Skip to content

Fix Flash Attention 3 interface for new FA3 return format #13173

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
DN6 merged 2 commits into from
Feb 26, 2026
+10 −2
Merged

Fix Flash Attention 3 interface for new FA3 return format #13173

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d7e2c07
Fix Flash Attention 3 interface compatibility for new FA3 versions
veeceey Feb 23, 2026
c0961d5
Simplify _wrapped_flash_attn_3 return unpacking
veeceey Feb 24, 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
19 changes: 16 additions & 3 deletions src/diffusers/models/attention_dispatch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -676,7 +676,7 @@ def _wrapped_flash_attn_3(
) -> tuple[torch.Tensor, torch.Tensor]:
# Hardcoded for now because pytorch does not support tuple/int type hints
window_size = (-1, -1)
out, lse, *_ = flash_attn_3_func(
result = flash_attn_3_func(
q=q,
k=k,
v=v,
Expand All @@ -693,8 +693,15 @@ def _wrapped_flash_attn_3(
pack_gqa=pack_gqa,
deterministic=deterministic,
sm_margin=sm_margin,
return_attn_probs=True,
)
lse = lse.permute(0, 2, 1)
# Handle both old FA3 (always returns tuple) and new FA3 (returns tuple only with return_attn_probs=True)
if isinstance(result, tuple):
out, lse, *_ = result
lse = lse.permute(0, 2, 1)
else:
out = result
lse = torch.empty(q.shape[0], q.shape[2], q.shape[1], device=q.device, dtype=torch.float32)
Copy link
Copy Markdown
Collaborator

@DN6 DN6 Feb 23, 2026

Choose a reason for hiding this comment

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

Don't think we need this guard. In both cases (old vs new FA3) we're always returning a tuple since return_attn_probs=True? Why not just leave as out, lse, *_ = flash_attn_3_func

return out, lse


Expand Down Expand Up @@ -2623,7 +2630,7 @@ def _flash_varlen_attention_3(
key_packed = torch.cat(key_valid, dim=0)
value_packed = torch.cat(value_valid, dim=0)

out, lse, *_ = flash_attn_3_varlen_func(
result = flash_attn_3_varlen_func(
q=query_packed,
k=key_packed,
v=value_packed,
Expand All @@ -2633,7 +2640,13 @@ def _flash_varlen_attention_3(
max_seqlen_k=max_seqlen_k,
softmax_scale=scale,
causal=is_causal,
return_attn_probs=return_lse,
)
if isinstance(result, tuple):
out, lse, *_ = result
else:
out = result
lse = None
out = out.unflatten(0, (batch_size, -1))

return (out, lse) if return_lse else out
Expand Down