Skip to content

Fixed Flash Attention TypeError error "only integer tensors of a single element can be converted to an index" #1318

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

Fixed Flash Attention TypeError error "only integer tensors of a single element can be converted to an index" #1318

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
9 changes: 7 additions & 2 deletions wanvideo/modules/attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -101,8 +101,13 @@ def half(x):
[lk] * b, dtype=torch.int32).to(
device=k.device, non_blocking=True)
else:
k = half(torch.cat([u[:v] for u, v in zip(k, k_lens)]))
v = half(torch.cat([u[:v] for u, v in zip(v, k_lens)]))
# 20250927 pftq: Convert k_lens to a single tensor to avoid 'list' object has no attribute 'new_zeros' and optimize performance
k_lens_tensor = torch.tensor(k_lens, dtype=torch.int32, device=k.device) if not isinstance(k_lens, torch.Tensor) else k_lens.to(dtype=torch.int32, device=k.device)
k = half(torch.cat([u[:v] for u, v in zip(k, k_lens_tensor)])) # Use tensor directly
#k = half(torch.cat([u[:v.item()] for u, v in zip(k, k_lens)])) # original line
v = half(torch.cat([u[:v] for u, v in zip(v, k_lens_tensor)])) # Use tensor directly
#v = half(torch.cat([u[:v.item()] for u, v in zip(v, k_lens)])) # original line
k_lens = k_lens_tensor # 20250927 pftq: Reuse k_lens_tensor for cu_seqlens_k

q = q.to(v.dtype)
k = k.to(v.dtype)
Expand Down