vscale_expr(dtype: str | tvm_ffi.dtype, min_size: int = 128) -> PrimExpr:
     """
     Create a datatype dependent scalable expression.
 
diff --git a/python/tvm/tir/pipeline.py b/python/tvm/tir/pipeline.py
index c74b6ffde4c3..3cd633c99a4a 100644
--- a/python/tvm/tir/pipeline.py
+++ b/python/tvm/tir/pipeline.py
@@ -18,8 +18,6 @@
 # pylint: disable=invalid-name
 """The TIR backend compilation pipeline."""
 
-from typing import Optional
-
 import tvm
 from tvm import tir
 
@@ -49,7 +47,7 @@ def finalize_device_passes():  # pylint: disable=unused-argument
 PIPELINE_MAP = {}
 
 
-def get_tir_pipeline(name: Optional[str] = None, **kwargs) -> tvm.transform.Pass:
+def get_tir_pipeline(name: str | None = None, **kwargs) -> tvm.transform.Pass:
     """Get pre-build pipeline by name
 
     Parameters
diff --git a/python/tvm/tir/stmt.py b/python/tvm/tir/stmt.py
index 293ebed6a7c8..093adc4feddf 100644
--- a/python/tvm/tir/stmt.py
+++ b/python/tvm/tir/stmt.py
@@ -27,8 +27,8 @@
     assert(st.buffer == buffer)
 """
 
+from collections.abc import Mapping
 from enum import IntEnum
-from typing import List, Mapping, Optional, Union
 
 import tvm_ffi
 
@@ -66,9 +66,9 @@ class LetStmt(Stmt):
     var: Var
     value: PrimExpr
     body: Stmt
-    span: Optional[Span]
+    span: Span | None
 
-    def __init__(self, var: Var, value: PrimExpr, body: Stmt, span: Optional[Span] = None) -> None:
+    def __init__(self, var: Var, value: PrimExpr, body: Stmt, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.LetStmt,
             var,
@@ -90,15 +90,15 @@ class AssertStmt(Stmt):
     message : PrimExpr
         The error message.
 
-    span : Optional[Span]
+    span : Span | None
         The location of the stmt in the source code.
     """
 
     condition: PrimExpr
     message: PrimExpr
-    span: Optional[Span]
+    span: Span | None
 
-    def __init__(self, condition: PrimExpr, message: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, condition: PrimExpr, message: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.AssertStmt,
             condition,
@@ -164,10 +164,10 @@ class For(Stmt):
     extent: PrimExpr
     kind: ForKind
     body: Stmt
-    thread_binding: Optional[IterVar]
+    thread_binding: IterVar | None
     annotations: Mapping[str, Object]
-    step: Optional[PrimExpr]
-    span: Optional[Span]
+    step: PrimExpr | None
+    span: Span | None
 
     def __init__(
         self,
@@ -176,10 +176,10 @@ def __init__(
         extent: PrimExpr,
         kind: ForKind,
         body: Stmt,
-        thread_binding: Optional[IterVar] = None,
-        annotations: Optional[Mapping[str, Object]] = None,
-        step: Optional[PrimExpr] = None,
-        span: Optional[Span] = None,
+        thread_binding: IterVar | None = None,
+        annotations: Mapping[str, Object] | None = None,
+        step: PrimExpr | None = None,
+        span: Span | None = None,
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.For,  # type: ignore
@@ -213,9 +213,9 @@ class While(Stmt):
 
     condition: PrimExpr
     body: Stmt
-    span: Optional[Span]
+    span: Span | None
 
-    def __init__(self, condition: PrimExpr, body: Stmt, span: Optional[Span] = None) -> None:
+    def __init__(self, condition: PrimExpr, body: Stmt, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.While, condition, body, span)  # type: ignore
 
 
@@ -245,17 +245,17 @@ class BufferStore(Stmt):
 
     buffer: Buffer
     value: PrimExpr
-    indices: List[PrimExpr]
-    predicate: Optional[PrimExpr]
-    span: Optional[Span]
+    indices: list[PrimExpr]
+    predicate: PrimExpr | None
+    span: Span | None
 
     def __init__(
         self,
         buffer: Buffer,
         value: PrimExpr,
-        indices: List[PrimExpr],
-        predicate: Optional[PrimExpr] = None,
-        span: Optional[Span] = None,
+        indices: list[PrimExpr],
+        predicate: PrimExpr | None = None,
+        span: Span | None = None,
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.BufferStore,
@@ -297,21 +297,21 @@ class Allocate(Stmt):
 
     buffer_var: Var
     dtype: str
-    extents: List[PrimExpr]
+    extents: list[PrimExpr]
     condition: PrimExpr
     body: Stmt
     annotations: Mapping[str, Object]
-    span: Optional[Span]
+    span: Span | None
 
     def __init__(
         self,
         buffer_var: Var,
         dtype: str,
-        extents: List[PrimExpr],
+        extents: list[PrimExpr],
         condition: PrimExpr,
         body: Stmt,
-        annotations: Optional[Mapping[str, Object]] = None,
-        span: Optional[Span] = None,
+        annotations: Mapping[str, Object] | None = None,
+        span: Span | None = None,
     ) -> None:
         if annotations is None:
             annotations = dict()
@@ -345,9 +345,9 @@ class DeclBuffer(Stmt):
 
     buffer: Buffer
     body: Stmt
-    span: Optional[Span]
+    span: Span | None
 
-    def __init__(self, buffer: Buffer, body: Stmt, span: Optional[Span] = None) -> None:
+    def __init__(self, buffer: Buffer, body: Stmt, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.DeclBuffer, buffer, body, span)
 
 
@@ -377,10 +377,10 @@ class AttrStmt(Stmt):
     attr_key: str
     value: PrimExpr
     body: Stmt
-    span: Optional[Span]
+    span: Span | None
 
     def __init__(
-        self, node: Object, attr_key: str, value: PrimExpr, body: Stmt, span: Optional[Span] = None
+        self, node: Object, attr_key: str, value: PrimExpr, body: Stmt, span: Span | None = None
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.AttrStmt,
@@ -405,10 +405,10 @@ class SeqStmt(Stmt):
         The location of the stmt in the source code.
     """
 
-    seq: List[Stmt]
-    span: Optional[Span]
+    seq: list[Stmt]
+    span: Span | None
 
-    def __init__(self, seq: List[Stmt], span: Optional[Span] = None) -> None:
+    def __init__(self, seq: list[Stmt], span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.SeqStmt, seq, span)  # type: ignore
 
     def __getitem__(self, i: int):
@@ -439,14 +439,14 @@ class IfThenElse(Stmt):
 
     condition: PrimExpr
     then_case: Stmt
-    else_case: Optional[Stmt]
+    else_case: Stmt | None
 
     def __init__(
         self,
         condition: PrimExpr,
         then_case: Stmt,
-        else_case: Optional[Stmt],
-        span: Optional[Span] = None,
+        else_case: Stmt | None,
+        span: Span | None = None,
     ) -> None:
         self.__init_handle_by_constructor__(
             _ffi_api.IfThenElse,
@@ -471,9 +471,9 @@ class Evaluate(Stmt):
     """
 
     value: PrimExpr
-    span: Optional[Span]
+    span: Span | None
 
-    def __init__(self, value: PrimExpr, span: Optional[Span] = None) -> None:
+    def __init__(self, value: PrimExpr, span: Span | None = None) -> None:
         self.__init_handle_by_constructor__(_ffi_api.Evaluate, value, span)  # type: ignore
 
 
@@ -491,9 +491,9 @@ class BufferRegion(Object, Scriptable):
     """
 
     buffer: Buffer
-    region: List[Range]
+    region: list[Range]
 
-    def __init__(self, buffer: Buffer, region: List[Range]) -> None:
+    def __init__(self, buffer: Buffer, region: list[Range]) -> None:
         self.__init_handle_by_constructor__(_ffi_api.BufferRegion, buffer, region)  # type: ignore
 
 
@@ -558,29 +558,29 @@ class SBlock(Stmt):
         The location of this block in the source code.
     """
 
-    iter_vars: List[IterVar]
-    reads: List[BufferRegion]
-    writes: List[BufferRegion]
+    iter_vars: list[IterVar]
+    reads: list[BufferRegion]
+    writes: list[BufferRegion]
     name_hint: str
     body: Stmt
-    init: Optional[Stmt]
-    alloc_buffers: List[Buffer]
-    match_buffers: List[MatchBufferRegion]
+    init: Stmt | None
+    alloc_buffers: list[Buffer]
+    match_buffers: list[MatchBufferRegion]
     annotations: Mapping[str, Object]
-    span: Optional[Span]
+    span: Span | None
 
     def __init__(
         self,
-        iter_vars: List[IterVar],
-        reads: List[BufferRegion],
-        writes: List[BufferRegion],
+        iter_vars: list[IterVar],
+        reads: list[BufferRegion],
+        writes: list[BufferRegion],
         name_hint: str,
         body: Stmt,
-        init: Optional[Stmt] = None,
-        alloc_buffers: Optional[List[Buffer]] = None,
-        match_buffers: Optional[List[MatchBufferRegion]] = None,
-        annotations: Optional[Mapping[str, Object]] = None,
-        span: Optional[Span] = None,
+        init: Stmt | None = None,
+        alloc_buffers: list[Buffer] | None = None,
+        match_buffers: list[MatchBufferRegion] | None = None,
+        annotations: Mapping[str, Object] | None = None,
+        span: Span | None = None,
     ) -> None:
         if alloc_buffers is None:
             alloc_buffers = []
@@ -622,17 +622,17 @@ class SBlockRealize(Stmt):
         The location of this block_realize in the source code.
     """
 
-    iter_values: List[PrimExpr]
+    iter_values: list[PrimExpr]
     predicate: PrimExpr
     block: SBlock
-    span: Optional[Span]
+    span: Span | None
 
     def __init__(
         self,
-        iter_values: List[PrimExpr],
-        predicate: Union[PrimExpr, bool],
+        iter_values: list[PrimExpr],
+        predicate: PrimExpr | bool,
         block: SBlock,
-        span: Optional[Span] = None,
+        span: Span | None = None,
     ) -> None:
         if isinstance(predicate, bool):
             predicate = const(predicate, "bool")
@@ -645,7 +645,7 @@ def __init__(
         )  # type: ignore
 
 
-def stmt_seq(*args: Union[PrimExpr, Stmt]) -> SeqStmt:
+def stmt_seq(*args: PrimExpr | Stmt) -> SeqStmt:
     """Make sequence of statements
 
     Parameters
@@ -668,7 +668,7 @@ def stmt_seq(*args: Union[PrimExpr, Stmt]) -> SeqStmt:
     return SeqStmt(ret)
 
 
-def stmt_list(stmt: Stmt) -> List[Stmt]:
+def stmt_list(stmt: Stmt) -> list[Stmt]:
     """Make list of stmt from blocks.
 
     Parameters
diff --git a/python/tvm/tir/transform/function_pass.py b/python/tvm/tir/transform/function_pass.py
index fc3aee0c2a7b..bb79914c0fd0 100644
--- a/python/tvm/tir/transform/function_pass.py
+++ b/python/tvm/tir/transform/function_pass.py
@@ -18,7 +18,7 @@
 
 import functools
 import inspect
-from typing import Callable, List, Optional, Union
+from collections.abc import Callable
 
 import tvm_ffi
 
@@ -69,11 +69,11 @@ def __getattr__(self, name):
 
 def prim_func_pass(
     pass_func=None,
-    opt_level: Optional[int] = None,
-    name: Optional[str] = None,
-    required: Optional[List[str]] = None,
+    opt_level: int | None = None,
+    name: str | None = None,
+    required: list[str] | None = None,
     traceable=False,
-) -> Union[Callable, PrimFuncPass]:
+) -> Callable | PrimFuncPass:
     """Decorate a function pass.
 
     This function returns a callback when pass_func
@@ -145,7 +145,7 @@ def transform(func, mod, ctx):
         raise ValueError("Please provide opt_level for the function pass.")
 
     required = required if required else []
-    if not isinstance(required, (list, tuple)):
+    if not isinstance(required, list | tuple):
         raise TypeError("Required is expected to be the type of " + "list/tuple.")
 
     def create_function_pass(pass_arg):
diff --git a/python/tvm/tir/transform/transform.py b/python/tvm/tir/transform/transform.py
index f5343e9c5603..ceee1edec0ce 100644
--- a/python/tvm/tir/transform/transform.py
+++ b/python/tvm/tir/transform/transform.py
@@ -18,7 +18,7 @@
 # pylint: disable=invalid-name, unsupported-binary-operation
 
 import enum
-from typing import Callable
+from collections.abc import Callable
 
 from ... import ffi as _ffi
 from ... import ir as _ir
diff --git a/python/tvm/topi/broadcast.py b/python/tvm/topi/broadcast.py
index cb0db387ae00..c00495b03237 100644
--- a/python/tvm/topi/broadcast.py
+++ b/python/tvm/topi/broadcast.py
@@ -16,8 +16,6 @@
 # under the License.
 """Broadcast operators"""
 
-from __future__ import absolute_import as _abs
-
 from . import cpp as _cpp
 
 
diff --git a/python/tvm/topi/gpu/scan.py b/python/tvm/topi/gpu/scan.py
index c009a5b2db04..cb0503a754cc 100644
--- a/python/tvm/topi/gpu/scan.py
+++ b/python/tvm/topi/gpu/scan.py
@@ -17,7 +17,7 @@
 # pylint: disable=invalid-name, too-many-locals, too-many-statements
 "Scan related operators"
 
-from typing import Callable, Optional, Union
+from collections.abc import Callable
 
 import tvm
 from tvm import te
@@ -610,11 +610,11 @@ def inclusive_scan(
 def scanop(
     data: tvm.te.Tensor,
     binop: Callable[["tvm.Expr", "tvm.Expr"], "tvm.Expr"],
-    identity_value: Union[float, int],
-    axis: Optional[int] = None,
-    dtype: Optional[str] = None,
-    exclusive: Optional[bool] = None,
-    workspace: Optional[tvm.te.Tensor] = None,
+    identity_value: float | int,
+    axis: int | None = None,
+    dtype: str | None = None,
+    exclusive: bool | None = None,
+    workspace: tvm.te.Tensor | None = None,
 ) -> tvm.te.Tensor:
     """Cumulative binary operator (scan) with similar axis behavior as np.cumsum and np.cumprod.
 
@@ -684,10 +684,10 @@ def scanop(
 
 def cumsum(
     data: tvm.te.Tensor,
-    axis: Optional[int] = None,
-    dtype: Optional[int] = None,
-    exclusive: Optional[bool] = None,
-    workspace: Optional[tvm.te.Tensor] = None,
+    axis: int | None = None,
+    dtype: int | None = None,
+    exclusive: bool | None = None,
+    workspace: tvm.te.Tensor | None = None,
 ) -> tvm.te.Tensor:
     """Numpy style cumsum op. Return the cumulative sum of the elements along a given axis.
 
@@ -734,10 +734,10 @@ def cumsum(
 
 def cumprod(
     data: tvm.te.Tensor,
-    axis: Optional[int] = None,
-    dtype: Optional[int] = None,
-    exclusive: Optional[bool] = None,
-    workspace: Optional[tvm.te.Tensor] = None,
+    axis: int | None = None,
+    dtype: int | None = None,
+    exclusive: bool | None = None,
+    workspace: tvm.te.Tensor | None = None,
 ):
     """Numpy style cumprod op. Return the cumulative product of the elements along a given axis.
 
diff --git a/python/tvm/topi/image/__init__.py b/python/tvm/topi/image/__init__.py
index ee39f2087ff3..8a2cbad028dd 100644
--- a/python/tvm/topi/image/__init__.py
+++ b/python/tvm/topi/image/__init__.py
@@ -19,8 +19,6 @@
 # pylint: disable=wildcard-import
 """IMAGE network operators"""
 
-from __future__ import absolute_import as _abs
-
 from .resize import *
 from .dilation2d import *
 from .grid_sample import *
diff --git a/python/tvm/topi/image/dilation2d.py b/python/tvm/topi/image/dilation2d.py
index cf8a59a5d132..e635c4e5df6d 100644
--- a/python/tvm/topi/image/dilation2d.py
+++ b/python/tvm/topi/image/dilation2d.py
@@ -18,8 +18,6 @@
 # pylint: disable=unused-argument, redefined-builtin
 """Dilation2D operators"""
 
-from __future__ import absolute_import as _abs
-
 from tvm import te
 from tvm.topi.utils import simplify
 
diff --git a/python/tvm/topi/index_put.py b/python/tvm/topi/index_put.py
index c0cf0d216300..d3e5ba42be12 100644
--- a/python/tvm/topi/index_put.py
+++ b/python/tvm/topi/index_put.py
@@ -49,7 +49,7 @@ def index_put(data, indices, values, accumulate=False):
     -------
     ret : tvm.te.Tensor
     """
-    if not isinstance(indices, (list, tuple)):
+    if not isinstance(indices, list | tuple):
         indices = [indices]
 
     # Check indices match data dimensions
diff --git a/python/tvm/topi/nn/__init__.py b/python/tvm/topi/nn/__init__.py
index 7b8dd59900b6..c6a24f92fd3e 100644
--- a/python/tvm/topi/nn/__init__.py
+++ b/python/tvm/topi/nn/__init__.py
@@ -19,8 +19,6 @@
 # pylint: disable=wildcard-import
 """Neural network operators"""
 
-from __future__ import absolute_import as _abs
-
 from .conv1d import *
 from .conv2d import *
 from .conv3d import *
diff --git a/python/tvm/topi/nn/batch_norm.py b/python/tvm/topi/nn/batch_norm.py
index 7de22a65b54e..8da1341d1b6a 100644
--- a/python/tvm/topi/nn/batch_norm.py
+++ b/python/tvm/topi/nn/batch_norm.py
@@ -16,7 +16,6 @@
 # under the License.
 """Batch normalization."""
 
-import typing
 from functools import reduce
 
 from tvm import te, topi
@@ -28,13 +27,13 @@ def batch_norm(
     beta: te.Tensor,
     moving_mean: te.Tensor,
     moving_var: te.Tensor,
-    axis: typing.Optional[int] = None,
-    epsilon: typing.Optional[float] = None,
-    center: typing.Optional[bool] = None,
-    scale: typing.Optional[bool] = None,
-    training: typing.Optional[bool] = None,
-    momentum: typing.Optional[float] = None,
-) -> typing.List[te.Tensor]:
+    axis: int | None = None,
+    epsilon: float | None = None,
+    center: bool | None = None,
+    scale: bool | None = None,
+    training: bool | None = None,
+    momentum: float | None = None,
+) -> list[te.Tensor]:
     """Batch normalization layer (Ioffe and Szegedy, 2014).
 
     Normalizes the input at each batch, i.e. applies a transformation
diff --git a/python/tvm/topi/nn/bitserial_conv2d.py b/python/tvm/topi/nn/bitserial_conv2d.py
index 0e2d9074d9e3..76a1940412d9 100644
--- a/python/tvm/topi/nn/bitserial_conv2d.py
+++ b/python/tvm/topi/nn/bitserial_conv2d.py
@@ -83,7 +83,7 @@ def bitserial_conv2d_nchw(
     batch, in_channel, activation_bits, in_height, in_width = Input_q.shape
     num_filter, _, kernel_h, kernel_w, weight_bits = Filter_q.shape
 
-    if isinstance(padding, int) or (isinstance(padding, (tuple, list)) and len(padding) == 2):
+    if isinstance(padding, int) or (isinstance(padding, tuple | list) and len(padding) == 2):
         TPAD, LPAD, DPAD, RPAD = get_pad_tuple(padding, kernel)
     else:
         TPAD, LPAD, DPAD, RPAD = padding
@@ -207,7 +207,7 @@ def bitserial_conv2d_nhwc(
         kernel_h, kernel_w, _, _, num_filter = get_const_tuple(Filter_q.shape)
     batch, in_height, in_width, in_channel_q, _ = get_const_tuple(Input_q.shape)
 
-    if isinstance(padding, int) or (isinstance(padding, (tuple, list)) and len(padding) == 2):
+    if isinstance(padding, int) or (isinstance(padding, tuple | list) and len(padding) == 2):
         TPAD, LPAD, DPAD, RPAD = get_pad_tuple(padding, kernel)
     else:
         TPAD, LPAD, DPAD, RPAD = padding
diff --git a/python/tvm/topi/nn/bnn.py b/python/tvm/topi/nn/bnn.py
index b98cb2f64c41..2e79d5118b12 100644
--- a/python/tvm/topi/nn/bnn.py
+++ b/python/tvm/topi/nn/bnn.py
@@ -16,8 +16,6 @@
 # under the License.
 """Binary Neural Network (BNN) Operators"""
 
-from __future__ import absolute_import as _abs
-
 import tvm
 from tvm import te
 
diff --git a/python/tvm/topi/nn/conv1d_transpose.py b/python/tvm/topi/nn/conv1d_transpose.py
index 24dfd790f038..2a850955934f 100644
--- a/python/tvm/topi/nn/conv1d_transpose.py
+++ b/python/tvm/topi/nn/conv1d_transpose.py
@@ -63,9 +63,9 @@ def _conv1d_transpose_ncw_preprocess(data, kernel, stride, padding, out_dtype, o
         out_dtype = data.dtype
 
     # dilate and pad
-    if isinstance(stride, (tuple, list)):
+    if isinstance(stride, tuple | list):
         stride = stride[0]
-    if isinstance(output_padding, (tuple, list)):
+    if isinstance(output_padding, tuple | list):
         output_padding = output_padding[0]
 
     _, channels_in, _ = data.shape
diff --git a/python/tvm/topi/nn/conv2d.py b/python/tvm/topi/nn/conv2d.py
index 4a3c017c9051..7e46ea553925 100644
--- a/python/tvm/topi/nn/conv2d.py
+++ b/python/tvm/topi/nn/conv2d.py
@@ -19,11 +19,9 @@
 # ruff: noqa: F841, RUF005
 """Conv2D operators"""
 
-from __future__ import absolute_import as _abs
-
 import re
 from collections import namedtuple
-from typing import Optional, Sequence, Union
+from collections.abc import Sequence
 
 import numpy as np
 
@@ -118,14 +116,14 @@ def _get_workload(data, kernel, stride, padding, dilation, out_dtype, data_layou
         KH, KW, CIG, CO = get_const_tuple(kernel.shape)
 
     dilation_h, dilation_w = (
-        dilation if isinstance(dilation, (tuple, list)) else (dilation, dilation)
+        dilation if isinstance(dilation, tuple | list) else (dilation, dilation)
     )
     pt, pl, pb, pr = get_pad_tuple(
         padding,
         (get_const_int((KH - 1) * dilation_h + 1), get_const_int((KW - 1) * dilation_w + 1)),
     )
     GRPS = CI // CIG
-    if isinstance(stride, (tuple, list)):
+    if isinstance(stride, tuple | list):
         HSTR, WSTR = stride
     else:
         HSTR, WSTR = stride, stride
@@ -315,9 +313,9 @@ def conv2d_NCHWc(data, kernel, stride, padding, dilation, layout, out_layout, ou
 
     # layout and out_layout are not used here,
     # we keep them for debug convenience when dumping autotvm workload
-    HSTR, WSTR = stride if isinstance(stride, (tuple, list)) else (stride, stride)
+    HSTR, WSTR = stride if isinstance(stride, tuple | list) else (stride, stride)
     dilation_h, dilation_w = (
-        dilation if isinstance(dilation, (tuple, list)) else (dilation, dilation)
+        dilation if isinstance(dilation, tuple | list) else (dilation, dilation)
     )
 
     n, ic_chunk, ih, iw, ic_bn = get_const_tuple(data.shape)
@@ -440,9 +438,9 @@ def conv2d_NCHWc_OIHWo(
 
     # layout and out_layout are not used here,
     # we keep them for debug convenience when dumping autotvm workload
-    HSTR, WSTR = stride if isinstance(stride, (tuple, list)) else (stride, stride)
+    HSTR, WSTR = stride if isinstance(stride, tuple | list) else (stride, stride)
     dilation_h, dilation_w = (
-        dilation if isinstance(dilation, (tuple, list)) else (dilation, dilation)
+        dilation if isinstance(dilation, tuple | list) else (dilation, dilation)
     )
 
     n, ic_chunk, ih, iw, ic_bn = get_const_tuple(data.shape)
@@ -572,9 +570,9 @@ def conv2d_NCHWc_int8(
 
     # layout and out_layout are not used here,
     # we keep them for debug convenience when dumping autotvm workload
-    HSTR, WSTR = stride if isinstance(stride, (tuple, list)) else (stride, stride)
+    HSTR, WSTR = stride if isinstance(stride, tuple | list) else (stride, stride)
     dilation_h, dilation_w = (
-        dilation if isinstance(dilation, (tuple, list)) else (dilation, dilation)
+        dilation if isinstance(dilation, tuple | list) else (dilation, dilation)
     )
 
     n, ic_chunk, ih, iw, ic_bn = get_const_tuple(data.shape)
@@ -735,14 +733,14 @@ def group_conv2d_nchw(Input, Filter, stride, padding, dilation, groups, out_dtyp
 def conv(
     inp: te.Tensor,
     filt: te.Tensor,
-    stride: Union[int, Sequence[int]],
-    padding: Union[int, Sequence[int]],
-    dilation: Union[int, Sequence[int]],
+    stride: int | Sequence[int],
+    padding: int | Sequence[int],
+    dilation: int | Sequence[int],
     groups: int,
     data_layout: str,
     kernel_layout: str = "",
-    out_dtype: Union[str, None] = None,
-    auto_scheduler_rewritten_layout: Optional[str] = None,
+    out_dtype: str | None = None,
+    auto_scheduler_rewritten_layout: str | None = None,
     meta_schedule_original_shape=None,
     auto_scheduler_should_rewrite_layout: bool = False,
 ):
diff --git a/python/tvm/topi/nn/conv3d_transpose.py b/python/tvm/topi/nn/conv3d_transpose.py
index 8db8759e9fb9..308619bb178f 100644
--- a/python/tvm/topi/nn/conv3d_transpose.py
+++ b/python/tvm/topi/nn/conv3d_transpose.py
@@ -156,7 +156,7 @@ def group_conv3d_transpose_ncdhw(data, kernel, strides, padding, out_dtype, outp
     Output : tvm.te.Tensor
         5-D with shape [batch, out_channel, out_depth, out_height, out_width]
     """
-    if not isinstance(strides, (tuple, list)):
+    if not isinstance(strides, tuple | list):
         strides = (strides, strides, strides)
 
     if groups == 1:
diff --git a/python/tvm/topi/nn/correlation.py b/python/tvm/topi/nn/correlation.py
index dcc885fba47f..1d0c06e5d985 100644
--- a/python/tvm/topi/nn/correlation.py
+++ b/python/tvm/topi/nn/correlation.py
@@ -65,7 +65,7 @@ def correlation_nchw(
     data_shape = get_const_tuple(data1.shape)
     assert get_const_tuple(data2.shape) == data_shape, "data1 and data2 should have the same shape"
     assert kernel_size > 0 and kernel_size % 2, "kernel_size should be non-negative odd number"
-    if isinstance(padding, (tuple, list)):
+    if isinstance(padding, tuple | list):
         if len(padding) == 2:
             pad_before_h = pad_after_h = padding[0]
             pad_before_w = pad_after_w = padding[1]
diff --git a/python/tvm/topi/nn/depthwise_conv2d.py b/python/tvm/topi/nn/depthwise_conv2d.py
index 89ce8ed1e2dd..7b2f3bedb613 100644
--- a/python/tvm/topi/nn/depthwise_conv2d.py
+++ b/python/tvm/topi/nn/depthwise_conv2d.py
@@ -18,8 +18,6 @@
 # ruff: noqa: F841
 """Depthwise convolution operators"""
 
-from __future__ import absolute_import as _abs
-
 from collections import namedtuple
 
 import numpy as np
@@ -91,9 +89,9 @@ def _get_workload(data, kernel, stride, padding, dilation, out_dtype, data_layou
 
     out_channel = filter_channel * channel_multiplier
     dilation_h, dilation_w = (
-        dilation if isinstance(dilation, (tuple, list)) else (dilation, dilation)
+        dilation if isinstance(dilation, tuple | list) else (dilation, dilation)
     )
-    if isinstance(stride, (tuple, list)):
+    if isinstance(stride, tuple | list):
         HSTR, WSTR = stride
     else:
         HSTR, WSTR = stride, stride
diff --git a/python/tvm/topi/nn/elemwise.py b/python/tvm/topi/nn/elemwise.py
index 7856260698a4..1c9d5b693f90 100644
--- a/python/tvm/topi/nn/elemwise.py
+++ b/python/tvm/topi/nn/elemwise.py
@@ -16,8 +16,6 @@
 # under the License.
 """Elementwise operators"""
 
-from __future__ import absolute_import as _abs
-
 import tvm
 from tvm import te
 
diff --git a/python/tvm/topi/nn/fifo_buffer.py b/python/tvm/topi/nn/fifo_buffer.py
index dd185c17c280..a07d0e64e1a7 100644
--- a/python/tvm/topi/nn/fifo_buffer.py
+++ b/python/tvm/topi/nn/fifo_buffer.py
@@ -18,8 +18,6 @@
 
 """FIFO buffer op"""
 
-from __future__ import absolute_import as _abs
-
 import tvm
 from tvm import te
 
diff --git a/python/tvm/topi/nn/mapping.py b/python/tvm/topi/nn/mapping.py
index 12d15be8f58a..dbc3ab56d950 100644
--- a/python/tvm/topi/nn/mapping.py
+++ b/python/tvm/topi/nn/mapping.py
@@ -17,8 +17,6 @@
 # pylint: disable=invalid-name, line-too-long
 """Operators of one-to-one-mapping on the first input"""
 
-from __future__ import absolute_import as _abs
-
 import tvm
 from tvm import te
 
diff --git a/python/tvm/topi/nn/pad.py b/python/tvm/topi/nn/pad.py
index f7964f7a23c3..813c4421ab55 100644
--- a/python/tvm/topi/nn/pad.py
+++ b/python/tvm/topi/nn/pad.py
@@ -16,8 +16,6 @@
 # under the License.
 """Pad the data by constant value"""
 
-from __future__ import absolute_import as _abs
-
 import tvm
 from tvm import te
 from tvm.tir import if_then_else
diff --git a/python/tvm/topi/nn/utils.py b/python/tvm/topi/nn/utils.py
index eec68f718210..0198654c5a13 100644
--- a/python/tvm/topi/nn/utils.py
+++ b/python/tvm/topi/nn/utils.py
@@ -141,7 +141,7 @@ def get_pad_tuple(padding, kernel):
         Padding size on right.
     """
     # compute the padding size
-    if isinstance(padding, (tuple, list)):
+    if isinstance(padding, tuple | list):
         if len(padding) == 2:
             pad_h = padding[0] * 2
             pad_w = padding[1] * 2
@@ -190,7 +190,7 @@ def get_pad_tuple_generic(padding, kernel):
         Padding size on right.
     """
     # compute the padding size
-    if isinstance(padding, (tuple, list)):
+    if isinstance(padding, tuple | list):
         if len(padding) == len(kernel):
             pad_dimensions = [p * 2 for p in padding]
         elif len(padding) == len(kernel) * 2:
@@ -244,7 +244,7 @@ def get_pad_tuple3d(padding, kernel):
         Padding size on right.
     """
     # compute the padding size
-    if isinstance(padding, (tuple, list)):
+    if isinstance(padding, tuple | list):
         if len(padding) == 3:
             pad_d = padding[0] * 2
             pad_h = padding[1] * 2
@@ -291,7 +291,7 @@ def get_pad_tuple1d(padding, kernel):
         Padding size on right.
     """
     # compute the padding size
-    if isinstance(padding, (tuple, list)):
+    if isinstance(padding, tuple | list):
         if len(padding) == 1:
             pad_w = padding[0] * 2
         elif len(padding) == 2:
diff --git a/python/tvm/topi/reduction.py b/python/tvm/topi/reduction.py
index 5d1cb714b33f..2531b58494b7 100644
--- a/python/tvm/topi/reduction.py
+++ b/python/tvm/topi/reduction.py
@@ -17,8 +17,6 @@
 # pylint: disable=redefined-builtin,consider-using-enumerate,no-member
 """Reduce operators"""
 
-from __future__ import absolute_import as _abs
-
 from . import cpp
 
 
@@ -29,7 +27,7 @@ def _get_real_axis(ndim, axis):
         if isinstance(axis, int):
             axis = [axis]
         else:
-            assert isinstance(axis, (list, tuple))
+            assert isinstance(axis, list | tuple)
         real_axis = []
         for ele in axis:
             if ele < 0:
diff --git a/python/tvm/topi/scan.py b/python/tvm/topi/scan.py
index 756b694206d3..c18376ef63c2 100644
--- a/python/tvm/topi/scan.py
+++ b/python/tvm/topi/scan.py
@@ -17,7 +17,7 @@
 # pylint: disable=invalid-name
 """Scan (cumulative binary) operators"""
 
-from typing import Callable, Optional
+from collections.abc import Callable
 
 import tvm
 from tvm.script.ir_builder import IRBuilder
@@ -34,9 +34,9 @@ def scanop(
     binop: Callable[["tvm.Expr", "tvm.Expr"], "tvm.Expr"],
     identity_value: "tvm.Expr",
     op_name: str,
-    axis: Optional[int] = None,
-    dtype: Optional[str] = None,
-    exclusive: Optional[bool] = None,
+    axis: int | None = None,
+    dtype: str | None = None,
+    exclusive: bool | None = None,
 ) -> tvm.te.Tensor:
     """Cumulative binary operator (scan) with similar axis behavior as np.cumsum and np.cumprod.
 
@@ -153,9 +153,9 @@ def gen_ir(data_buf, out_buf):
 
 def cumsum(
     data: tvm.te.Tensor,
-    axis: Optional[int] = None,
-    dtype: Optional[str] = None,
-    exclusive: Optional[bool] = None,
+    axis: int | None = None,
+    dtype: str | None = None,
+    exclusive: bool | None = None,
 ) -> tvm.te.Tensor:
     """Numpy style cumsum op. Return the cumulative sum of the elements along a given axis.
 
@@ -197,9 +197,9 @@ def cumsum(
 
 def cumprod(
     data: tvm.te.Tensor,
-    axis: Optional[int] = None,
-    dtype: Optional[int] = None,
-    exclusive: Optional[bool] = None,
+    axis: int | None = None,
+    dtype: int | None = None,
+    exclusive: bool | None = None,
 ) -> tvm.te.Tensor:
     """Numpy style cumprod op. Return the cumulative product of the elements along a given axis.
 
diff --git a/python/tvm/topi/tensor.py b/python/tvm/topi/tensor.py
index 1f08b15fc4f9..3231023503f5 100644
--- a/python/tvm/topi/tensor.py
+++ b/python/tvm/topi/tensor.py
@@ -18,7 +18,6 @@
 """Elementwise operators"""
 
 import math as _math
-from typing import Optional
 
 from tvm import te
 
@@ -59,9 +58,7 @@ def full(shape, dtype, fill_value):
         The result.
     """
 
-    if isinstance(fill_value, (int, float)) and (
-        _math.isinf(fill_value) or _math.isnan(fill_value)
-    ):
+    if isinstance(fill_value, int | float) and (_math.isinf(fill_value) or _math.isnan(fill_value)):
         if not ("float" in dtype or "bfloat16" in dtype):
             raise ValueError("Infinite and NaN require a floating-point dtype.")
 
@@ -87,7 +84,7 @@ def full_like(x, fill_value):
     return cpp.full_like(x, fill_value)
 
 
-def eye(n: int, m: Optional[int] = None, k: int = 0, dtype: str = "float32") -> te.Tensor:
+def eye(n: int, m: int | None = None, k: int = 0, dtype: str = "float32") -> te.Tensor:
     """Generate an identity matrix or a matrix with ones on the k-th diagonal.
 
     Parameters
diff --git a/python/tvm/topi/testing/__init__.py b/python/tvm/topi/testing/__init__.py
index c94fce144af4..d9fd005921ed 100644
--- a/python/tvm/topi/testing/__init__.py
+++ b/python/tvm/topi/testing/__init__.py
@@ -21,8 +21,6 @@
 Used to verify the correctness of operators in TOPI .
 """
 
-from __future__ import absolute_import as _abs
-
 from .conv1d_ncw_python import conv1d_ncw_python, group_conv1d_ncw_python
 from .conv2d_hwcn_python import conv2d_hwcn_python
 from .conv2d_nchw_python import conv2d_nchw_python
diff --git a/python/tvm/topi/testing/attention_python.py b/python/tvm/topi/testing/attention_python.py
index d06092be5b43..0352dd2040dc 100644
--- a/python/tvm/topi/testing/attention_python.py
+++ b/python/tvm/topi/testing/attention_python.py
@@ -17,8 +17,6 @@
 
 """Attention operator in python"""
 
-from typing import Optional
-
 import numpy as np
 
 from .softmax_python import softmax_python
@@ -28,10 +26,10 @@ def attention_python(
     q: np.ndarray,
     k: np.ndarray,
     v: np.ndarray,
-    bias: Optional[np.ndarray],
+    bias: np.ndarray | None,
     qk_scale: float,
     causal: str,
-    window_size: Optional[int] = None,
+    window_size: int | None = None,
     layout: str = "BSNH",
 ):  # pylint: disable=too-many-arguments, too-many-locals, invalid-name
     """Attention operator in python
diff --git a/python/tvm/topi/testing/conv1d_ncw_python.py b/python/tvm/topi/testing/conv1d_ncw_python.py
index e514a376a02f..11942ce58a1d 100644
--- a/python/tvm/topi/testing/conv1d_ncw_python.py
+++ b/python/tvm/topi/testing/conv1d_ncw_python.py
@@ -87,9 +87,9 @@ def conv1d_ncw_python(a_np, w_np, stride, padding, dilation):
     """
     batch, in_c, in_w = a_np.shape
     out_c, _, filter_w = w_np.shape
-    if isinstance(stride, (tuple, list)):
+    if isinstance(stride, tuple | list):
         stride = stride[0]
-    if isinstance(dilation, (tuple, list)):
+    if isinstance(dilation, tuple | list):
         dilation = dilation[0]
 
     dilated_filter_w = (filter_w - 1) * dilation + 1
diff --git a/python/tvm/topi/testing/matrix_set_diag.py b/python/tvm/topi/testing/matrix_set_diag.py
index 052d9a3677d4..ef22053748af 100644
--- a/python/tvm/topi/testing/matrix_set_diag.py
+++ b/python/tvm/topi/testing/matrix_set_diag.py
@@ -55,7 +55,7 @@ def matrix_set_diag(input_np, diagonal, k=0, align="RIGHT_LEFT"):
     rows = input_np.shape[-2]
 
     onlyOneDiagonal = True
-    if isinstance(k, (tuple, list)):
+    if isinstance(k, tuple | list):
         if len(k) < 2 or k[0] == k[1]:
             k = k[0]
         else:
diff --git a/python/tvm/topi/testing/poolnd_python.py b/python/tvm/topi/testing/poolnd_python.py
index 39e77a1500d6..5ab39f97af3c 100644
--- a/python/tvm/topi/testing/poolnd_python.py
+++ b/python/tvm/topi/testing/poolnd_python.py
@@ -20,7 +20,6 @@
 
 import itertools
 import math
-from typing import List, Optional, Tuple
 
 import numpy as np
 
@@ -72,11 +71,11 @@ def _convert_from_layout(input_tensor: np.ndarray, layout: str) -> np.ndarray:
 def get_slice(
     spatial_dimensions: int,
     pad_np: np.array,
-    dim_coord: Tuple[int],
-    kernel: Tuple[int],
-    strides: Tuple[int],
-    dilation: Tuple[int],
-) -> Tuple[slice]:
+    dim_coord: tuple[int],
+    kernel: tuple[int],
+    strides: tuple[int],
+    dilation: tuple[int],
+) -> tuple[slice]:
     """
     Programmatically create a slice object of the right dimensions for pad_np.
 
@@ -103,8 +102,8 @@ def get_slice(
 def pad_tensor(
     np_arr: np.array,
     pad_value: float,
-    padding_before: List[int],
-    padding_after: List[int],
+    padding_before: list[int],
+    padding_after: list[int],
     dtype: str,
 ) -> np.array:
     """Pad the spatial dimensions of the given array."""
@@ -126,16 +125,16 @@ def pad_tensor(
 
 def poolnd_python(
     np_data: np.array,
-    kernel: Tuple[int],
-    strides: Tuple[int],
-    dilation: Tuple[int],
-    padding_before: Tuple[int],
-    padding_after: Tuple[int],
+    kernel: tuple[int],
+    strides: tuple[int],
+    dilation: tuple[int],
+    padding_before: tuple[int],
+    padding_after: tuple[int],
     pool_type: str,
     count_include_pad: bool = True,
     ceil_mode: bool = False,
     dtype: str = "float32",
-    layout: Optional[str] = None,
+    layout: str | None = None,
 ) -> np.array:
     """Ground truth pooling operator impelmented in numpy."""
 
diff --git a/python/tvm/topi/transform.py b/python/tvm/topi/transform.py
index ec6242098d96..bc187e3f269b 100644
--- a/python/tvm/topi/transform.py
+++ b/python/tvm/topi/transform.py
@@ -17,8 +17,6 @@
 # pylint: disable=invalid-name,consider-using-enumerate,redefined-outer-name
 """Injective transformation operators"""
 
-from __future__ import absolute_import as _abs
-
 from math import pi
 
 import numpy as np
@@ -931,7 +929,7 @@ def matrix_set_diag(data, diagonal, k=0, align="RIGHT_LEFT"):
               [7, 5, 7, 7],
               [7, 7, 6, 7]]]
     """
-    if isinstance(k, (tuple, list)):
+    if isinstance(k, tuple | list):
         k_one = k[0]
         if len(k) >= 2:
             k_two = k[1]
diff --git a/python/tvm/topi/utils.py b/python/tvm/topi/utils.py
index 72e5a4479980..2e2b1b01ad16 100644
--- a/python/tvm/topi/utils.py
+++ b/python/tvm/topi/utils.py
@@ -18,8 +18,6 @@
 # ruff: noqa: RUF005
 """Common topi utilities"""
 
-from __future__ import absolute_import as _abs
-
 from numbers import Integral
 
 import numpy as np
@@ -193,7 +191,7 @@ def get_const_tuple(in_tuple):
     for elem in in_tuple:
         if isinstance(elem, tvm.tir.Var):
             ret.append(elem)
-        elif not isinstance(elem, (tvm.tir.IntImm, int)):
+        elif not isinstance(elem, tvm.tir.IntImm | int):
             ana = tvm.arith.Analyzer() if ana is None else ana
             elem = ana.simplify(elem)
             if not isinstance(elem, tvm.tir.IntImm):
diff --git a/python/tvm/topi/vision/nms.py b/python/tvm/topi/vision/nms.py
index 1140892c9645..6a9964e4b24d 100644
--- a/python/tvm/topi/vision/nms.py
+++ b/python/tvm/topi/vision/nms.py
@@ -59,7 +59,7 @@ def get_valid_counts(data, score_threshold=0, id_index=0, score_index=1):  # pyl
     out_indices: tvm.te.Tensor or numpy NDArray
         Related index in input data.
     """
-    if isinstance(score_threshold, (float, int)):
+    if isinstance(score_threshold, float | int):
         score_threshold = tvm.tir.const(score_threshold, dtype=data.dtype)
     # id_index_const = tvm.tir.const(id_index, "int32")  # Unused
     # score_index_const = tvm.tir.const(score_index, "int32")  # Unused
diff --git a/tests/conda/build-environment.yaml b/tests/conda/build-environment.yaml
index 28650499ea7c..ebd45ff4c422 100644
--- a/tests/conda/build-environment.yaml
+++ b/tests/conda/build-environment.yaml
@@ -29,7 +29,7 @@ dependencies:
   - ninja
   - make
   - llvmdev >=11
-  - python >=3.9
+  - python >=3.10
   - pip
   - git
   - bzip2
diff --git a/tests/python/arith/test_arith_intset.py b/tests/python/arith/test_arith_intset.py
index 0832a397cf14..78871d065cb4 100644
--- a/tests/python/arith/test_arith_intset.py
+++ b/tests/python/arith/test_arith_intset.py
@@ -160,7 +160,7 @@ def check_region_bound(expect_region, var_dom, mode, predicate=None):
     region = []
     expect = []
     for k, v in expect_region.items():
-        if not isinstance(k, (tuple, list)):
+        if not isinstance(k, tuple | list):
             k = (k, k + 1)
         region.append(tvm.ir.Range.from_min_extent(k[0], Analyzer().simplify(k[1] - k[0])))
         expect.append(v)
diff --git a/tests/python/ci/test_tvmbot.py b/tests/python/ci/test_tvmbot.py
index e89edbfd9517..362a87d4fa9b 100644
--- a/tests/python/ci/test_tvmbot.py
+++ b/tests/python/ci/test_tvmbot.py
@@ -21,7 +21,7 @@
 
 import json
 from pathlib import Path
-from typing import Any, Dict
+from typing import Any
 
 import tvm
 
@@ -40,7 +40,7 @@
 class _TvmBotTest:
     NUMBER = 10786
 
-    def preprocess_data(self, data: Dict[str, Any]):
+    def preprocess_data(self, data: dict[str, Any]):
         """
         Used to pre-process PR data before running the test. Override as
         necessary to edit data for specific test cases.
@@ -108,7 +108,7 @@ class TestNoRequest(_TvmBotTest):
     USER = "abc"
     EXPECTED = "Command 'do something else' did not match anything"
 
-    def preprocess_data(self, data: Dict[str, Any]):
+    def preprocess_data(self, data: dict[str, Any]):
         data["reviews"]["nodes"][0]["body"] = "nothing"
         return data
 
@@ -132,7 +132,7 @@ class TestBadCI(_TvmBotTest):
     USER = "abc"
     EXPECTED = "Cannot merge, these CI jobs are not successful on"
 
-    def preprocess_data(self, data: Dict[str, Any]):
+    def preprocess_data(self, data: dict[str, Any]):
         # Mark the Jenkins build as failed
         contexts = data["commits"]["nodes"][0]["commit"]["statusCheckRollup"]["contexts"]["nodes"]
         for context in contexts:
@@ -150,7 +150,7 @@ class TestOldReview(_TvmBotTest):
     USER = "abc"
     EXPECTED = "Cannot merge, did not find any approving reviews"
 
-    def preprocess_data(self, data: Dict[str, Any]) -> Dict[str, Any]:
+    def preprocess_data(self, data: dict[str, Any]) -> dict[str, Any]:
         data["reviews"]["nodes"][0]["commit"]["oid"] = "abc12345"
         return data
 
@@ -164,7 +164,7 @@ class TestMissingJob(_TvmBotTest):
     USER = "abc"
     EXPECTED = "Cannot merge, missing expected jobs"
 
-    def preprocess_data(self, data: Dict[str, Any]) -> Dict[str, Any]:
+    def preprocess_data(self, data: dict[str, Any]) -> dict[str, Any]:
         contexts = data["commits"]["nodes"][0]["commit"]["statusCheckRollup"]["contexts"]["nodes"]
         for context in contexts:
             if "context" in context and context["context"] == "tvm-ci/pr-head":
@@ -201,7 +201,7 @@ class TestNoReview(_TvmBotTest):
     USER = "abc"
     EXPECTED = "Cannot merge, did not find any approving reviews from users with write access"
 
-    def preprocess_data(self, data: Dict[str, Any]) -> Dict[str, Any]:
+    def preprocess_data(self, data: dict[str, Any]) -> dict[str, Any]:
         data["reviews"]["nodes"] = []
         return data
 
@@ -215,7 +215,7 @@ class TestChangesRequested(_TvmBotTest):
     USER = "abc"
     EXPECTED = "Cannot merge, found [this review]"
 
-    def preprocess_data(self, data: Dict[str, Any]) -> Dict[str, Any]:
+    def preprocess_data(self, data: dict[str, Any]) -> dict[str, Any]:
         data["reviews"]["nodes"][0]["state"] = "CHANGES_REQUESTED"
         data["reviews"]["nodes"][0]["url"] = "http://example.com"
         return data
@@ -230,7 +230,7 @@ class TestCoAuthors(_TvmBotTest):
     USER = "abc"
     EXPECTED = "Co-authored-by: Some One <someone@email.com>"
 
-    def preprocess_data(self, data: Dict[str, Any]) -> Dict[str, Any]:
+    def preprocess_data(self, data: dict[str, Any]) -> dict[str, Any]:
         data["authorCommits"]["nodes"][0]["commit"]["authors"]["nodes"].append(
             {"name": "Some One", "email": "someone@email.com"}
         )
diff --git a/tests/python/ci/test_utils.py b/tests/python/ci/test_utils.py
index 63143f773419..7354c7a50b80 100644
--- a/tests/python/ci/test_utils.py
+++ b/tests/python/ci/test_utils.py
@@ -21,7 +21,7 @@
 
 import pathlib
 import subprocess
-from typing import Any, List
+from typing import Any
 
 REPO_ROOT = pathlib.Path(__file__).resolve().parent.parent.parent.parent
 GITHUB_SCRIPT_ROOT = REPO_ROOT / "ci" / "scripts" / "github"
@@ -54,7 +54,7 @@ def run(self, *args, **kwargs):
         return proc
 
 
-def run_script(command: List[Any], check: bool = True, **kwargs):
+def run_script(command: list[Any], check: bool = True, **kwargs):
     """
     Wrapper to run a script and print its output if there was an error
     """
diff --git a/tests/python/codegen/test_target_codegen_cuda_fp8.py b/tests/python/codegen/test_target_codegen_cuda_fp8.py
index 56bcab61a400..5b3dec2ad35c 100644
--- a/tests/python/codegen/test_target_codegen_cuda_fp8.py
+++ b/tests/python/codegen/test_target_codegen_cuda_fp8.py
@@ -17,7 +17,6 @@
 # ruff: noqa: F401, F821, F841
 
 from itertools import product
-from typing import List, Tuple
 
 import numpy as np
 import pytest
@@ -418,7 +417,7 @@ def create_dequantize_func(
     @classmethod
     def quantize_fp8x4_e4m3(  # pylint: disable=too-many-locals
         cls,
-        weight_shape: List[tir.PrimExpr],
+        weight_shape: list[tir.PrimExpr],
         model_dtype,
         quantize_dtype,
         storage_dtype,
@@ -427,7 +426,7 @@ def quantize_fp8x4_e4m3(  # pylint: disable=too-many-locals
         max_int_value,
         axis: int = -1,
         output_transpose: bool = False,
-    ) -> Tuple[te.Tensor, te.Tensor]:
+    ) -> tuple[te.Tensor, te.Tensor]:
         """Group quantization for weight tensor, defined in tensor expression."""
         max_int = tir.const(max_int_value, model_dtype)
         shape = weight_shape  # pylint: disable=invalid-name
@@ -505,7 +504,7 @@ def compute_transpose(quantized_weight: te.Tensor, scale: te.Tensor):
     @classmethod
     def dequantize_fp8x4_e4m3(  # pylint: disable=too-many-locals
         cls,
-        packed_weight_shape: List[tir.PrimExpr],
+        packed_weight_shape: list[tir.PrimExpr],
         scale_shape,
         dequant_shape,
         model_dtype,
@@ -514,7 +513,7 @@ def dequantize_fp8x4_e4m3(  # pylint: disable=too-many-locals
         group_size,
         num_elem_per_storage,
         axis: int = -1,
-    ) -> Tuple[te.Tensor, te.Tensor]:
+    ) -> tuple[te.Tensor, te.Tensor]:
         """Group quantization for weight tensor, defined in tensor expression."""
         axis = axis if axis >= 0 else len(shape) + axis
 
diff --git a/tests/python/contrib/test_cutlass_gemm.py b/tests/python/contrib/test_cutlass_gemm.py
index 951085e8530c..327353cca288 100644
--- a/tests/python/contrib/test_cutlass_gemm.py
+++ b/tests/python/contrib/test_cutlass_gemm.py
@@ -14,7 +14,6 @@
 # KIND, either express or implied.  See the License for the
 # specific language governing permissions and limitations
 # under the License.
-from typing import Tuple
 
 import ml_dtypes
 import numpy as np
@@ -134,7 +133,7 @@ def test_group_gemm_sm100():
     )
 
 
-def rowwise_quant_fp8_e4m3(shape: Tuple[int, int], block_size: Tuple[int, int], dtype: str):
+def rowwise_quant_fp8_e4m3(shape: tuple[int, int], block_size: tuple[int, int], dtype: str):
     x_full_np = (np.random.rand(*shape) * 2 - 1).astype(dtype)
     x_scale_shape = (
         *shape[:-1],
@@ -172,7 +171,7 @@ def rowwise_quant_fp8_e4m3(shape: Tuple[int, int], block_size: Tuple[int, int],
     return x_np, x_scale_np
 
 
-def blockwise_quant_fp8_e4m3(shape: Tuple[int, int], block_size: Tuple[int, int], dtype: str):
+def blockwise_quant_fp8_e4m3(shape: tuple[int, int], block_size: tuple[int, int], dtype: str):
     w_full_np = (np.random.rand(*shape) * 2 - 1).astype(dtype)
     w_scale_shape = (
         *shape[:-2],
@@ -247,7 +246,7 @@ def blockwise_matmul(
     x_scale_np: np.ndarray,
     w_np: np.ndarray,
     w_scale_np: np.ndarray,
-    block_size: Tuple[int, int],
+    block_size: tuple[int, int],
     dtype: str,
 ):
     o_np = np.zeros((x_fp8_np.shape[0], w_np.shape[0]), dtype=dtype)
@@ -274,7 +273,7 @@ def blockwise_bmm(
     x_scale_np: np.ndarray,
     w_np: np.ndarray,
     w_scale_np: np.ndarray,
-    block_size: Tuple[int, int],
+    block_size: tuple[int, int],
     dtype: str,
 ):
     o_np = np.zeros((x_fp8_np.shape[0], x_fp8_np.shape[1], w_np.shape[1]), dtype=dtype)
diff --git a/tests/python/contrib/test_hexagon/benchmark_util.py b/tests/python/contrib/test_hexagon/benchmark_util.py
index 26eea5d8a156..c9be25efd3c4 100644
--- a/tests/python/contrib/test_hexagon/benchmark_util.py
+++ b/tests/python/contrib/test_hexagon/benchmark_util.py
@@ -201,7 +201,7 @@ def get_benchmark_id(keys_dict):
     # Sniff for shape-like lists, because we want them in a form that's both
     # readable and filesystem-friendly...
     for k, v in keys_dict_copy.items():
-        if isinstance(v, (list, tuple)):
+        if isinstance(v, list | tuple):
             v_str = "_".join([str(x) for x in v])
             keys_dict_copy[k] = v_str
 
diff --git a/tests/python/contrib/test_hexagon/pytest_util.py b/tests/python/contrib/test_hexagon/pytest_util.py
index d3ff86a79d24..1c7b92a529b1 100644
--- a/tests/python/contrib/test_hexagon/pytest_util.py
+++ b/tests/python/contrib/test_hexagon/pytest_util.py
@@ -18,12 +18,11 @@
 """Hexagon pytest utility functions"""
 
 import collections
-from typing import List, Optional, Union
 
 import numpy as np
 
 
-def get_test_id(*test_params, test_param_descs: Optional[List[Optional[str]]] = None) -> str:
+def get_test_id(*test_params, test_param_descs: list[str | None] | None = None) -> str:
     """
     An opinionated alternative to pytest's default algorithm for generating a
     test's ID string.  Intended to make it easier for human readers to
@@ -47,7 +46,7 @@ def get_test_id(*test_params, test_param_descs: Optional[List[Optional[str]]] =
     else:
         assert len(test_param_descs) == len(test_params)
 
-    def get_single_param_chunk(param_val, param_desc: Optional[str]):
+    def get_single_param_chunk(param_val, param_desc: str | None):
         if isinstance(param_val, list):
             # Like str(list), but avoid the whitespace padding.
             val_str = "[" + ",".join(str(x) for x in param_val) + "]"
@@ -99,8 +98,8 @@ def get_single_param_chunk(param_val, param_desc: Optional[str]):
 
 
 def get_multitest_ids(
-    multitest_params_list: List[List], param_descs: Optional[List[Optional[str]]]
-) -> List[str]:
+    multitest_params_list: list[list], param_descs: list[str | None] | None
+) -> list[str]:
     """
     A convenience function for classes that use both 'tvm.testing.parameters' and 'get_test_id'.
 
@@ -114,7 +113,7 @@ def get_multitest_ids(
     ]
 
 
-def get_numpy_dtype_info(dtype) -> Union[np.finfo, np.iinfo]:
+def get_numpy_dtype_info(dtype) -> np.finfo | np.iinfo:
     """
     Return an appropriate 'np.iinfo' or 'np.finfo' object corresponding to
     the specified Numpy dtype.
@@ -142,7 +141,7 @@ def get_numpy_dtype_info(dtype) -> Union[np.finfo, np.iinfo]:
 
 
 def create_populated_numpy_tensor(
-    input_shape: Union[list, tuple], dtype: str, input_tensor_populator
+    input_shape: list | tuple, dtype: str, input_tensor_populator
 ) -> np.ndarray:
     """
     Create a numpy tensor with the specified shape, dtype, and content.
diff --git a/tests/python/contrib/test_hexagon/test_2d_physical_buffers.py b/tests/python/contrib/test_hexagon/test_2d_physical_buffers.py
index 8b1202847f4b..13b3b07354fc 100644
--- a/tests/python/contrib/test_hexagon/test_2d_physical_buffers.py
+++ b/tests/python/contrib/test_hexagon/test_2d_physical_buffers.py
@@ -171,7 +171,7 @@ def extract_buffers(stmt):
     buffers = []
 
     def visitor(node):
-        if isinstance(node, (tvm.tir.BufferLoad, tvm.tir.BufferStore)):
+        if isinstance(node, tvm.tir.BufferLoad | tvm.tir.BufferStore):
             buffers.append(node.buffer)
 
     post_order_visit(stmt, visitor)
diff --git a/tests/python/contrib/test_hexagon/test_benchmark_maxpool2d.py b/tests/python/contrib/test_hexagon/test_benchmark_maxpool2d.py
index ba53ba7af2a2..374f47300827 100644
--- a/tests/python/contrib/test_hexagon/test_benchmark_maxpool2d.py
+++ b/tests/python/contrib/test_hexagon/test_benchmark_maxpool2d.py
@@ -45,7 +45,6 @@
 
 import copy
 import os
-from typing import List
 
 import numpy as np
 import pytest
@@ -83,7 +82,7 @@ def _int8_nhwc_8h8w32c_map(n_batch, height, width, channel):
     ]
 
 
-def _int8_nhwc_8h8w32c_shape(n_batch, height, width, channel) -> List[int]:
+def _int8_nhwc_8h8w32c_shape(n_batch, height, width, channel) -> list[int]:
     return [
         n_batch,
         _ceil_div(height, 8),
diff --git a/tests/python/contrib/test_msc/test_translate_relax.py b/tests/python/contrib/test_msc/test_translate_relax.py
index f763865f6d5f..2b616eb33adb 100644
--- a/tests/python/contrib/test_msc/test_translate_relax.py
+++ b/tests/python/contrib/test_msc/test_translate_relax.py
@@ -42,7 +42,7 @@ def _tvm_runtime_to_np(obj):
             return obj.numpy()
         elif isinstance(obj, tvm.runtime.ShapeTuple):
             return np.array(obj, dtype="int64")
-        elif isinstance(obj, (list, tvm.ir.container.Array)):
+        elif isinstance(obj, list | tvm.ir.container.Array):
             return [_tvm_runtime_to_np(item) for item in obj]
         elif isinstance(obj, tuple):
             return tuple(_tvm_runtime_to_np(item) for item in obj)
@@ -63,9 +63,9 @@ def _run_relax(relax_mod):
 
     orig_output = _run_relax(orig_mod)
     rt_output = _run_relax(rt_mod)
-    if not isinstance(orig_output, (list, tuple)):
+    if not isinstance(orig_output, list | tuple):
         orig_output = [orig_output]
-    if not isinstance(rt_output, (list, tuple)):
+    if not isinstance(rt_output, list | tuple):
         rt_output = [rt_output]
     for o_out, r_out in zip(orig_output, rt_output):
         tvm.testing.assert_allclose(o_out, r_out)
diff --git a/tests/python/contrib/test_msc/test_translate_tensorrt.py b/tests/python/contrib/test_msc/test_translate_tensorrt.py
index bd9a6127955f..422411fa3576 100644
--- a/tests/python/contrib/test_msc/test_translate_tensorrt.py
+++ b/tests/python/contrib/test_msc/test_translate_tensorrt.py
@@ -94,7 +94,7 @@ def verify_model(torch_model, input_info, **trans_config):
     with torch.no_grad():
         golden = torch_model(*torch_datas)
         mod = from_fx(graph_model, input_info)
-    if not isinstance(golden, (list, tuple)):
+    if not isinstance(golden, list | tuple):
         golden = [golden]
     golden = [g.detach().cpu().numpy() for g in golden]
     # partition module for tensorrt
diff --git a/tests/python/contrib/test_msc/test_translate_torch.py b/tests/python/contrib/test_msc/test_translate_torch.py
index 826ee4a8d289..584ca414fbf7 100644
--- a/tests/python/contrib/test_msc/test_translate_torch.py
+++ b/tests/python/contrib/test_msc/test_translate_torch.py
@@ -40,9 +40,9 @@ def verify_model(torch_model, input_info):
             result = model()
         else:
             result = model(*torch_datas)
-    if not isinstance(golden, (list, tuple)):
+    if not isinstance(golden, list | tuple):
         golden = [golden]
-    if not isinstance(result, (list, tuple)):
+    if not isinstance(result, list | tuple):
         result = [result]
     assert len(golden) == len(result), f"golden {len(golden)} mismatch with result {len(result)}"
     for gol_r, new_r in zip(golden, result):
diff --git a/tests/python/disco/test_nvshmem.py b/tests/python/disco/test_nvshmem.py
index 7aa72fbef9d0..df285c9c65c3 100644
--- a/tests/python/disco/test_nvshmem.py
+++ b/tests/python/disco/test_nvshmem.py
@@ -24,8 +24,9 @@
 import sys
 import tempfile
 import threading
+from collections.abc import Callable
 from multiprocessing import Process
-from typing import Any, Callable, List
+from typing import Any
 
 import numpy as np
 import pytest
diff --git a/tests/python/ir/analysis/test_collect_call_map.py b/tests/python/ir/analysis/test_collect_call_map.py
index f1dcd3b23537..8990aea7f78d 100644
--- a/tests/python/ir/analysis/test_collect_call_map.py
+++ b/tests/python/ir/analysis/test_collect_call_map.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 
-from typing import Dict, List
 
 import tvm
 import tvm.testing
@@ -26,7 +25,7 @@
 from tvm.script import tir as T
 
 
-def _build_str_map(call_map: Dict[GlobalVar, List[GlobalVar]]) -> Dict[str, List[str]]:
+def _build_str_map(call_map: dict[GlobalVar, list[GlobalVar]]) -> dict[str, list[str]]:
     return {
         caller.name_hint: [callee.name_hint for callee in callees]
         for caller, callees in call_map.items()
diff --git a/tests/python/nightly/test_nnapi/infrastructure.py b/tests/python/nightly/test_nnapi/infrastructure.py
index ac3ddcd84e94..917af437b44f 100644
--- a/tests/python/nightly/test_nnapi/infrastructure.py
+++ b/tests/python/nightly/test_nnapi/infrastructure.py
@@ -26,8 +26,6 @@
 
 # pylint: disable=import-outside-toplevel,missing-function-docstring
 def reshape_matmul(mod: tvm.IRModule):
-    from typing import Dict
-
     from tvm.relax import Expr
     from tvm.relax.dpl import DFPattern, rewrite_call
     from tvm.relax.dpl.pattern import is_op, wildcard
@@ -36,7 +34,7 @@ def reshape_matmul(mod: tvm.IRModule):
     input1 = wildcard()
     pattern = is_op("relax.matmul")(input0, input1)
 
-    def _rewriter(expr: Expr, matches: Dict[DFPattern, Expr]):
+    def _rewriter(expr: Expr, matches: dict[DFPattern, Expr]):
         i0 = matches[input0]
         i1 = matches[input1]
         if len(i0.struct_info.shape) == 2 and len(i1.struct_info.shape) == 2:
@@ -51,8 +49,6 @@ def _rewriter(expr: Expr, matches: Dict[DFPattern, Expr]):
 
 
 def decompose_clip(mod: tvm.IRModule) -> tvm.IRModule:
-    from typing import Dict
-
     from tvm.relax import Expr
     from tvm.relax.dpl import DFPattern, rewrite_call
     from tvm.relax.dpl.pattern import is_op, wildcard
@@ -62,7 +58,7 @@ def decompose_clip(mod: tvm.IRModule) -> tvm.IRModule:
     max_pattern = wildcard()
     pattern = is_op("relax.clip")(input_pattern, min_pattern, max_pattern)
 
-    def _rewriter(expr: Expr, matches: Dict[DFPattern, Expr]) -> Expr:  # pylint: disable=unused-argument
+    def _rewriter(expr: Expr, matches: dict[DFPattern, Expr]) -> Expr:  # pylint: disable=unused-argument
         dtype = matches[input_pattern].struct_info.dtype
         return R.minimum(
             R.maximum(
diff --git a/tests/python/nightly/test_nnapi/test_network.py b/tests/python/nightly/test_nnapi/test_network.py
index 80d820a29ca8..dcee4e161b5c 100644
--- a/tests/python/nightly/test_nnapi/test_network.py
+++ b/tests/python/nightly/test_nnapi/test_network.py
@@ -16,8 +16,6 @@
 # under the License.
 """NNAPI network tests."""
 
-from typing import List
-
 import numpy as np
 import onnx
 import pytest
@@ -125,7 +123,7 @@ def test_network(name, dtype):
     for _name, (shape, _dtype) in inputs.items():
         input_data[_name] = np.random.uniform(-1.0, 1.0, shape).astype(_dtype)
 
-    inputs_tvm: List[tvm.runtime.Tensor] = [tvm.runtime.tensor(v) for k, v in input_data.items()]
+    inputs_tvm: list[tvm.runtime.Tensor] = [tvm.runtime.tensor(v) for k, v in input_data.items()]
     outputs = _build_and_run_network(remote_obj, tracker, mod, inputs_tvm)
     nnapi_out = outputs[0]
     expected_out = outputs[1]
diff --git a/tests/python/nightly/test_nnapi/test_ops.py b/tests/python/nightly/test_nnapi/test_ops.py
index 3944b0001224..21453c360ee4 100644
--- a/tests/python/nightly/test_nnapi/test_ops.py
+++ b/tests/python/nightly/test_nnapi/test_ops.py
@@ -17,8 +17,6 @@
 # ruff: noqa: F841
 """NNAPI integration operator tests."""
 
-from typing import List
-
 import numpy as np
 import pytest
 
@@ -352,7 +350,7 @@ def main(
 
 
 def verify(remote_obj, tracker, mod, inputs):
-    inputs_tvm: List[tvm.runtime.Tensor] = [tvm.runtime.tensor(v) for v in inputs]
+    inputs_tvm: list[tvm.runtime.Tensor] = [tvm.runtime.tensor(v) for v in inputs]
     outputs = _build_and_run_network(remote_obj, tracker, mod, inputs_tvm)
     nnapi_out = outputs[0]
     expected_out = outputs[1]
diff --git a/tests/python/relax/distributed/test_distributed_tvmscript_parser.py b/tests/python/relax/distributed/test_distributed_tvmscript_parser.py
index 7deef9c7d36b..c9c2c50bd85b 100644
--- a/tests/python/relax/distributed/test_distributed_tvmscript_parser.py
+++ b/tests/python/relax/distributed/test_distributed_tvmscript_parser.py
@@ -33,8 +33,8 @@
 
 
 def _check(
-    parsed: Union[relax.Function, IRModule],
-    expect: Optional[Union[relax.Function, IRModule]] = None,
+    parsed: relax.Function | IRModule,
+    expect: relax.Function | IRModule | None = None,
 ):
     test = parsed.script(show_meta=True)
     roundtrip_mod = tvm.script.from_source(test)
diff --git a/tests/python/relax/nvshmem/test_runtime_builtin_kv_cache_transfer.py b/tests/python/relax/nvshmem/test_runtime_builtin_kv_cache_transfer.py
index 2bf58576a840..b9d9e76d062b 100644
--- a/tests/python/relax/nvshmem/test_runtime_builtin_kv_cache_transfer.py
+++ b/tests/python/relax/nvshmem/test_runtime_builtin_kv_cache_transfer.py
@@ -17,7 +17,7 @@
 # ruff: noqa: E501, E741, F401
 import enum
 import itertools
-from typing import Dict, List, Optional, Tuple, Union
+from typing import Optional, Union
 
 import numpy as np
 import pytest
@@ -277,7 +277,7 @@ def verify_cached_kv(kv_cache, seq_ids, expected_k, expected_v):
         )
 
 
-def f_apply_rotary(x, offset, scale, theta, offset_list: Optional[List[int]] = None):
+def f_apply_rotary(x, offset, scale, theta, offset_list: list[int] | None = None):
     # x: (N, H, D)
     assert len(x.shape) == 3
     nfeat = x.shape[-1]
@@ -307,13 +307,13 @@ def f_apply_rotary(x, offset, scale, theta, offset_list: Optional[List[int]] = N
 def apply_attention(
     kv_cache,
     rope_mode: RopeMode,
-    batch: List[Tuple[Union[int, Tuple[int, int, int]], int]],
-    cached_k: Dict[int, torch.Tensor],
-    cached_v: Dict[int, torch.Tensor],
-    sliding_window_sizes: Optional[List[int]] = None,
-    attn_sink_sizes: Optional[List[int]] = None,
-    token_tree_parent_ptr_list: Optional[List[List[int]]] = None,
-    accepted_leaf_indices: Optional[List[int]] = None,
+    batch: list[tuple[int | tuple[int, int, int], int]],
+    cached_k: dict[int, torch.Tensor],
+    cached_v: dict[int, torch.Tensor],
+    sliding_window_sizes: list[int] | None = None,
+    attn_sink_sizes: list[int] | None = None,
+    token_tree_parent_ptr_list: list[list[int]] | None = None,
+    accepted_leaf_indices: list[int] | None = None,
     only_update_host=False,
     skip_add_sequence=False,
 ) -> None:
@@ -349,7 +349,7 @@ def apply_attention(
             )
 
     flattened_token_tree_parent_ptr = None
-    token_tree_node_depths_list: List[Optional[List[int]]] = [None for _ in batch]
+    token_tree_node_depths_list: list[list[int] | None] = [None for _ in batch]
     if token_tree_parent_ptr_list:
         assert len(token_tree_node_depths_list) == len(seq_ids)
         if accepted_leaf_indices is not None:
diff --git a/tests/python/relax/test_analysis.py b/tests/python/relax/test_analysis.py
index f9747b284472..7d8cfd093d91 100644
--- a/tests/python/relax/test_analysis.py
+++ b/tests/python/relax/test_analysis.py
@@ -16,7 +16,6 @@
 # under the License.
 # ruff: noqa: F811, F841
 
-from typing import List, Set, Union
 
 import pytest
 
@@ -40,7 +39,7 @@
 from tvm.script import tir as T
 
 
-def var_name_set(vars: List[Union[rx.Var, rx.GlobalVar]]) -> Set[str]:
+def var_name_set(vars: list[rx.Var | rx.GlobalVar]) -> set[str]:
     return set(map(lambda v: v.name_hint, vars))
 
 
diff --git a/tests/python/relax/test_analysis_computable_at_compile_time.py b/tests/python/relax/test_analysis_computable_at_compile_time.py
index d1160642b117..fd592a989450 100644
--- a/tests/python/relax/test_analysis_computable_at_compile_time.py
+++ b/tests/python/relax/test_analysis_computable_at_compile_time.py
@@ -16,7 +16,6 @@
 # under the License.
 # ruff: noqa: F841
 
-from typing import List
 
 import tvm
 import tvm.testing
@@ -24,7 +23,7 @@
 from tvm.script import tir as T
 
 
-def _analyze_func(func: tvm.relax.Function) -> List[str]:
+def _analyze_func(func: tvm.relax.Function) -> list[str]:
     return [var.name_hint for var in tvm.relax.analysis.computable_at_compile_time(func)]
 
 
diff --git a/tests/python/relax/test_analysis_detect_recursion.py b/tests/python/relax/test_analysis_detect_recursion.py
index c2a82e8c5d08..7f5b38d54b41 100644
--- a/tests/python/relax/test_analysis_detect_recursion.py
+++ b/tests/python/relax/test_analysis_detect_recursion.py
@@ -14,7 +14,6 @@
 # KIND, either express or implied.  See the License for the
 # specific language governing permissions and limitations
 # under the License.
-from typing import List
 
 import tvm
 import tvm.testing
@@ -24,7 +23,7 @@
 from tvm.script import tir as T
 
 
-def assert_groups(groups: List[List[rx.GlobalVar]], expected: List[List[str]]) -> None:
+def assert_groups(groups: list[list[rx.GlobalVar]], expected: list[list[str]]) -> None:
     assert len(groups) == len(expected)
 
     # disregard order, search only by name for convenience
diff --git a/tests/python/relax/test_ast_printer.py b/tests/python/relax/test_ast_printer.py
index 7c943e5d3951..bba5c6aa402b 100644
--- a/tests/python/relax/test_ast_printer.py
+++ b/tests/python/relax/test_ast_printer.py
@@ -17,7 +17,6 @@
 # ruff: noqa: F811, F841
 import re
 from functools import partial
-from typing import Dict
 
 import numpy as np
 
@@ -59,7 +58,7 @@ class DefaultMutator(rx.PyExprMutator):
     return mod["main"]
 
 
-def assert_fields(nodename: str, fields: Dict[str, str], target: str) -> None:
+def assert_fields(nodename: str, fields: dict[str, str], target: str) -> None:
     """
     Given a target string, ensure that the string defines the specified node
     and that the given mappings of fields to values are present in the string.
diff --git a/tests/python/relax/test_dataflow_inplace.py b/tests/python/relax/test_dataflow_inplace.py
index 7fcc0e48e798..57b178680ed0 100644
--- a/tests/python/relax/test_dataflow_inplace.py
+++ b/tests/python/relax/test_dataflow_inplace.py
@@ -16,7 +16,6 @@
 # under the License.
 # ruff: noqa: F841
 
-from typing import List, Set, Tuple
 
 import numpy as np
 
@@ -347,7 +346,7 @@ def main(x: R.Tensor((2, 3), "int32"), y: R.Tensor((2, 3), "int32")) -> R.Tensor
 
     # order does not matter for the listing of candidates, so we have to implement as sets
     def assert_candidate_list(
-        actual: List[Tuple[int, Set[int]]], expected: List[Tuple[int, Set[int]]]
+        actual: list[tuple[int, set[int]]], expected: list[tuple[int, set[int]]]
     ) -> None:
         assert len(actual) == len(expected)
         for i in range(len(actual)):
diff --git a/tests/python/relax/test_expr_args_converter.py b/tests/python/relax/test_expr_args_converter.py
index 5e8cfd434d55..d156245452ec 100644
--- a/tests/python/relax/test_expr_args_converter.py
+++ b/tests/python/relax/test_expr_args_converter.py
@@ -16,7 +16,8 @@
 # under the License.
 # ruff: noqa: E731
 
-from typing import Any, Callable, List, Optional, Union
+from collections.abc import Callable
+from typing import Any
 
 import pytest
 
@@ -43,20 +44,18 @@ def _test_expr(arg: Expr, *args: Expr, **kwargs: Expr) -> None:
     _test_base(f_checker, arg, *args, **kwargs)
 
 
-def _test_optional_expr(
-    arg: Optional[Expr], *args: Optional[Expr], **kwargs: Optional[Expr]
-) -> None:
+def _test_optional_expr(arg: Expr | None, *args: Expr | None, **kwargs: Expr | None) -> None:
     f_checker = lambda x: x is None or isinstance(x, Expr)
     _test_base(f_checker, arg, *args, **kwargs)
 
 
-def _test_list_expr(arg: List[Expr], *args: List[Expr], **kwargs: List[Expr]) -> None:
+def _test_list_expr(arg: list[Expr], *args: list[Expr], **kwargs: list[Expr]) -> None:
     f_checker = lambda x: isinstance(x, list) and all([isinstance(arg, Expr) for arg in x])
     _test_base(f_checker, arg, *args, **kwargs)
 
 
 def _test_optional_list_expr(
-    arg: Optional[List[Expr]], *args: Optional[List[Expr]], **kwargs: Optional[List[Expr]]
+    arg: list[Expr] | None, *args: list[Expr] | None, **kwargs: list[Expr] | None
 ) -> None:
     f_checker = lambda x: (
         x is None or (isinstance(x, list) and all([isinstance(arg, Expr) for arg in x]))
@@ -128,7 +127,7 @@ def test_auto_convert():
 
 
 def test_auto_convert_skip():
-    def _test_expr_skip(arg: int, *args: Union[str, Expr], **kwargs: List[Optional[Expr]]) -> None:
+    def _test_expr_skip(arg: int, *args: str | Expr, **kwargs: list[Expr | None]) -> None:
         f_checker = lambda x: not isinstance(x, Expr)
         _test_base(f_checker, arg, *args, **kwargs)
 
diff --git a/tests/python/relax/test_frontend_nn_jit.py b/tests/python/relax/test_frontend_nn_jit.py
index 3c536dbc0230..31328104c7a1 100644
--- a/tests/python/relax/test_frontend_nn_jit.py
+++ b/tests/python/relax/test_frontend_nn_jit.py
@@ -14,7 +14,6 @@
 # KIND, either express or implied.  See the License for the
 # specific language governing permissions and limitations
 # under the License.
-from typing import List, Tuple
 
 import pytest
 import torch
@@ -108,7 +107,7 @@ class Layer(nn.Module):
         def __init__(self):
             pass
 
-        def forward(self, x: Tuple[nn.Tensor, nn.Tensor]):
+        def forward(self, x: tuple[nn.Tensor, nn.Tensor]):
             assert isinstance(x, tuple)
             x0 = x[0]
             x1 = x[1]
@@ -143,7 +142,7 @@ class Layer(nn.Module):
         def __init__(self):
             pass
 
-        def forward(self, x: List[nn.Tensor]):
+        def forward(self, x: list[nn.Tensor]):
             assert isinstance(x, list)
             x0 = x[0]
             x1 = x[1]
@@ -178,7 +177,7 @@ class Layer(nn.Module):
         def __init__(self):
             pass
 
-        def forward(self, x: Tuple[nn.Tensor, nn.Tensor, int]):
+        def forward(self, x: tuple[nn.Tensor, nn.Tensor, int]):
             x0 = x[0]
             x1 = x[1]
             y0 = nn.add(x0, x1)
diff --git a/tests/python/relax/test_frontend_nn_modules.py b/tests/python/relax/test_frontend_nn_modules.py
index 80073dcaa6cb..3fb81ff08c0c 100644
--- a/tests/python/relax/test_frontend_nn_modules.py
+++ b/tests/python/relax/test_frontend_nn_modules.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 # ruff: noqa: E501, F401, F841
-from typing import List, Tuple
 
 import numpy as np
 import pytest
@@ -651,7 +650,7 @@ class Layer(nn.Module):
         def __init__(self):
             pass
 
-        def forward(self, x: Tuple[nn.Tensor, nn.Tensor]):
+        def forward(self, x: tuple[nn.Tensor, nn.Tensor]):
             x0 = x[0]
             x1 = x[1]
             y0 = nn.add(x0, x1)
@@ -690,7 +689,7 @@ class Layer(nn.Module):
         def __init__(self):
             pass
 
-        def forward(self, x: List[nn.Tensor]):
+        def forward(self, x: list[nn.Tensor]):
             x0 = x[0]
             x1 = x[1]
             y0 = nn.add(x0, x1)
diff --git a/tests/python/relax/test_frontend_onnx.py b/tests/python/relax/test_frontend_onnx.py
index dff9a36aa746..2d7f57224fd9 100644
--- a/tests/python/relax/test_frontend_onnx.py
+++ b/tests/python/relax/test_frontend_onnx.py
@@ -22,7 +22,7 @@
 This file is a test script to test Relax ONNX frontend coverage.
 """
 
-from typing import Dict, List, Literal, Optional
+from typing import Literal
 
 import numpy as np
 import onnx
@@ -43,8 +43,8 @@
 
 
 def generate_random_inputs(
-    model: ModelProto, inputs: Optional[Dict[str, np.ndarray]] = None
-) -> Dict[str, np.ndarray]:
+    model: ModelProto, inputs: dict[str, np.ndarray] | None = None
+) -> dict[str, np.ndarray]:
     input_values = {}
     # Iterate through model inputs and extract their shape.
     for i in model.graph.input:
@@ -83,7 +83,7 @@ def generate_random_value(shape, elem_type) -> np.ndarray:
 
 def check_correctness(
     model: ModelProto,
-    inputs: Optional[Dict[str, np.ndarray]] = None,
+    inputs: dict[str, np.ndarray] | None = None,
     ir_version: int = 8,
     opset: int = 14,
     rtol: float = 1e-7,
@@ -169,7 +169,7 @@ def _get_numpy_subdtype(narray):
             return "other"
 
     def _check_output(tvm_out, ort_out):
-        if isinstance(tvm_out, tuple) and isinstance(ort_out, (tvm.runtime.ShapeTuple, list)):
+        if isinstance(tvm_out, tuple) and isinstance(ort_out, tvm.runtime.ShapeTuple | list):
             assert len(tvm_out) == len(ort_out), "Unequal number of outputs"
             for tvm_out_i, ort_out_i in zip(tvm_out, ort_out):
                 _check_output(tvm_out_i, ort_out_i)
@@ -182,7 +182,7 @@ def _check_output(tvm_out, ort_out):
             if check_dtypes:
                 assert _get_numpy_subdtype(shape_out.numpy()) == _get_numpy_subdtype(ort_out)
             tvm.testing.assert_allclose(shape_out.numpy(), ort_out, rtol=rtol, atol=atol)
-        elif isinstance(tvm_out, (int, float, bool)) and isinstance(ort_out, np.ndarray):
+        elif isinstance(tvm_out, int | float | bool) and isinstance(ort_out, np.ndarray):
             if check_dtypes:
                 assert _get_numpy_subdtype(np.array(tvm_out)) == _get_numpy_subdtype(ort_out)
             tvm.testing.assert_allclose(np.array(tvm_out), ort_out, rtol=rtol, atol=atol)
@@ -567,7 +567,7 @@ def test_gather():
     def _verify_gather(data_shape, indices, out_shape, axis=0):
         gather_node = helper.make_node("Gather", ["data", "indices"], ["y"], axis=axis)
 
-        if isinstance(indices, (list, tuple)):
+        if isinstance(indices, list | tuple):
             indices_shape = np.asarray(indices).shape
         else:
             indices_shape = []
@@ -717,9 +717,9 @@ def verify_scatter_nd(data_shape, indices_shape, updates_shape):
 @pytest.mark.parametrize("condition_shape", [None, [8], [16]])
 @pytest.mark.parametrize("axis", [None, 0, 1])
 def test_compress(
-    tensor_shape: List[int],
-    condition_shape: Optional[List[int]],
-    axis: Optional[int],
+    tensor_shape: list[int],
+    condition_shape: list[int] | None,
+    axis: int | None,
 ):
     if condition_shape is None and axis is None:
         pytest.skip("Either condition_shape or axis must be specified")
@@ -2908,11 +2908,11 @@ def test_batch_norm():
 )
 def test_pool(
     pool_name: str,
-    shape: List[int],
+    shape: list[int],
     auto_pad: str,
-    kernel_shape: List[int],
-    strides: List[int],
-    pads: List[int],
+    kernel_shape: list[int],
+    strides: list[int],
+    pads: list[int],
 ):
     verify_unary(
         pool_name,
@@ -3015,7 +3015,7 @@ def test_onehot():
 @pytest.mark.parametrize("axis", [None, 0, 1, -1])
 @pytest.mark.parametrize("sorted", [0, 1])
 @pytest.mark.parametrize("num_outputs", [1, 2, 3, 4])
-def test_unique(axis: Optional[int], sorted: int, num_outputs: int):
+def test_unique(axis: int | None, sorted: int, num_outputs: int):
     input_shape = [8, 8]
     if axis is None:
         output_shape = [-1]
@@ -3085,7 +3085,7 @@ def test_space_to_depth():
     check_correctness(model)
 
 
-def construct_sequence(input_shape: List[int], num_tensors: int, name: str = "sequence"):
+def construct_sequence(input_shape: list[int], num_tensors: int, name: str = "sequence"):
     inputs = [f"data{i}" for i in range(num_tensors)]
     sequence_construct_node = helper.make_node("SequenceConstruct", inputs, [name])
     graph_inputs = [
@@ -3095,7 +3095,7 @@ def construct_sequence(input_shape: List[int], num_tensors: int, name: str = "se
     return sequence_construct_node, graph_inputs
 
 
-def make_constant_node(name: str, data_type: int, dims: List[int], vals: List[int]):
+def make_constant_node(name: str, data_type: int, dims: list[int], vals: list[int]):
     return helper.make_node(
         "Constant",
         inputs=[],
@@ -3581,7 +3581,7 @@ def test_nms():
     vm.invoke_stateful("main")
     tvm_output = vm.get_outputs("main")
 
-    if isinstance(tvm_output, (list, tuple)):
+    if isinstance(tvm_output, list | tuple):
         tvm_selected = tvm_output[0].numpy()
     else:
         tvm_selected = tvm_output.numpy()
@@ -3745,7 +3745,7 @@ def test_nms_iou_suppression():
     tvm_output = vm.get_outputs("main")
 
     # Custom NMS output comparison
-    if isinstance(tvm_output, (list, tuple)):
+    if isinstance(tvm_output, list | tuple):
         tvm_selected = tvm_output[0].numpy()
     else:
         tvm_selected = tvm_output.numpy()
@@ -3840,7 +3840,7 @@ def test_nms_max_boxes_limit():
     tvm_output = vm.get_outputs("main")
 
     # Custom NMS output comparison
-    if isinstance(tvm_output, (list, tuple)):
+    if isinstance(tvm_output, list | tuple):
         tvm_selected = tvm_output[0].numpy()
     else:
         tvm_selected = tvm_output.numpy()
@@ -3932,7 +3932,7 @@ def test_nms_score_threshold():
     tvm_output = vm.get_outputs("main")
 
     # Custom NMS output comparison
-    if isinstance(tvm_output, (list, tuple)):
+    if isinstance(tvm_output, list | tuple):
         tvm_selected = tvm_output[0].numpy()
     else:
         tvm_selected = tvm_output.numpy()
diff --git a/tests/python/relax/test_frontend_stablehlo.py b/tests/python/relax/test_frontend_stablehlo.py
index 398997fd6382..2261b32a9c5e 100644
--- a/tests/python/relax/test_frontend_stablehlo.py
+++ b/tests/python/relax/test_frontend_stablehlo.py
@@ -19,7 +19,6 @@
 # pylint: disable=c-extension-no-member
 
 import functools
-from typing import List, Tuple, Union
 
 import numpy as np
 import pytest
@@ -34,8 +33,8 @@
 
 
 def generate_np_inputs(
-    input_shapes: Union[Tuple, List[Tuple]], dtype: str = "float32"
-) -> Union[np.ndarray, List[np.ndarray]]:
+    input_shapes: tuple | list[tuple], dtype: str = "float32"
+) -> np.ndarray | list[np.ndarray]:
     """Generate numpy data as the inputs of model
 
     Parameters
@@ -50,7 +49,7 @@ def generate_np_inputs(
     out: List[np.ndarray]
         numpy input data
     """
-    if not isinstance(input_shapes[0], (list, tuple)):
+    if not isinstance(input_shapes[0], list | tuple):
         return [np.random.uniform(size=input_shapes).astype(dtype)]
     out = []
     for input_shape in input_shapes:
@@ -58,7 +57,7 @@ def generate_np_inputs(
     return out
 
 
-def np2jnp(inputs_np: Union[np.ndarray, List[np.ndarray]]):
+def np2jnp(inputs_np: np.ndarray | list[np.ndarray]):
     """Convert data from numpy to jax.numpy
 
     Parameters
@@ -75,7 +74,7 @@ def np2jnp(inputs_np: Union[np.ndarray, List[np.ndarray]]):
 
     # Use jnp.asarray to avoid unnecessary memory copies
     inputs_jnp = []
-    if isinstance(inputs_np, (tuple, list)):
+    if isinstance(inputs_np, tuple | list):
         for input_np in inputs_np:
             inputs_jnp.append(jnp.asarray(input_np))
         return inputs_jnp
@@ -84,7 +83,7 @@ def np2jnp(inputs_np: Union[np.ndarray, List[np.ndarray]]):
 
 def check_correctness(
     jax_jit_mod,
-    input_shapes: Union[Tuple, List[Tuple]],
+    input_shapes: tuple | list[tuple],
     dtype: str = "float32",
 ) -> None:
     """Run a jax model and the translated TVM IRModule,
@@ -139,8 +138,8 @@ def check_correctness(
 
 
 def get_vm_res(
-    ir_mod: tvm.IRModule, weights: Union[np.ndarray, List[np.ndarray]]
-) -> Union[tvm.runtime.Tensor, List[tvm.runtime.Tensor]]:
+    ir_mod: tvm.IRModule, weights: np.ndarray | list[np.ndarray]
+) -> tvm.runtime.Tensor | list[tvm.runtime.Tensor]:
     """Compile and run an ir_module on Relax VM
 
     Parameters
diff --git a/tests/python/relax/test_op_binary.py b/tests/python/relax/test_op_binary.py
index 25d32249c503..90f71b7f388a 100644
--- a/tests/python/relax/test_op_binary.py
+++ b/tests/python/relax/test_op_binary.py
@@ -15,7 +15,7 @@
 # specific language governing permissions and limitations
 # under the License.
 # ruff: noqa: F841
-from typing import Callable
+from collections.abc import Callable
 
 import pytest
 
diff --git a/tests/python/relax/test_op_gradient_numeric.py b/tests/python/relax/test_op_gradient_numeric.py
index 4e9e8842ba83..b525e9a744fb 100644
--- a/tests/python/relax/test_op_gradient_numeric.py
+++ b/tests/python/relax/test_op_gradient_numeric.py
@@ -15,7 +15,8 @@
 # specific language governing permissions and limitations
 # under the License.
 # ruff: noqa: F401, F811, RUF005
-from typing import Callable, List, Tuple, Union
+from collections.abc import Callable
+from typing import Union
 
 import numpy as np
 import pytest
@@ -32,11 +33,11 @@
 
 def relax_check_gradients(
     op_func: Callable,
-    inputs_numpy: List[np.array],
-    target: Union[str, tvm.target.Target],
+    inputs_numpy: list[np.array],
+    target: str | tvm.target.Target,
     dev: tvm.runtime.Device,
     tuple_input: bool = False,
-    ignore_grads: List[int] = [],
+    ignore_grads: list[int] = [],
     **kwargs,  # attr for operators
 ):
     """Generate the forward and the gradient module. Then run them and check numeric gradients.
diff --git a/tests/python/relax/test_op_search.py b/tests/python/relax/test_op_search.py
index 3c7455ff0a6d..53d2c3fa683b 100644
--- a/tests/python/relax/test_op_search.py
+++ b/tests/python/relax/test_op_search.py
@@ -14,7 +14,7 @@
 # KIND, either express or implied.  See the License for the
 # specific language governing permissions and limitations
 # under the License.
-from typing import Callable
+from collections.abc import Callable
 
 import pytest
 
diff --git a/tests/python/relax/test_op_statistical.py b/tests/python/relax/test_op_statistical.py
index 54a5704c72fe..37793ecc73c7 100644
--- a/tests/python/relax/test_op_statistical.py
+++ b/tests/python/relax/test_op_statistical.py
@@ -14,7 +14,7 @@
 # KIND, either express or implied.  See the License for the
 # specific language governing permissions and limitations
 # under the License.
-from typing import Callable
+from collections.abc import Callable
 
 import pytest
 
diff --git a/tests/python/relax/test_op_unary.py b/tests/python/relax/test_op_unary.py
index 1e30f45050d5..ca4e65f89c2f 100644
--- a/tests/python/relax/test_op_unary.py
+++ b/tests/python/relax/test_op_unary.py
@@ -14,7 +14,7 @@
 # KIND, either express or implied.  See the License for the
 # specific language governing permissions and limitations
 # under the License.
-from typing import Callable
+from collections.abc import Callable
 
 import pytest
 
diff --git a/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_cpu.py b/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_cpu.py
index 19eacfa67c17..8f75804cbba6 100644
--- a/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_cpu.py
+++ b/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_cpu.py
@@ -17,7 +17,6 @@
 # ruff: noqa: E501, E741
 import enum
 import itertools
-from typing import Dict, List, Optional, Tuple, Union
 
 import numpy as np
 import pytest
@@ -255,7 +254,7 @@ def verify_cached_kv(kv_cache, seq_ids, expected_k, expected_v):
         tvm.testing.assert_allclose(values.numpy(), values_expected, rtol=1e-3, atol=1e-3)
 
 
-def f_apply_rotary(x, offset, scale, theta, offset_list: Optional[List[int]] = None):
+def f_apply_rotary(x, offset, scale, theta, offset_list: list[int] | None = None):
     # x: (N, H, D)
     assert len(x.shape) == 3
     nfeat = x.shape[-1]
@@ -280,13 +279,13 @@ def f_apply_rotary(x, offset, scale, theta, offset_list: Optional[List[int]] = N
 def apply_attention(
     kv_cache,
     rope_mode: RopeMode,
-    batch: List[Tuple[Union[int, Tuple[int, int, int]], int]],
-    cached_k: Dict[int, np.ndarray],
-    cached_v: Dict[int, np.ndarray],
-    sliding_window_sizes: Optional[List[int]] = None,
-    attn_sink_sizes: Optional[List[int]] = None,
-    token_tree_parent_ptr_list: Optional[List[List[int]]] = None,
-    accepted_leaf_indices: Optional[List[int]] = None,
+    batch: list[tuple[int | tuple[int, int, int], int]],
+    cached_k: dict[int, np.ndarray],
+    cached_v: dict[int, np.ndarray],
+    sliding_window_sizes: list[int] | None = None,
+    attn_sink_sizes: list[int] | None = None,
+    token_tree_parent_ptr_list: list[list[int]] | None = None,
+    accepted_leaf_indices: list[int] | None = None,
 ) -> None:
     seq_ids = []
     append_lengths = []
@@ -314,7 +313,7 @@ def apply_attention(
             cached_v[seq_id] = np.zeros((num_layers, 0, num_kv_heads, head_dim), dtype)
 
     flattened_token_tree_parent_ptr = None
-    token_tree_node_depths_list: List[Optional[List[int]]] = [None for _ in batch]
+    token_tree_node_depths_list: list[list[int] | None] = [None for _ in batch]
 
     if token_tree_parent_ptr_list:
         assert len(token_tree_node_depths_list) == len(seq_ids)
diff --git a/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_flashinfer.py b/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_flashinfer.py
index 68f9035f2378..fdc706d4d457 100644
--- a/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_flashinfer.py
+++ b/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_flashinfer.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 # ruff: noqa: E741
-from typing import Dict, List, Tuple, Union
 
 import pytest
 import torch
@@ -243,9 +242,9 @@ def f_apply_rotary(x, offset, scale, theta):
 def apply_attention(
     kv_cache,
     rope_mode: RopeMode,
-    batch: List[Tuple[Union[int, Tuple[int, int, int]], int]],
-    cached_k: Dict[int, torch.Tensor],
-    cached_v: Dict[int, torch.Tensor],
+    batch: list[tuple[int | tuple[int, int, int], int]],
+    cached_k: dict[int, torch.Tensor],
+    cached_v: dict[int, torch.Tensor],
 ) -> None:
     seq_ids = []
     append_lengths = []
diff --git a/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_mla_flashinfer.py b/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_mla_flashinfer.py
index cb27159fe0e8..07da9da8f7ba 100644
--- a/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_mla_flashinfer.py
+++ b/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_mla_flashinfer.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 import itertools
-from typing import Dict, List, Tuple, Union
 
 import numpy as np
 import pytest
@@ -245,8 +244,8 @@ def verify_cached_kv(kv_cache, seq_ids, expected_kv):
 
 def apply_attention(
     kv_cache,
-    batch: List[Tuple[Union[int, Tuple[int, int, int]], int]],
-    cached_kv: Dict[int, torch.Tensor],
+    batch: list[tuple[int | tuple[int, int, int], int]],
+    cached_kv: dict[int, torch.Tensor],
 ) -> None:
     seq_ids = []
     append_lengths = []
diff --git a/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_mla_tir.py b/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_mla_tir.py
index 34a9b2d26419..38c5cda9b1d3 100644
--- a/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_mla_tir.py
+++ b/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_mla_tir.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 import itertools
-from typing import Dict, List, Tuple, Union
 
 import numpy as np
 import pytest
@@ -227,8 +226,8 @@ def verify_cached_kv(kv_cache, seq_ids, expected_kv):
 
 def apply_attention(
     kv_cache,
-    batch: List[Tuple[Union[int, Tuple[int, int, int]], int]],
-    cached_kv: Dict[int, torch.Tensor],
+    batch: list[tuple[int | tuple[int, int, int], int]],
+    cached_kv: dict[int, torch.Tensor],
 ) -> None:
     seq_ids = []
     append_lengths = []
diff --git a/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_tir.py b/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_tir.py
index 85714199f691..f385e53db083 100644
--- a/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_tir.py
+++ b/tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_tir.py
@@ -16,7 +16,6 @@
 # under the License.
 # ruff: noqa: E501, E741
 import itertools
-from typing import Dict, List, Optional, Tuple, Union
 
 import pytest
 import torch
@@ -250,7 +249,7 @@ def verify_cached_kv(kv_cache, seq_ids, expected_k, expected_v):
         )
 
 
-def f_apply_rotary(x, offset, scale, theta, offset_list: Optional[List[int]] = None):
+def f_apply_rotary(x, offset, scale, theta, offset_list: list[int] | None = None):
     # x: (N, H, D)
     assert len(x.shape) == 3
     nfeat = x.shape[-1]
@@ -280,13 +279,13 @@ def f_apply_rotary(x, offset, scale, theta, offset_list: Optional[List[int]] = N
 def apply_attention(
     kv_cache,
     rope_mode: RopeMode,
-    batch: List[Tuple[Union[int, Tuple[int, int, int]], int]],
-    cached_k: Dict[int, torch.Tensor],
-    cached_v: Dict[int, torch.Tensor],
-    sliding_window_sizes: Optional[List[int]] = None,
-    attn_sink_sizes: Optional[List[int]] = None,
-    token_tree_parent_ptr_list: Optional[List[List[int]]] = None,
-    accepted_leaf_indices: Optional[List[int]] = None,
+    batch: list[tuple[int | tuple[int, int, int], int]],
+    cached_k: dict[int, torch.Tensor],
+    cached_v: dict[int, torch.Tensor],
+    sliding_window_sizes: list[int] | None = None,
+    attn_sink_sizes: list[int] | None = None,
+    token_tree_parent_ptr_list: list[list[int]] | None = None,
+    accepted_leaf_indices: list[int] | None = None,
 ) -> None:
     seq_ids = []
     append_lengths = []
@@ -318,7 +317,7 @@ def apply_attention(
             )
 
     flattened_token_tree_parent_ptr = None
-    token_tree_node_depths_list: List[Optional[List[int]]] = [None for _ in batch]
+    token_tree_node_depths_list: list[list[int] | None] = [None for _ in batch]
     if token_tree_parent_ptr_list:
         assert len(token_tree_node_depths_list) == len(seq_ids)
         if accepted_leaf_indices is not None:
diff --git a/tests/python/relax/test_runtime_builtin_rnn_state.py b/tests/python/relax/test_runtime_builtin_rnn_state.py
index 7a9d8afd0be0..f67b42a7ee2c 100644
--- a/tests/python/relax/test_runtime_builtin_rnn_state.py
+++ b/tests/python/relax/test_runtime_builtin_rnn_state.py
@@ -15,7 +15,7 @@
 # specific language governing permissions and limitations
 # under the License.
 # pylint: disable=missing-docstring,
-from typing import Sequence, Union
+from collections.abc import Sequence
 
 import numpy as np
 import pytest
@@ -218,7 +218,7 @@ def _rnn_state_get(
 
 
 def rnn_state_set(
-    shape: Sequence[Union[int, tir.Var]],
+    shape: Sequence[int | tir.Var],
     dtype: str,
 ):
     # fmt: off
diff --git a/tests/python/relax/test_runtime_sampling_flashinfer.py b/tests/python/relax/test_runtime_sampling_flashinfer.py
index b29eb7c0a2ea..c5092a149a08 100644
--- a/tests/python/relax/test_runtime_sampling_flashinfer.py
+++ b/tests/python/relax/test_runtime_sampling_flashinfer.py
@@ -18,7 +18,6 @@
 
 
 import random
-from typing import List
 
 import numpy as np
 import pytest
@@ -31,7 +30,7 @@
 
 @pytest.mark.skip(reason="Requires FlashInfer enabled and proper setup")
 def test_sampling():
-    def load_module(name: str, static_modules: List[tvm.runtime.Module]):
+    def load_module(name: str, static_modules: list[tvm.runtime.Module]):
         assert len(static_modules) > 0
         if len(static_modules) == 1:
             return static_modules[0]
diff --git a/tests/python/relax/test_training_optimizer_numeric.py b/tests/python/relax/test_training_optimizer_numeric.py
index 6aa566bf6fc3..c9ca6b209788 100644
--- a/tests/python/relax/test_training_optimizer_numeric.py
+++ b/tests/python/relax/test_training_optimizer_numeric.py
@@ -16,7 +16,7 @@
 # under the License.
 """Numeric tests for relax optimizer APIs."""
 
-from typing import Callable, List
+from collections.abc import Callable
 
 import numpy as np
 
@@ -36,20 +36,20 @@ def _legalize_and_build(mod: IRModule, target, dev):
 
 
 def _numpy_to_tvm(data):
-    if isinstance(data, (list, tuple)):
+    if isinstance(data, list | tuple):
         return [_numpy_to_tvm(_data) for _data in data]
     return tvm.runtime.tensor(data)
 
 
 def _tvm_to_numpy(data):
-    if isinstance(data, (list, tuple, tvm.ir.Array)):
+    if isinstance(data, list | tuple | tvm.ir.Array):
         return [_tvm_to_numpy(_data) for _data in data]
     return data.numpy()
 
 
 def _assert_allclose_nested(data1, data2):
-    if isinstance(data1, (list, tuple)):
-        assert isinstance(data2, (list, tuple))
+    if isinstance(data1, list | tuple):
+        assert isinstance(data2, list | tuple)
         assert len(data1) == len(data2)
         for x, y in zip(data1, data2):
             _assert_allclose_nested(x, y)
@@ -57,7 +57,7 @@ def _assert_allclose_nested(data1, data2):
         assert_allclose(data1, data2)
 
 
-def _assert_run_result_same(tvm_func: Callable, np_func: Callable, np_inputs: List):
+def _assert_run_result_same(tvm_func: Callable, np_func: Callable, np_inputs: list):
     result = _tvm_to_numpy(tvm_func(*[_numpy_to_tvm(i) for i in np_inputs]))
     expected = np_func(*np_inputs)
     _assert_allclose_nested(result, expected)
diff --git a/tests/python/relax/test_transform_few_shot_tuning.py b/tests/python/relax/test_transform_few_shot_tuning.py
index 4d79a9c40dcb..6c8ee372908a 100644
--- a/tests/python/relax/test_transform_few_shot_tuning.py
+++ b/tests/python/relax/test_transform_few_shot_tuning.py
@@ -16,7 +16,6 @@
 # under the License.
 # pylint: disable=invalid-name,,missing-function-docstring
 # ruff: noqa: E501, F403
-from typing import List, Tuple
 
 import numpy as np
 import pytest
@@ -330,7 +329,7 @@ def _get_single_prim_func(mod: tvm.ir.IRModule) -> tvm.tir.PrimFunc:
     return funcs[0]
 
 
-def _get_input_output_info(func: tvm.tir.PrimFunc) -> Tuple[List[np.ndarray], Tuple, str]:
+def _get_input_output_info(func: tvm.tir.PrimFunc) -> tuple[list[np.ndarray], tuple, str]:
     args = ArgInfo.from_prim_func(func)
     inputs = [generate_input_data(x.shape, x.dtype) for x in args[:-1]]
     output_shape = args[-1].shape
@@ -339,7 +338,7 @@ def _get_input_output_info(func: tvm.tir.PrimFunc) -> Tuple[List[np.ndarray], Tu
 
 
 def _expected_results(
-    mod: tvm.ir.IRModule, inputs: List[np.ndarray], output_shape: Tuple, output_dtype: str
+    mod: tvm.ir.IRModule, inputs: list[np.ndarray], output_shape: tuple, output_dtype: str
 ) -> np.ndarray:
     func = _get_single_prim_func(mod)
     func = func.with_attr("global_symbol", "main")
@@ -356,7 +355,7 @@ def _expected_results(
 
 
 def _actual_results(
-    actual: tvm.ir.IRModule, inputs: List[np.ndarray], output_shape: Tuple, output_dtype: str
+    actual: tvm.ir.IRModule, inputs: list[np.ndarray], output_shape: tuple, output_dtype: str
 ):
     target = _target()
     actual_rt_mod = tvm.compile(actual, target=target)
diff --git a/tests/python/relax/test_transform_legalize_ops_unary.py b/tests/python/relax/test_transform_legalize_ops_unary.py
index c5fd494468b2..55fa34ab2204 100644
--- a/tests/python/relax/test_transform_legalize_ops_unary.py
+++ b/tests/python/relax/test_transform_legalize_ops_unary.py
@@ -16,7 +16,7 @@
 # under the License.
 # ruff: noqa: F401
 
-from typing import Callable
+from collections.abc import Callable
 
 import pytest
 
diff --git a/tests/python/relax/test_transform_update_param_struct_info.py b/tests/python/relax/test_transform_update_param_struct_info.py
index b56320046873..30200955322d 100644
--- a/tests/python/relax/test_transform_update_param_struct_info.py
+++ b/tests/python/relax/test_transform_update_param_struct_info.py
@@ -16,7 +16,6 @@
 # under the License.
 
 import inspect
-from typing import Optional
 
 import pytest
 
@@ -37,13 +36,13 @@ def test_compare(self):
             after = transform(self.Before)
             tvm.ir.assert_structural_equal(self.Expected, after)
 
-    def update_sinfo(self, var: relax.Var) -> Optional[relax.StructInfo]:
+    def update_sinfo(self, var: relax.Var) -> relax.StructInfo | None:
         """The struct info update function provided to the transform"""
         raise NotImplementedError("Should be implemented in derived class")
 
 
 class TestSimple(Base):
-    def update_sinfo(self, var: relax.Var) -> Optional[relax.StructInfo]:
+    def update_sinfo(self, var: relax.Var) -> relax.StructInfo | None:
         if var.name_hint == "weight":
             return relax.TensorStructInfo([64, 16], "float32")
 
diff --git a/tests/python/relax/test_tvmscript_parser.py b/tests/python/relax/test_tvmscript_parser.py
index 860f10e19775..720b8290e385 100644
--- a/tests/python/relax/test_tvmscript_parser.py
+++ b/tests/python/relax/test_tvmscript_parser.py
@@ -31,8 +31,8 @@
 
 
 def _check(
-    parsed: Union[relax.Function, IRModule],
-    expect: Optional[Union[relax.Function, IRModule]] = None,
+    parsed: relax.Function | IRModule,
+    expect: relax.Function | IRModule | None = None,
 ):
     test = parsed.script(show_meta=True)
     roundtrip_mod = tvm.script.from_source(test)
diff --git a/tests/python/relax/test_tvmscript_parser_op_arith_cmp.py b/tests/python/relax/test_tvmscript_parser_op_arith_cmp.py
index 961273ea5e3e..830d2cf7d472 100644
--- a/tests/python/relax/test_tvmscript_parser_op_arith_cmp.py
+++ b/tests/python/relax/test_tvmscript_parser_op_arith_cmp.py
@@ -15,7 +15,7 @@
 # specific language governing permissions and limitations
 # under the License.
 
-from typing import Callable, Optional, Union
+from collections.abc import Callable
 
 import tvm
 import tvm.script
@@ -25,8 +25,8 @@
 
 
 def _check(
-    parsed: Union[relax.Function, IRModule],
-    expect: Optional[Union[relax.Function, IRModule]],
+    parsed: relax.Function | IRModule,
+    expect: relax.Function | IRModule | None,
 ):
     test = parsed.script(show_meta=True)
     roundtrip_mod = tvm.script.from_source(test)
diff --git a/tests/python/relax/test_tvmscript_parser_op_create.py b/tests/python/relax/test_tvmscript_parser_op_create.py
index e2def5f61606..90c2b709b2c2 100644
--- a/tests/python/relax/test_tvmscript_parser_op_create.py
+++ b/tests/python/relax/test_tvmscript_parser_op_create.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 
-from typing import Optional, Union
 
 import tvm
 import tvm.script
@@ -25,8 +24,8 @@
 
 
 def _check(
-    parsed: Union[relax.Function, IRModule],
-    expect: Optional[Union[relax.Function, IRModule]],
+    parsed: relax.Function | IRModule,
+    expect: relax.Function | IRModule | None,
 ):
     test = parsed.script(show_meta=True)
     roundtrip_mod = tvm.script.from_source(test)
diff --git a/tests/python/relax/test_tvmscript_parser_op_datatype.py b/tests/python/relax/test_tvmscript_parser_op_datatype.py
index 85c5faa8667b..0f0646d1d5dc 100644
--- a/tests/python/relax/test_tvmscript_parser_op_datatype.py
+++ b/tests/python/relax/test_tvmscript_parser_op_datatype.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 
-from typing import Optional, Union
 
 import tvm
 import tvm.script
@@ -25,8 +24,8 @@
 
 
 def _check(
-    parsed: Union[relax.Function, IRModule],
-    expect: Optional[Union[relax.Function, IRModule]],
+    parsed: relax.Function | IRModule,
+    expect: relax.Function | IRModule | None,
 ):
     test = parsed.script(show_meta=True)
     roundtrip_mod = tvm.script.from_source(test)
diff --git a/tests/python/relax/test_tvmscript_parser_op_grad.py b/tests/python/relax/test_tvmscript_parser_op_grad.py
index 9dd1f01fc763..d16d0175b274 100644
--- a/tests/python/relax/test_tvmscript_parser_op_grad.py
+++ b/tests/python/relax/test_tvmscript_parser_op_grad.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 
-from typing import Optional, Union
 
 import tvm
 import tvm.testing
@@ -24,8 +23,8 @@
 
 
 def _check(
-    parsed: Union[relax.Function, IRModule],
-    expect: Optional[Union[relax.Function, IRModule]],
+    parsed: relax.Function | IRModule,
+    expect: relax.Function | IRModule | None,
 ):
     test = parsed.script(show_meta=True)
     roundtrip_mod = tvm.script.from_source(test)
diff --git a/tests/python/relax/test_tvmscript_parser_op_image.py b/tests/python/relax/test_tvmscript_parser_op_image.py
index a90da37812ef..47a7e3f19601 100644
--- a/tests/python/relax/test_tvmscript_parser_op_image.py
+++ b/tests/python/relax/test_tvmscript_parser_op_image.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 
-from typing import Optional, Union
 
 import tvm
 import tvm.script
@@ -25,8 +24,8 @@
 
 
 def _check(
-    parsed: Union[relax.Function, IRModule],
-    expect: Optional[Union[relax.Function, IRModule]],
+    parsed: relax.Function | IRModule,
+    expect: relax.Function | IRModule | None,
 ):
     test = parsed.script(show_meta=True)
     roundtrip_mod = tvm.script.from_source(test)
diff --git a/tests/python/relax/test_tvmscript_parser_op_index.py b/tests/python/relax/test_tvmscript_parser_op_index.py
index e9821eba6d23..f7148cdf00be 100644
--- a/tests/python/relax/test_tvmscript_parser_op_index.py
+++ b/tests/python/relax/test_tvmscript_parser_op_index.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 
-from typing import Optional, Union
 
 import tvm
 import tvm.script
@@ -25,8 +24,8 @@
 
 
 def _check(
-    parsed: Union[relax.Function, IRModule],
-    expect: Optional[Union[relax.Function, IRModule]],
+    parsed: relax.Function | IRModule,
+    expect: relax.Function | IRModule | None,
 ):
     test = parsed.script(show_meta=True)
     roundtrip_mod = tvm.script.from_source(test)
diff --git a/tests/python/relax/test_tvmscript_parser_op_linear_algebra.py b/tests/python/relax/test_tvmscript_parser_op_linear_algebra.py
index 89665b0dce86..ab57b757dfd0 100644
--- a/tests/python/relax/test_tvmscript_parser_op_linear_algebra.py
+++ b/tests/python/relax/test_tvmscript_parser_op_linear_algebra.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 
-from typing import Optional, Union
 
 import tvm
 import tvm.script
@@ -25,8 +24,8 @@
 
 
 def _check(
-    parsed: Union[relax.Function, IRModule],
-    expect: Optional[Union[relax.Function, IRModule]],
+    parsed: relax.Function | IRModule,
+    expect: relax.Function | IRModule | None,
 ):
     test = parsed.script(show_meta=True)
     roundtrip_mod = tvm.script.from_source(test)
diff --git a/tests/python/relax/test_tvmscript_parser_op_manipulate.py b/tests/python/relax/test_tvmscript_parser_op_manipulate.py
index ef30270c0d00..5b20534e622b 100644
--- a/tests/python/relax/test_tvmscript_parser_op_manipulate.py
+++ b/tests/python/relax/test_tvmscript_parser_op_manipulate.py
@@ -16,7 +16,6 @@
 # under the License.
 # ruff: noqa: E731
 
-from typing import Optional, Union
 
 import tvm
 import tvm.script
@@ -27,8 +26,8 @@
 
 
 def _check(
-    parsed: Union[relax.Function, IRModule],
-    expect: Optional[Union[relax.Function, IRModule]],
+    parsed: relax.Function | IRModule,
+    expect: relax.Function | IRModule | None,
 ):
     test = parsed.script(show_meta=True)
     roundtrip_mod = tvm.script.from_source(test)
diff --git a/tests/python/relax/test_tvmscript_parser_op_nn.py b/tests/python/relax/test_tvmscript_parser_op_nn.py
index a8d98903d252..b9aeb9efe596 100644
--- a/tests/python/relax/test_tvmscript_parser_op_nn.py
+++ b/tests/python/relax/test_tvmscript_parser_op_nn.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 
-from typing import Optional, Union
 
 import tvm
 import tvm.script
@@ -25,8 +24,8 @@
 
 
 def _check(
-    parsed: Union[relax.Function, IRModule],
-    expect: Optional[Union[relax.Function, IRModule]],
+    parsed: relax.Function | IRModule,
+    expect: relax.Function | IRModule | None,
 ):
     test = parsed.script(show_meta=True)
     roundtrip_mod = tvm.script.from_source(test)
diff --git a/tests/python/relax/test_tvmscript_parser_op_search.py b/tests/python/relax/test_tvmscript_parser_op_search.py
index 86d0cfc9bcf9..86693b17d20d 100644
--- a/tests/python/relax/test_tvmscript_parser_op_search.py
+++ b/tests/python/relax/test_tvmscript_parser_op_search.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 
-from typing import Optional, Union
 
 import tvm
 import tvm.script
@@ -25,8 +24,8 @@
 
 
 def _check(
-    parsed: Union[relax.Function, IRModule],
-    expect: Optional[Union[relax.Function, IRModule]],
+    parsed: relax.Function | IRModule,
+    expect: relax.Function | IRModule | None,
 ):
     test = parsed.script(show_meta=True)
     roundtrip_mod = tvm.script.from_source(test)
diff --git a/tests/python/relax/test_tvmscript_parser_op_set.py b/tests/python/relax/test_tvmscript_parser_op_set.py
index 5476be6d6bc0..505dabde8fb4 100644
--- a/tests/python/relax/test_tvmscript_parser_op_set.py
+++ b/tests/python/relax/test_tvmscript_parser_op_set.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 
-from typing import Optional, Union
 
 import tvm
 import tvm.script
@@ -25,8 +24,8 @@
 
 
 def _check(
-    parsed: Union[relax.Function, IRModule],
-    expect: Optional[Union[relax.Function, IRModule]],
+    parsed: relax.Function | IRModule,
+    expect: relax.Function | IRModule | None,
 ):
     test = parsed.script(show_meta=True)
     roundtrip_mod = tvm.script.from_source(test)
diff --git a/tests/python/relax/test_tvmscript_parser_op_sort.py b/tests/python/relax/test_tvmscript_parser_op_sort.py
index 1e025ed60b52..e537a862d7db 100644
--- a/tests/python/relax/test_tvmscript_parser_op_sort.py
+++ b/tests/python/relax/test_tvmscript_parser_op_sort.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 
-from typing import Optional, Union
 
 import tvm
 import tvm.script
@@ -25,8 +24,8 @@
 
 
 def _check(
-    parsed: Union[relax.Function, IRModule],
-    expect: Optional[Union[relax.Function, IRModule]],
+    parsed: relax.Function | IRModule,
+    expect: relax.Function | IRModule | None,
 ):
     test = parsed.script(show_meta=True)
     roundtrip_mod = tvm.script.from_source(test)
diff --git a/tests/python/relax/test_tvmscript_parser_op_statistical.py b/tests/python/relax/test_tvmscript_parser_op_statistical.py
index 8b3917d88739..2621e1d7cf2b 100644
--- a/tests/python/relax/test_tvmscript_parser_op_statistical.py
+++ b/tests/python/relax/test_tvmscript_parser_op_statistical.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 
-from typing import Optional, Union
 
 import tvm
 import tvm.script
@@ -25,8 +24,8 @@
 
 
 def _check(
-    parsed: Union[relax.Function, IRModule],
-    expect: Optional[Union[relax.Function, IRModule]],
+    parsed: relax.Function | IRModule,
+    expect: relax.Function | IRModule | None,
 ):
     test = parsed.script(show_meta=True)
     roundtrip_mod = tvm.script.from_source(test)
diff --git a/tests/python/relax/test_tvmscript_parser_op_vision.py b/tests/python/relax/test_tvmscript_parser_op_vision.py
index cab0e731a648..10817c1dc45d 100644
--- a/tests/python/relax/test_tvmscript_parser_op_vision.py
+++ b/tests/python/relax/test_tvmscript_parser_op_vision.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 
-from typing import Optional, Union
 
 import tvm
 import tvm.script
@@ -25,8 +24,8 @@
 
 
 def _check(
-    parsed: Union[relax.Function, IRModule],
-    expect: Optional[Union[relax.Function, IRModule]],
+    parsed: relax.Function | IRModule,
+    expect: relax.Function | IRModule | None,
 ):
     test = parsed.script(show_meta=True)
     roundtrip_mod = tvm.script.from_source(test)
diff --git a/tests/python/relax/test_vm_build.py b/tests/python/relax/test_vm_build.py
index c0840f188a52..d2ab2a7c41ad 100644
--- a/tests/python/relax/test_vm_build.py
+++ b/tests/python/relax/test_vm_build.py
@@ -17,7 +17,7 @@
 # ruff: noqa: F841
 
 import ctypes
-from typing import Callable, Tuple
+from collections.abc import Callable
 
 import numpy as np
 import pytest
@@ -1109,7 +1109,7 @@ def set_input_attempt_get(vm: relax.VirtualMachine, device: tvm.runtime.Device)
     _ = vm.get_outputs("main")
 
 
-def make_vm(mod, exec_mode, temp) -> Tuple[relax.VirtualMachine, tvm.runtime.Device]:
+def make_vm(mod, exec_mode, temp) -> tuple[relax.VirtualMachine, tvm.runtime.Device]:
     """Returns a local VM for the given mod and the device"""
     target = tvm.target.Target("llvm", host="llvm")
     exec = relax.build(mod, target, exec_mode=exec_mode)
diff --git a/tests/python/relax/test_vm_multi_device.py b/tests/python/relax/test_vm_multi_device.py
index ec0c94c05bca..d065d0078e4d 100644
--- a/tests/python/relax/test_vm_multi_device.py
+++ b/tests/python/relax/test_vm_multi_device.py
@@ -16,8 +16,6 @@
 # under the License.
 """Test eliminate common subexpr pass"""
 
-from typing import List
-
 import numpy as np
 
 import tvm
@@ -31,7 +29,7 @@
 
 def compile(
     mod: IRModule,
-    device: List[Device] = [
+    device: list[Device] = [
         tvm.cpu(),
     ],
 ) -> relax.VirtualMachine:
diff --git a/tests/python/s_tir/meta_schedule/test_meta_schedule_builder.py b/tests/python/s_tir/meta_schedule/test_meta_schedule_builder.py
index 9a1d356e448c..b6e2eed3e186 100644
--- a/tests/python/s_tir/meta_schedule/test_meta_schedule_builder.py
+++ b/tests/python/s_tir/meta_schedule/test_meta_schedule_builder.py
@@ -20,7 +20,6 @@
 import os
 import sys
 import time
-from typing import List
 
 import pytest
 from tvm_ffi import register_global_func
@@ -100,7 +99,7 @@ def batch_matmul(  # pylint: disable=no-self-argument
 # pylint: enable=invalid-name,no-member,line-too-long,too-many-nested-blocks,missing-docstring
 
 
-def _check_build_results(builder_results: List[BuilderResult]):
+def _check_build_results(builder_results: list[BuilderResult]):
     """Simple check whether the build is successful"""
     for result in builder_results:
         artifact_path = result.artifact_path
@@ -140,8 +139,8 @@ def test_meta_schedule_error_handle_test_builder():
     class TestBuilder(PyBuilder):
         def build(  # pylint: disable=no-self-use
             self,
-            build_inputs: List[BuilderInput],
-        ) -> List[BuilderResult]:
+            build_inputs: list[BuilderInput],
+        ) -> list[BuilderResult]:
             return [BuilderResult(None, "error") for w in build_inputs]
 
     builder = TestBuilder()
diff --git a/tests/python/s_tir/meta_schedule/test_meta_schedule_cost_model.py b/tests/python/s_tir/meta_schedule/test_meta_schedule_cost_model.py
index 27139750e212..57ee5db6f07d 100644
--- a/tests/python/s_tir/meta_schedule/test_meta_schedule_cost_model.py
+++ b/tests/python/s_tir/meta_schedule/test_meta_schedule_cost_model.py
@@ -22,7 +22,6 @@
 import tempfile
 import unittest
 from functools import partial
-from typing import List
 
 import numpy as np
 
@@ -84,12 +83,12 @@ def save(self, path: str) -> None:
         def update(
             self,
             context: TuneContext,
-            candidates: List[MeasureCandidate],
-            results: List[RunnerResult],
+            candidates: list[MeasureCandidate],
+            results: list[RunnerResult],
         ) -> None:
             pass
 
-        def predict(self, context: TuneContext, candidates: List[MeasureCandidate]) -> np.ndarray:
+        def predict(self, context: TuneContext, candidates: list[MeasureCandidate]) -> np.ndarray:
             return np.random.rand(10)
 
     model = FancyCostModel()
@@ -114,12 +113,12 @@ def save(self, path: str) -> None:
         def update(
             self,
             context: TuneContext,
-            candidates: List[MeasureCandidate],
-            results: List[RunnerResult],
+            candidates: list[MeasureCandidate],
+            results: list[RunnerResult],
         ) -> None:
             pass
 
-        def predict(self, context: TuneContext, candidates: List[MeasureCandidate]) -> np.ndarray:
+        def predict(self, context: TuneContext, candidates: list[MeasureCandidate]) -> np.ndarray:
             return np.random.rand(10)
 
     cost_model = NotSoFancyCostModel()
diff --git a/tests/python/s_tir/meta_schedule/test_meta_schedule_database.py b/tests/python/s_tir/meta_schedule/test_meta_schedule_database.py
index 3d3e026e18fc..f7ddc44b22f6 100644
--- a/tests/python/s_tir/meta_schedule/test_meta_schedule_database.py
+++ b/tests/python/s_tir/meta_schedule/test_meta_schedule_database.py
@@ -20,7 +20,8 @@
 
 import os.path as osp
 import tempfile
-from typing import Callable, List, Optional
+from collections.abc import Callable
+from typing import Optional
 
 import pytest
 
@@ -116,8 +117,8 @@ def _equal_record(a: ms.database.TuningRecord, b: ms.database.TuningRecord):
 class PyMemoryDatabaseDefault(ms.database.PyDatabase):
     def __init__(self):
         super().__init__()
-        self.tuning_records_: List[TuningRecord] = []
-        self.workloads_: List[Workload] = []
+        self.tuning_records_: list[TuningRecord] = []
+        self.workloads_: list[Workload] = []
 
     def has_workload(self, mod: IRModule) -> bool:
         for workload in self.workloads_:
@@ -137,10 +138,10 @@ def commit_workload(self, mod: IRModule) -> ms.database.Workload:
     def commit_tuning_record(self, record: TuningRecord) -> None:
         self.tuning_records_.append(record)
 
-    def get_all_tuning_records(self) -> List[TuningRecord]:
+    def get_all_tuning_records(self) -> list[TuningRecord]:
         return self.tuning_records_
 
-    def get_top_k(self, workload: ms.database.Workload, top_k: int) -> List[TuningRecord]:
+    def get_top_k(self, workload: ms.database.Workload, top_k: int) -> list[TuningRecord]:
         return sorted(
             list(
                 filter(
@@ -159,8 +160,8 @@ def __len__(self) -> int:
 class PyMemoryDatabaseOverride(ms.database.PyDatabase):
     def __init__(self):
         super().__init__()
-        self.tuning_records_: List[TuningRecord] = []
-        self.workloads_: List[Workload] = []
+        self.tuning_records_: list[TuningRecord] = []
+        self.workloads_: list[Workload] = []
 
     def has_workload(self, mod: IRModule) -> bool:
         for workload in self.workloads_:
@@ -180,10 +181,10 @@ def commit_workload(self, mod: IRModule) -> ms.database.Workload:
     def commit_tuning_record(self, record: TuningRecord) -> None:
         self.tuning_records_.append(record)
 
-    def get_all_tuning_records(self) -> List[TuningRecord]:
+    def get_all_tuning_records(self) -> list[TuningRecord]:
         return self.tuning_records_
 
-    def get_top_k(self, workload: ms.database.Workload, top_k: int) -> List[TuningRecord]:
+    def get_top_k(self, workload: ms.database.Workload, top_k: int) -> list[TuningRecord]:
         return sorted(
             list(
                 filter(
@@ -198,8 +199,8 @@ def __len__(self) -> int:
         return len(self.tuning_records_)
 
     def query_tuning_record(
-        self, mod: IRModule, target: Target, workload_name: Optional[str] = None
-    ) -> Optional[TuningRecord]:
+        self, mod: IRModule, target: Target, workload_name: str | None = None
+    ) -> TuningRecord | None:
         if self.has_workload(mod):
             records = self.get_top_k(self.commit_workload(mod), 2)
             if len(records) == 1:
@@ -209,8 +210,8 @@ def query_tuning_record(
         return None
 
     def query_schedule(
-        self, mod: IRModule, target: Target, workload_name: Optional[str] = None
-    ) -> Optional[Schedule]:
+        self, mod: IRModule, target: Target, workload_name: str | None = None
+    ) -> Schedule | None:
         record = self.query_tuning_record(mod, target, workload_name)
         if record is not None:
             sch = Schedule(record.workload.mod)
@@ -219,8 +220,8 @@ def query_schedule(
         return None
 
     def query_ir_module(
-        self, mod: IRModule, target: Target, workload_name: Optional[str] = None
-    ) -> Optional[IRModule]:
+        self, mod: IRModule, target: Target, workload_name: str | None = None
+    ) -> IRModule | None:
         record = self.query_tuning_record(mod, target, workload_name)
         if record is not None:
             sch = Schedule(record.workload.mod)
diff --git a/tests/python/s_tir/meta_schedule/test_meta_schedule_feature_extractor.py b/tests/python/s_tir/meta_schedule/test_meta_schedule_feature_extractor.py
index dd7dbfac3666..1d336d9b5aa0 100644
--- a/tests/python/s_tir/meta_schedule/test_meta_schedule_feature_extractor.py
+++ b/tests/python/s_tir/meta_schedule/test_meta_schedule_feature_extractor.py
@@ -16,7 +16,6 @@
 # under the License.
 # pylint: disable=missing-module-docstring,missing-function-docstring,missing-class-docstring
 import re
-from typing import List
 
 import numpy as np
 
@@ -33,8 +32,8 @@ class FancyFeatureExtractor(PyFeatureExtractor):
         def extract_from(
             self,
             context: TuneContext,  # pylint: disable = unused-argument
-            candidates: List[MeasureCandidate],  # pylint: disable = unused-argument
-        ) -> List[np.ndarray]:
+            candidates: list[MeasureCandidate],  # pylint: disable = unused-argument
+        ) -> list[np.ndarray]:
             return [tvm.runtime.tensor(np.random.rand(4, 5))]
 
     extractor = FancyFeatureExtractor()
@@ -49,8 +48,8 @@ class NotSoFancyFeatureExtractor(PyFeatureExtractor):
         def extract_from(
             self,
             context: TuneContext,  # pylint: disable = unused-argument
-            candidates: List[MeasureCandidate],  # pylint: disable = unused-argument
-        ) -> List[np.ndarray]:
+            candidates: list[MeasureCandidate],  # pylint: disable = unused-argument
+        ) -> list[np.ndarray]:
             return []
 
     feature_extractor = NotSoFancyFeatureExtractor()
diff --git a/tests/python/s_tir/meta_schedule/test_meta_schedule_feature_extractor_per_store_feature.py b/tests/python/s_tir/meta_schedule/test_meta_schedule_feature_extractor_per_store_feature.py
index 090c8e36a02b..24657f5ba8d5 100644
--- a/tests/python/s_tir/meta_schedule/test_meta_schedule_feature_extractor_per_store_feature.py
+++ b/tests/python/s_tir/meta_schedule/test_meta_schedule_feature_extractor_per_store_feature.py
@@ -17,7 +17,7 @@
 # pylint: disable=missing-module-docstring,missing-function-docstring,missing-class-docstring
 # ruff: noqa: E501, F401
 import sys
-from typing import Callable, List
+from collections.abc import Callable
 
 import pytest
 from numpy.testing import assert_allclose
@@ -93,7 +93,7 @@ def _make_candidate(f_sch: Callable[[], s_tir.Schedule]) -> ms.MeasureCandidate:
 def _feature_names(  # pylint: disable=invalid-name
     buffers_per_store: int = 5,
     arith_intensity_curve_num_samples: int = 10,
-) -> List[str]:
+) -> list[str]:
     result = [
         "float_mad",
         "float_addsub",
diff --git a/tests/python/s_tir/meta_schedule/test_meta_schedule_measure_callback.py b/tests/python/s_tir/meta_schedule/test_meta_schedule_measure_callback.py
index 0a730b0a5968..d2e5845e45a7 100644
--- a/tests/python/s_tir/meta_schedule/test_meta_schedule_measure_callback.py
+++ b/tests/python/s_tir/meta_schedule/test_meta_schedule_measure_callback.py
@@ -17,7 +17,6 @@
 # pylint: disable=missing-module-docstring,missing-function-docstring,missing-class-docstring
 import re
 import tempfile
-from typing import List
 
 import pytest
 
@@ -55,9 +54,9 @@ def apply(
             self,
             task_scheduler: ms.task_scheduler.TaskScheduler,
             task_id: int,
-            measure_candidates: List[ms.MeasureCandidate],
-            builder_results: List[ms.builder.BuilderResult],
-            runner_results: List[ms.runner.RunnerResult],
+            measure_candidates: list[ms.MeasureCandidate],
+            builder_results: list[ms.builder.BuilderResult],
+            runner_results: list[ms.runner.RunnerResult],
         ) -> None:
             assert len(measure_candidates) == 1
             tvm.ir.assert_structural_equal(measure_candidates[0].sch.mod, Matmul)
@@ -89,9 +88,9 @@ def apply(
             self,
             task_scheduler: ms.task_scheduler.TaskScheduler,
             task_id: int,
-            measure_candidates: List[ms.MeasureCandidate],
-            builder_results: List[ms.builder.BuilderResult],
-            runner_results: List[ms.runner.RunnerResult],
+            measure_candidates: list[ms.MeasureCandidate],
+            builder_results: list[ms.builder.BuilderResult],
+            runner_results: list[ms.runner.RunnerResult],
         ) -> None:
             raise ValueError("test")
 
@@ -113,9 +112,9 @@ def apply(
             self,
             task_scheduler: ms.task_scheduler.TaskScheduler,
             task_id: int,
-            measure_candidates: List[ms.MeasureCandidate],
-            builder_results: List[ms.builder.BuilderResult],
-            runner_results: List[ms.runner.RunnerResult],
+            measure_candidates: list[ms.MeasureCandidate],
+            builder_results: list[ms.builder.BuilderResult],
+            runner_results: list[ms.runner.RunnerResult],
         ) -> None:
             pass
 
@@ -135,7 +134,7 @@ def result(self) -> ms.runner.RunnerResult:
 
     @ms.derived_object
     class AllZeroRunner(ms.runner.PyRunner):
-        def run(self, runner_inputs: List[ms.runner.RunnerInput]) -> List[ms.runner.RunnerResult]:
+        def run(self, runner_inputs: list[ms.runner.RunnerInput]) -> list[ms.runner.RunnerResult]:
             return [AllZeroRunnerFuture() for _ in runner_inputs]
 
     with tempfile.TemporaryDirectory() as work_dir:
@@ -160,7 +159,7 @@ def result(self) -> ms.runner.RunnerResult:
 
     @ms.derived_object
     class EmptyRunner(ms.runner.PyRunner):
-        def run(self, runner_inputs: List[ms.runner.RunnerInput]) -> List[ms.runner.RunnerResult]:
+        def run(self, runner_inputs: list[ms.runner.RunnerInput]) -> list[ms.runner.RunnerResult]:
             return [EmptyRunnerFuture() for _ in runner_inputs]
 
     with tempfile.TemporaryDirectory() as work_dir:
diff --git a/tests/python/s_tir/meta_schedule/test_meta_schedule_mutator_mutate_parallel.py b/tests/python/s_tir/meta_schedule/test_meta_schedule_mutator_mutate_parallel.py
index a8eb9897e565..7678ee7a418b 100644
--- a/tests/python/s_tir/meta_schedule/test_meta_schedule_mutator_mutate_parallel.py
+++ b/tests/python/s_tir/meta_schedule/test_meta_schedule_mutator_mutate_parallel.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 # pylint: disable=missing-module-docstring,missing-function-docstring,missing-class-docstring
-from typing import List
 
 from tvm.s_tir import Schedule
 from tvm.s_tir import meta_schedule as ms
@@ -41,7 +40,7 @@ def matmul(a: T.handle, b: T.handle, c: T.handle) -> None:
 # pylint: enable=invalid-name, no-member
 
 
-def _sch(decisions: List[List[int]], ann_val: int) -> Schedule:
+def _sch(decisions: list[list[int]], ann_val: int) -> Schedule:
     sch = Schedule(matmul, debug_mask="all")
     # pylint: disable=invalid-name
     d0, d1, d2 = decisions
diff --git a/tests/python/s_tir/meta_schedule/test_meta_schedule_mutator_mutate_tile_size.py b/tests/python/s_tir/meta_schedule/test_meta_schedule_mutator_mutate_tile_size.py
index 2547eefe3760..bfb0ff3eb2cb 100644
--- a/tests/python/s_tir/meta_schedule/test_meta_schedule_mutator_mutate_tile_size.py
+++ b/tests/python/s_tir/meta_schedule/test_meta_schedule_mutator_mutate_tile_size.py
@@ -17,7 +17,6 @@
 # pylint: disable=missing-module-docstring,missing-function-docstring,missing-class-docstring
 import operator
 from functools import reduce
-from typing import List
 
 from tvm.s_tir import Schedule
 from tvm.s_tir import meta_schedule as ms
@@ -43,7 +42,7 @@ def matmul(a: T.handle, b: T.handle, c: T.handle) -> None:
 # pylint: enable=invalid-name, no-member
 
 
-def _sch(decisions: List[List[int]]) -> Schedule:
+def _sch(decisions: list[list[int]]) -> Schedule:
     sch = Schedule(matmul, debug_mask="all")
     # pylint: disable=invalid-name
     (d0,) = decisions
diff --git a/tests/python/s_tir/meta_schedule/test_meta_schedule_mutator_mutate_unroll.py b/tests/python/s_tir/meta_schedule/test_meta_schedule_mutator_mutate_unroll.py
index 766eced8799e..bec385163652 100644
--- a/tests/python/s_tir/meta_schedule/test_meta_schedule_mutator_mutate_unroll.py
+++ b/tests/python/s_tir/meta_schedule/test_meta_schedule_mutator_mutate_unroll.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 # pylint: disable=missing-module-docstring,missing-function-docstring,missing-class-docstring
-from typing import List
 
 from tvm.s_tir import Schedule
 from tvm.s_tir import meta_schedule as ms
@@ -41,7 +40,7 @@ def matmul(a: T.handle, b: T.handle, c: T.handle) -> None:
 # pylint: enable=invalid-name, no-member
 
 
-def _sch(decisions: List[List[int]]) -> Schedule:
+def _sch(decisions: list[list[int]]) -> Schedule:
     sch = Schedule(matmul, debug_mask="all")
     # pylint: disable=invalid-name
     d0, d1, d2 = decisions
diff --git a/tests/python/s_tir/meta_schedule/test_meta_schedule_post_order_apply.py b/tests/python/s_tir/meta_schedule/test_meta_schedule_post_order_apply.py
index b811b0e8e902..2fdd638b16c2 100644
--- a/tests/python/s_tir/meta_schedule/test_meta_schedule_post_order_apply.py
+++ b/tests/python/s_tir/meta_schedule/test_meta_schedule_post_order_apply.py
@@ -19,7 +19,6 @@
 
 import math
 import sys
-from typing import List
 
 import pytest
 from tvm_ffi import register_global_func
@@ -162,7 +161,7 @@ class WowSoFancyScheduleRule(PyScheduleRule):
     def _initialize_with_tune_context(self, context: "TuneContext") -> None:
         pass
 
-    def apply(self, sch: Schedule, block: SBlockRV) -> List[Schedule]:
+    def apply(self, sch: Schedule, block: SBlockRV) -> list[Schedule]:
         if _is_root(sch, block):
             return [sch]
         new_sch = sch.copy()
@@ -179,7 +178,7 @@ class DoubleScheduleRule(PyScheduleRule):
     def _initialize_with_tune_context(self, context: "TuneContext") -> None:
         pass
 
-    def apply(self, sch: Schedule, block: SBlockRV) -> List[Schedule]:
+    def apply(self, sch: Schedule, block: SBlockRV) -> list[Schedule]:
         if _is_root(sch, block):
             return [sch]
         new_sch = sch.copy()
@@ -204,7 +203,7 @@ class TrinityDoubleRule(PyScheduleRule):
     def _initialize_with_tune_context(self, context: "TuneContext") -> None:
         pass
 
-    def apply(self, sch: Schedule, block: SBlockRV) -> List[Schedule]:
+    def apply(self, sch: Schedule, block: SBlockRV) -> list[Schedule]:
         if _is_root(sch, block):
             return [sch]
         new_sch = sch.copy()
@@ -227,7 +226,7 @@ class ReorderScheduleRule(PyScheduleRule):
     def _initialize_with_tune_context(self, context: "TuneContext") -> None:
         pass
 
-    def apply(self, sch: Schedule, block: SBlockRV) -> List[Schedule]:
+    def apply(self, sch: Schedule, block: SBlockRV) -> list[Schedule]:
         if _is_root(sch, block):
             return [sch]
         new_sch = sch.copy()
@@ -326,7 +325,7 @@ class RemoveBlock(PyScheduleRule):
         def _initialize_with_tune_context(self, context: "TuneContext") -> None:
             pass
 
-        def apply(self, sch: Schedule, block: SBlockRV) -> List[Schedule]:
+        def apply(self, sch: Schedule, block: SBlockRV) -> list[Schedule]:
             if _is_root(sch, block):
                 return [sch]
             sch = sch.copy()
diff --git a/tests/python/s_tir/meta_schedule/test_meta_schedule_runner.py b/tests/python/s_tir/meta_schedule/test_meta_schedule_runner.py
index cecd561f9d2c..6ad493dde0a4 100644
--- a/tests/python/s_tir/meta_schedule/test_meta_schedule_runner.py
+++ b/tests/python/s_tir/meta_schedule/test_meta_schedule_runner.py
@@ -20,7 +20,7 @@
 import itertools
 import sys
 import time
-from typing import Any, List
+from typing import Any
 
 import numpy as np
 import pytest
@@ -384,7 +384,7 @@ def test_meta_schedule_py_runner():
 
     @derived_object
     class TestRunner(PyRunner):
-        def run(self, runner_inputs: List[RunnerInput]) -> List[RunnerFuture]:
+        def run(self, runner_inputs: list[RunnerInput]) -> list[RunnerFuture]:
             raise ValueError("TestRunner")
 
     runner = TestRunner()
@@ -596,8 +596,8 @@ def test_meta_schedule_runner_matmul_test():
     """Test meta schedule runner with add module"""
 
     def _check_correct_matmul(
-        args_before: List[np.ndarray],
-        args_after: List[np.ndarray],
+        args_before: list[np.ndarray],
+        args_after: list[np.ndarray],
     ) -> None:
         a_before, b_before, c_before = args_before
         a_after, b_after, c_after = args_after
@@ -611,7 +611,7 @@ def test_alloc_argument(
         device: Device,
         args_info: Any,
         alloc_repeat: int,
-    ) -> List[Any]:
+    ) -> list[Any]:
         global repeated_args_before  # pylint: disable=global-variable-undefined, invalid-name
         repeated_args_before = []  # type: ignore
         repeated_args = rpc_default_alloc_argument(session, device, args_info, alloc_repeat)
@@ -624,8 +624,8 @@ def test_run_evaluator(
         rt_mod: Module,
         device: Device,
         evaluator_config: EvaluatorConfig,
-        repeated_args: List[Any],
-    ) -> List[float]:
+        repeated_args: list[Any],
+    ) -> list[float]:
         global repeated_args_before  # pylint: disable=global-variable-undefined, invalid-name
         repeated_args_after = []
         evaluator = rt_mod.time_evaluator(
@@ -638,7 +638,7 @@ def test_run_evaluator(
                 "cache_flush_cpu_non_first_arg" if evaluator_config.enable_cpu_cache_flush else ""
             ),
         )
-        repeated_costs: List[List[float]] = []
+        repeated_costs: list[list[float]] = []
         for args in repeated_args:
             device.sync()
             profile_result = evaluator(*args)
@@ -705,7 +705,7 @@ def test_run_evaluator(
 def test_meta_schedule_runner_add_test():
     """Test meta schedule runner with add module"""
 
-    def _check_correct_add(args_before: List[np.ndarray], args_after: List[np.ndarray]) -> None:
+    def _check_correct_add(args_before: list[np.ndarray], args_after: list[np.ndarray]) -> None:
         a_before, b_before, c_before = args_before
         a_after, b_after, c_after = args_after
         c_before = a_before + b_before
@@ -718,7 +718,7 @@ def test_alloc_argument(
         device: Device,
         args_info: Any,
         alloc_repeat: int,
-    ) -> List[Any]:
+    ) -> list[Any]:
         global repeated_args_before  # pylint: disable=global-variable-undefined, invalid-name
         repeated_args_before = []  # type: ignore
         repeated_args = rpc_default_alloc_argument(
@@ -736,8 +736,8 @@ def test_run_evaluator(
         rt_mod: Module,
         device: Device,
         evaluator_config: EvaluatorConfig,
-        repeated_args: List[Any],
-    ) -> List[float]:
+        repeated_args: list[Any],
+    ) -> list[float]:
         global repeated_args_before  # pylint: disable=global-variable-undefined, invalid-name
         repeated_args_after = []
         evaluator = rt_mod.time_evaluator(
@@ -750,7 +750,7 @@ def test_run_evaluator(
                 "cache_flush_cpu_non_first_arg" if evaluator_config.enable_cpu_cache_flush else ""
             ),
         )
-        repeated_costs: List[List[float]] = []
+        repeated_costs: list[list[float]] = []
         for args in repeated_args:
             device.sync()
             profile_result = evaluator(*args)
@@ -817,7 +817,7 @@ def test_run_evaluator(
 def test_meta_schedule_local_runner_add_test():
     """Test meta schedule local runner with add module"""
 
-    def _check_correct_add(args_before: List[np.array], args_after: List[np.array]) -> None:
+    def _check_correct_add(args_before: list[np.array], args_after: list[np.array]) -> None:
         a_before, b_before, c_before = args_before
         a_after, b_after, c_after = args_after
         c_before = a_before + b_before
@@ -829,7 +829,7 @@ def test_alloc_argument(
         device: Device,
         args_info: T_ARG_INFO_JSON_OBJ_LIST,  # pylint: disable=unused-argument
         alloc_repeat: int,
-    ) -> List[T_ARGUMENT_LIST]:
+    ) -> list[T_ARGUMENT_LIST]:
         global repeated_args_before  # pylint: disable=global-variable-undefined, invalid-name
         repeated_args_before = []
         repeated_args = local_default_alloc_argument(device, args_info, alloc_repeat)
@@ -841,8 +841,8 @@ def test_run_evaluator(
         rt_mod: Module,
         device: Device,
         evaluator_config: EvaluatorConfig,
-        repeated_args: List[Any],
-    ) -> List[float]:
+        repeated_args: list[Any],
+    ) -> list[float]:
         global repeated_args_before  # pylint: disable=global-variable-undefined, invalid-name
         repeated_args_after = []
         evaluator = rt_mod.time_evaluator(
@@ -855,7 +855,7 @@ def test_run_evaluator(
                 "cache_flush_cpu_non_first_arg" if evaluator_config.enable_cpu_cache_flush else ""
             ),
         )
-        repeated_costs: List[List[float]] = []
+        repeated_costs: list[list[float]] = []
         for args in repeated_args:
             device.sync()
             profile_result = evaluator(*args)
diff --git a/tests/python/s_tir/meta_schedule/test_meta_schedule_schedule_rule_apply_custom_rule.py b/tests/python/s_tir/meta_schedule/test_meta_schedule_schedule_rule_apply_custom_rule.py
index 78646a1886dc..eee51ae93bb2 100644
--- a/tests/python/s_tir/meta_schedule/test_meta_schedule_schedule_rule_apply_custom_rule.py
+++ b/tests/python/s_tir/meta_schedule/test_meta_schedule_schedule_rule_apply_custom_rule.py
@@ -16,7 +16,6 @@
 # under the License.
 # pylint: disable=missing-module-docstring,missing-function-docstring,missing-class-docstring
 import tempfile
-from typing import List
 
 import pytest
 
@@ -44,7 +43,7 @@ def main(a: T.handle, b: T.handle, c: T.handle) -> None:
 
 
 @tvm.register_global_func("s_tir.meta_schedule.cpu.test_apply_custom_rule")
-def sch_fn(sch: tvm.s_tir.Schedule, block: tvm.tir.SBlock) -> List[tvm.s_tir.Schedule]:
+def sch_fn(sch: tvm.s_tir.Schedule, block: tvm.tir.SBlock) -> list[tvm.s_tir.Schedule]:
     raise ValueError("Intended for s_tir.meta_schedule.cpu.test_apply_custom_rule")
 
 
diff --git a/tests/python/s_tir/meta_schedule/test_meta_schedule_search_strategy.py b/tests/python/s_tir/meta_schedule/test_meta_schedule_search_strategy.py
index bfcd808105d0..83e943609627 100644
--- a/tests/python/s_tir/meta_schedule/test_meta_schedule_search_strategy.py
+++ b/tests/python/s_tir/meta_schedule/test_meta_schedule_search_strategy.py
@@ -18,7 +18,6 @@
 """Test Meta Schedule SearchStrategy"""
 
 # pylint: disable=missing-function-docstring
-from typing import List
 
 import pytest
 
@@ -101,11 +100,11 @@ def test_meta_schedule_replay_func(
     (correct_sch,) = ms.space_generator.ScheduleFn(sch_fn=_schedule_matmul).generate_design_space(
         Matmul
     )
-    num_trials_each_iter: List[int] = []
+    num_trials_each_iter: list[int] = []
     candidates = strategy.generate_measure_candidates()
     while candidates is not None:
         num_trials_each_iter.append(len(candidates))
-        runner_results: List[ms.runner.RunnerResult] = []
+        runner_results: list[ms.runner.RunnerResult] = []
         for candidate in candidates:
             _is_trace_equal(
                 candidate.sch,
@@ -167,11 +166,11 @@ def _schedule_matmul_small(sch: Schedule):
         database=ms.database.MemoryDatabase(),
         cost_model=ms.cost_model.RandomModel(),
     )
-    num_trials_each_iter: List[int] = []
+    num_trials_each_iter: list[int] = []
     candidates = strategy.generate_measure_candidates()
     while candidates is not None:
         num_trials_each_iter.append(len(candidates))
-        runner_results: List[ms.runner.RunnerResult] = []
+        runner_results: list[ms.runner.RunnerResult] = []
         for candidate in candidates:
             _is_trace_equal(
                 candidate.sch,
@@ -232,11 +231,11 @@ def _schedule_matmul_empty(sch: Schedule):
         database=ms.database.MemoryDatabase(),
         cost_model=ms.cost_model.RandomModel(),
     )
-    num_trials_each_iter: List[int] = []
+    num_trials_each_iter: list[int] = []
     candidates = strategy.generate_measure_candidates()
     while candidates is not None:
         num_trials_each_iter.append(len(candidates))
-        runner_results: List[ms.runner.RunnerResult] = []
+        runner_results: list[ms.runner.RunnerResult] = []
         for candidate in candidates:
             _is_trace_equal(
                 candidate.sch,
diff --git a/tests/python/s_tir/meta_schedule/test_meta_schedule_space_cuda_async_multiple_initialization.py b/tests/python/s_tir/meta_schedule/test_meta_schedule_space_cuda_async_multiple_initialization.py
index ca792004ab27..dc169946578c 100644
--- a/tests/python/s_tir/meta_schedule/test_meta_schedule_space_cuda_async_multiple_initialization.py
+++ b/tests/python/s_tir/meta_schedule/test_meta_schedule_space_cuda_async_multiple_initialization.py
@@ -16,10 +16,8 @@
 # under the License.
 """Tests for MetaSchedule search space on CUDA"""
 
-from typing import List, Optional, Tuple, Union
-
 # isort: off
-from typing_extensions import Literal
+from typing import Literal
 
 # isort: on
 from tvm.ir import IRModule
@@ -34,10 +32,10 @@ def generate_design_space(
     kind: Literal["llvm", "cuda", "cuda-tensorcore", "hexagon"],
     mod: IRModule,
     target: Target,
-    types: Union[type, Tuple[type, ...]],
-    sch_rules: Optional[List[ms.ScheduleRule]] = None,
+    types: type | tuple[type, ...],
+    sch_rules: list[ms.ScheduleRule] | None = None,
     initialize_time: int = 1,
-) -> List[Schedule]:
+) -> list[Schedule]:
     if sch_rules is None:
         sch_rules = get_rules(kind, types)
     else:
diff --git a/tests/python/s_tir/meta_schedule/test_meta_schedule_task_scheduler.py b/tests/python/s_tir/meta_schedule/test_meta_schedule_task_scheduler.py
index 3c5bf3e2dc83..e2c1e6314f1e 100644
--- a/tests/python/s_tir/meta_schedule/test_meta_schedule_task_scheduler.py
+++ b/tests/python/s_tir/meta_schedule/test_meta_schedule_task_scheduler.py
@@ -19,7 +19,6 @@
 
 import random
 import weakref
-from typing import Set
 
 import pytest
 
@@ -122,7 +121,7 @@ def _schedule_batch_matmul(sch: Schedule):
 
 @ms.derived_object
 class MyTaskScheduler(ms.task_scheduler.PyTaskScheduler):
-    done: Set = set()
+    done: set = set()
 
     def next_task_id(self) -> int:
         tasks = self._outer().tasks_
diff --git a/tests/python/s_tir/schedule/test_tir_schedule_analysis.py b/tests/python/s_tir/schedule/test_tir_schedule_analysis.py
index ea68c1a96dff..3c4e75d15ebd 100644
--- a/tests/python/s_tir/schedule/test_tir_schedule_analysis.py
+++ b/tests/python/s_tir/schedule/test_tir_schedule_analysis.py
@@ -16,7 +16,6 @@
 # under the License.
 # pylint: disable=missing-docstring
 # ruff: noqa: F401, F841
-from typing import List
 
 import pytest
 
@@ -53,11 +52,11 @@
 from tvm.tir.stmt_functor import pre_order_visit
 
 
-def _make_vars(*args: str) -> List[Var]:
+def _make_vars(*args: str) -> list[Var]:
     return [Var(arg, dtype="int32") for arg in args]
 
 
-def _make_loops(loop_vars: List[Var], extents: List[int]) -> List[For]:
+def _make_loops(loop_vars: list[Var], extents: list[int]) -> list[For]:
     assert len(loop_vars) == len(extents)
     return [
         For(
diff --git a/tests/python/testing/test_type_annotation_checker.py b/tests/python/testing/test_type_annotation_checker.py
index efe9b128a00b..b5d2afcb92be 100644
--- a/tests/python/testing/test_type_annotation_checker.py
+++ b/tests/python/testing/test_type_annotation_checker.py
@@ -18,7 +18,8 @@
 """Test type checker based on python's type annotations"""
 
 import sys
-from typing import Callable, Dict, List, Tuple, Union
+from collections.abc import Callable
+from typing import Union
 
 import _pytest
 import pytest
@@ -42,7 +43,7 @@ def str_func(x: str) -> str:
         "negative_cases": ["5"],
     },
     {
-        "type_annotation": List[int],
+        "type_annotation": list[int],
         "positive_cases": [
             [5],
             [],
@@ -57,14 +58,14 @@ def str_func(x: str) -> str:
         ],
     },
     {
-        "type_annotation": Dict[str, int],
+        "type_annotation": dict[str, int],
         "positive_cases": [
             {"key1": 0, "key2": 1, "key3": -1},
         ],
         "negative_cases": [None, [1], {1: "1"}],
     },
     {
-        "type_annotation": Tuple[int],
+        "type_annotation": tuple[int],
         "positive_cases": [
             (5,),
         ],
@@ -77,7 +78,7 @@ def str_func(x: str) -> str:
         ],
     },
     {
-        "type_annotation": Tuple[str, int],
+        "type_annotation": tuple[str, int],
         "positive_cases": [
             ("x", 5),
         ],
@@ -90,7 +91,7 @@ def str_func(x: str) -> str:
         ],
     },
     {
-        "type_annotation": Union[str, int],
+        "type_annotation": str | int,
         "positive_cases": [
             "x",
             5,
@@ -191,14 +192,14 @@ def func(_: type_annotation):
 @pytest.mark.parametrize(
     ["type_annotation", "expected_key", "expected_subtypes"],
     [
-        pytest.param(Union[str, int], "union", [str, int], id="Union[str, int]"),
-        pytest.param(List[str], "list", [str], id="List[str]"),
-        pytest.param(Dict[str, int], "dict", [str, int], id="Dict[str, int]"),
-        pytest.param(Tuple[str, int], "tuple", (str, int), id="Tuple[str, int]"),
+        pytest.param(str | int, "union", [str, int], id="str | int"),
+        pytest.param(list[str], "list", [str], id="List[str]"),
+        pytest.param(dict[str, int], "dict", [str, int], id="Dict[str, int]"),
+        pytest.param(tuple[str, int], "tuple", (str, int), id="Tuple[str, int]"),
         pytest.param(
-            Union[List[str], Dict[str, int]],
+            list[str] | dict[str, int],
             "union",
-            [List[str], Dict[str, int]],
+            [list[str], dict[str, int]],
             id="Union[List[str], Dict[str, int]]",
         ),
     ],
diff --git a/tests/python/tir-base/test_tir_imm_values.py b/tests/python/tir-base/test_tir_imm_values.py
index ae5f13bd8df7..de5798a31687 100644
--- a/tests/python/tir-base/test_tir_imm_values.py
+++ b/tests/python/tir-base/test_tir_imm_values.py
@@ -78,7 +78,7 @@ def test_tir_large_py_int_literals(dtype, literals):
     """
     for l in literals:
         x = tir.const(l, dtype)
-        if isinstance(x, (tir.IntImm, tir.FloatImm)):
+        if isinstance(x, tir.IntImm | tir.FloatImm):
             assert x.value == literals[l]
         else:
             # LargeUIntImm(low32, hi32)
@@ -208,7 +208,7 @@ def check_tir_const_fold(
 
     if x_range is None:
         x_range = (ninfo.min, ninfo.max)
-    if isinstance(x_range, (int, float)):
+    if isinstance(x_range, int | float):
         x = x_range
     elif dtype.startswith("int") or dtype.startswith("uint"):
         x = np.random.randint(x_range[0], x_range[1] + 1, dtype=dtype)
@@ -217,7 +217,7 @@ def check_tir_const_fold(
 
     if y_range is None:
         y_range = (ninfo.min, ninfo.max)
-    if isinstance(y_range, (int, float)):
+    if isinstance(y_range, int | float):
         y = y_range
     elif dtype.startswith("int") or dtype.startswith("uint"):
         y = np.random.randint(y_range[0], y_range[1] + 1, dtype=dtype)
@@ -226,7 +226,7 @@ def check_tir_const_fold(
 
     if skip_overflow:
         py_res = foldf(x, y)
-        if isinstance(py_res, (tir.IntImm, tir.FloatImm)):
+        if isinstance(py_res, tir.IntImm | tir.FloatImm):
             py_res = py_res.value
         if not (ninfo.min <= py_res <= ninfo.max):
             # If the result overflow, certain arithmetics is non-defined
diff --git a/tests/python/tir-base/test_tir_ops.py b/tests/python/tir-base/test_tir_ops.py
index fb67980b1924..9eeac7030164 100644
--- a/tests/python/tir-base/test_tir_ops.py
+++ b/tests/python/tir-base/test_tir_ops.py
@@ -34,7 +34,7 @@ def test_const_fold():
     def check(f, *args):
         x = f(*[tvm.tir.const(x, "int32") for x in args])
         y = f(*args)
-        if not isinstance(x, (tvm.tir.IntImm,)) or x.value != int(y):
+        if not isinstance(x, tvm.tir.IntImm) or x.value != int(y):
             raise ValueError(f"check error: {x} vs {y} ")
 
     tmod = tvm.tir.truncmod
diff --git a/tests/python/tvmscript/test_tvmscript_printer_underlining.py b/tests/python/tvmscript/test_tvmscript_printer_underlining.py
index 6901e86e14c7..f6818b4c935f 100644
--- a/tests/python/tvmscript/test_tvmscript_printer_underlining.py
+++ b/tests/python/tvmscript/test_tvmscript_printer_underlining.py
@@ -15,7 +15,6 @@
 # specific language governing permissions and limitations
 # under the License.
 
-from typing import Optional
 
 import pytest
 from tvm_ffi.access_path import AccessPath
@@ -36,7 +35,7 @@ def make_path(name: str) -> AccessPath:
     return AccessPath.root().attr(name)
 
 
-def make_id_doc(name: str, path_name: Optional[str] = None) -> IdDoc:
+def make_id_doc(name: str, path_name: str | None = None) -> IdDoc:
     if path_name is None:
         path_name = name
     doc = IdDoc(name)
diff --git a/tests/scripts/ci.py b/tests/scripts/ci.py
index dec9588687c4..c0ef2f09cbfe 100755
--- a/tests/scripts/ci.py
+++ b/tests/scripts/ci.py
@@ -33,8 +33,9 @@
 import sys
 import textwrap
 import typing
+from collections.abc import Callable
 from pathlib import Path
-from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+from typing import Any
 
 REPO_ROOT = Path(__file__).resolve().parent.parent.parent
 SCRIPT_DIR = REPO_ROOT / ".ci-py-scripts"
@@ -60,7 +61,7 @@ def print_color(color: str, msg: str, bold: bool, **kwargs: Any) -> None:
         print(msg, **kwargs)
 
 
-warnings: List[str] = []
+warnings: list[str] = []
 
 
 def clean_exit(msg: str) -> None:
@@ -72,7 +73,7 @@ def clean_exit(msg: str) -> None:
     exit(1)
 
 
-def cmd(commands: List[Any], **kwargs: Any):
+def cmd(commands: list[Any], **kwargs: Any):
     commands = [str(s) for s in commands]
     command_str = " ".join(commands)
     print_color(col.BLUE, command_str, bold=True)
@@ -149,10 +150,10 @@ def gen_name(s: str) -> str:
 def docker(
     name: str,
     image: str,
-    scripts: List[str],
-    env: Dict[str, str],
+    scripts: list[str],
+    env: dict[str, str],
     interactive: bool,
-    additional_flags: Optional[Dict[str, str]] = None,
+    additional_flags: dict[str, str] | None = None,
 ):
     """
     Invoke a set of bash scripts through docker/bash.sh
@@ -231,12 +232,12 @@ def docker(
 
 
 def docs(
-    tutorial_pattern: Optional[str] = None,
+    tutorial_pattern: str | None = None,
     cpu: bool = False,
     full: bool = False,
     interactive: bool = False,
     skip_build: bool = False,
-    docker_image: Optional[str] = None,
+    docker_image: str | None = None,
 ) -> None:
     """
     Build the documentation from gallery/ and docs/. By default this builds only
@@ -332,7 +333,7 @@ def serve_docs(directory: str = "_docs") -> None:
     cmd([sys.executable, "-m", "http.server"], cwd=directory_path)
 
 
-def lint(interactive: bool = False, fix: bool = False, docker_image: Optional[str] = None) -> None:
+def lint(interactive: bool = False, fix: bool = False, docker_image: str | None = None) -> None:
     """
     Run CI's Sanity Check step
 
@@ -355,17 +356,17 @@ def lint(interactive: bool = False, fix: bool = False, docker_image: Optional[st
     )
 
 
-Option = Tuple[str, List[str]]
+Option = tuple[str, list[str]]
 
 
 def generate_command(
     name: str,
-    options: Dict[str, Option],
+    options: dict[str, Option],
     help: str,
-    precheck: Optional[Callable[[], None]] = None,
-    post_build: Optional[List[str]] = None,
-    additional_flags: Optional[Dict[str, str]] = None,
-    env: Optional[Dict[str, str]] = None,
+    precheck: Callable[[], None] | None = None,
+    post_build: list[str] | None = None,
+    additional_flags: dict[str, str] | None = None,
+    env: dict[str, str] | None = None,
 ):
     """
     Helper to generate CLIs that:
@@ -376,10 +377,10 @@ def generate_command(
     """
 
     def fn(
-        tests: Optional[List[str]],
+        tests: list[str] | None,
         skip_build: bool = False,
         interactive: bool = False,
-        docker_image: Optional[str] = None,
+        docker_image: str | None = None,
         verbose: bool = False,
         **kwargs,
     ) -> None:
@@ -494,8 +495,8 @@ def is_optional_type(annotation):
 def add_subparser(
     func: Callable,
     subparsers: Any,
-    options: Optional[Dict[str, Option]] = None,
-    help: Optional[str] = None,
+    options: dict[str, Option] | None = None,
+    help: str | None = None,
 ) -> Any:
     """
     Utility function to make it so subparser commands can be defined locally
@@ -534,7 +535,7 @@ def add_subparser(
             continue
 
         arg_cli_name = cli_name(name)
-        kwargs: Dict[str, Union[str, bool]] = {"help": arg_help_texts[arg_cli_name]}
+        kwargs: dict[str, str | bool] = {"help": arg_help_texts[arg_cli_name]}
 
         is_optional = is_optional_type(value.annotation)
         if is_optional:
diff --git a/tests/scripts/release/gather_prs.py b/tests/scripts/release/gather_prs.py
index 96182bcab3e6..16829c497a18 100644
--- a/tests/scripts/release/gather_prs.py
+++ b/tests/scripts/release/gather_prs.py
@@ -22,8 +22,9 @@
 import os
 import pickle
 import sys
+from collections.abc import Callable
 from pathlib import Path
-from typing import Any, Callable, Dict, List
+from typing import Any
 
 REPO_ROOT = Path(__file__).resolve().parent.parent.parent.parent
 sys.path.append(str(REPO_ROOT / "ci" / "scripts" / "jenkins"))
@@ -137,7 +138,7 @@ def fetch_pr_data(args, cache):
 
 
 def write_csv(
-    filename: str, data: List[Dict[str, Any]], threshold_filter: Callable[[Dict[str, Any]], bool]
+    filename: str, data: list[dict[str, Any]], threshold_filter: Callable[[dict[str, Any]], bool]
 ) -> None:
     with open(filename, "w", newline="") as csvfile:
         writer = csv.writer(csvfile, quotechar='"')