Skip to content

perf: flydsl gdr decode default on #752

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
xytpai wants to merge 5 commits into
base: main
Choose a base branch
from
Open

perf: flydsl gdr decode default on #752

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
416f1b7
ATOM_USE_FLYDSL_GDR default ON
xytpai May 11, 2026
b7657d6
Update attention_gdn.py
xytpai May 12, 2026
539f31a
Merge branch 'main' into xyt/flydsl_gdr_on
xytpai May 12, 2026
b665af8
Merge branch 'main' into xyt/flydsl_gdr_on
xytpai May 13, 2026
c5b4024
Merge branch 'main' into xyt/flydsl_gdr_on
xytpai May 14, 2026
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
4 changes: 2 additions & 2 deletions atom/plugin/vllm/attention_backend/attention_gdn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -395,8 +395,8 @@ def forward(
query=query_non_spec,
key=key_non_spec,
value=value_non_spec,
a=a,
b=b,
a=a.unsqueeze(1),
b=b.unsqueeze(1),
dt_bias=self.dt_bias,
A_log=self.A_log,
indices=non_spec_state_indices_tensor,
Expand Down
2 changes: 1 addition & 1 deletion atom/utils/envs.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -89,7 +89,7 @@
"ATOM_USE_CUSTOM_ALL_GATHER": lambda: (
os.getenv("ATOM_USE_CUSTOM_ALL_GATHER", "1").lower() == "1"
),
"ATOM_USE_FLYDSL_GDR": lambda: os.getenv("ATOM_USE_FLYDSL_GDR", "0").lower() == "1",
"ATOM_USE_FLYDSL_GDR": lambda: os.getenv("ATOM_USE_FLYDSL_GDR", "1").lower() == "1",
# --- MoE (DeepSeek-style shared experts) ---
# Dual-stream MoE only when num_tokens <= threshold; 0 disables dual-stream registration.
"ATOM_DUAL_STREAM_MOE_TOKEN_THRESHOLD": lambda: int(
Expand Down
Loading