Skip to content

rocm/ci: disable gradient accumulation fusion for GLM-4-9B on ROCm #1166

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
sreerohi wants to merge 1 commit into
base: main
Choose a base branch
from
Open

rocm/ci: disable gradient accumulation fusion for GLM-4-9B on ROCm #1166

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions tests/e2e/megatron/test_quick_start_glm4_9B.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,7 @@
import os

import torch

from tests.ci.ci_register import register_cuda_ci

import miles.utils.external_utils.command_utils as U
Expand All @@ -8,6 +10,7 @@

ENABLE_EVAL = U.get_bool_env_var("MILES_TEST_ENABLE_EVAL", "1")
TIGHT_DEVICE_MEMORY = U.get_bool_env_var("MILES_TEST_TIGHT_DEVICE_MEMORY", "1")
IS_ROCM = hasattr(torch.version, "hip") and torch.version.hip is not None
Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist Bot May 21, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

medium

For consistency with other parts of the codebase (e.g., miles/backends/megatron_utils/model.py line 809, which uses if torch.version.hip:) and for conciseness, you can simplify this check. The torch.version.hip attribute is guaranteed to exist and will be None on non-ROCm builds, so the hasattr check is redundant.

Suggested change
IS_ROCM = hasattr(torch.version, "hip") and torch.version.hip is not None
IS_ROCM = torch.version.hip is not None
References
  1. Maintain consistency with existing patterns in the repository for checking ROCm availability and avoid redundant hasattr checks for attributes guaranteed to exist in the environment. (link)


MODEL_NAME = "GLM-Z1-9B-0414"
MODEL_TYPE = "glm4-9B"
Expand Down Expand Up @@ -103,6 +106,9 @@ def execute():
"--rollout-num-gpus 4 "
)

if IS_ROCM:
misc_args += "--no-gradient-accumulation-fusion "

train_args = (
f"{ckpt_args} "
f"{rollout_args} "
Expand Down