Skip to content

[Flex Attn] Fix lse x attention sinks logic #41249

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

Merged
merged 1 commit into from
Oct 2, 2025
Merged

[Flex Attn] Fix lse x attention sinks logic #41249

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: 3 additions & 3 deletions src/transformers/integrations/flex_attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -282,10 +282,10 @@ def score_mod(score, batch_idx, head_idx, q_idx, kv_idx):
# On CPU we must skip returning LSE due to a runtime issue; elsewhere, follow PyTorch API and return it
return_lse = query.device.type != "cpu"

# Validate that s_aux is not silently ignored
if not return_lse and s_aux is not None:
logger.warning_once("s_aux provided with return_lse=False - forcing return_lse=True to avoid silent failure")
return_lse = True
raise ValueError(
"Attention sinks cannot be run on CPU with flex attention. Please switch to a different device, e.g. CUDA"
)

flex_attention_output = compile_friendly_flex_attention(
query,
Expand Down