Skip to content

[BugFix] Fix flash_attn_with_kvcache with scalar cache_seqlen #1795

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
tridao merged 1 commit into from
Aug 15, 2025
+2 −2
Merged

[BugFix] Fix flash_attn_with_kvcache with scalar cache_seqlen #1795

Show file tree
Hide file tree
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
2 changes: 1 addition & 1 deletion flash_attn/flash_attn_interface.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1576,7 +1576,7 @@ def flash_attn_with_kvcache(
softmax_scale = q.shape[-1] ** (-0.5)
if cache_seqlens is not None and isinstance(cache_seqlens, int):
cache_seqlens = torch.full(
(k_cache.shape[0],), cache_seqlens, dtype=torch.int32, device=k_cache.device
(q.shape[0],), cache_seqlens, dtype=torch.int32, device=k_cache.device
)
cache_seqlens = maybe_contiguous(cache_seqlens)
cache_batch_idx = maybe_contiguous(cache_batch_idx)
Expand Down
2 changes: 1 addition & 1 deletion hopper/flash_attn_interface.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -751,7 +751,7 @@ def flash_attn_with_kvcache(
softmax_scale = (q.shape[-1] + (qv.shape[-1] if qv is not None else 0)) ** (-0.5)
if cache_seqlens is not None and isinstance(cache_seqlens, int):
cache_seqlens = torch.full(
(k_cache.shape[0],), cache_seqlens, dtype=torch.int32, device=k_cache.device
(q.shape[0],), cache_seqlens, dtype=torch.int32, device=k_cache.device
)
cache_seqlens = maybe_contiguous(cache_seqlens)
out, softmax_lse, *rest = _flash_attn_forward(
Expand Down