Skip to content

make trtllm attn backend's init_forward_metadat non blocking #13802

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
Fridge003 merged 1 commit into from
Nov 23, 2025
Merged

make trtllm attn backend's init_forward_metadat non blocking #13802

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
4 changes: 3 additions & 1 deletion python/sglang/srt/layers/attention/trtllm_mla_backend.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -564,7 +564,9 @@ def init_forward_metadata(self, forward_batch: ForwardBatch):
seq_lens = forward_batch.seq_lens - forward_batch.extend_prefix_lens
cum_seq_lens_q = torch.cat(
(
torch.tensor([0], device=forward_batch.seq_lens.device),
torch.zeros(
1, dtype=torch.int32, device=forward_batch.seq_lens.device
),
Comment on lines +567 to +569
Copy link
Contributor

@gemini-code-assist gemini-code-assist bot Nov 23, 2025

Choose a reason for hiding this comment

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

medium

While this change is a good optimization, it also makes the .int() call on line 572 redundant. Since torch.zeros now explicitly creates an int32 tensor, and torch.cumsum on an int32 tensor also returns an int32 tensor, the result of torch.cat will already be int32. The .int() call can be removed for a minor cleanup.

torch.cumsum(seq_lens, dim=0),
)
).int()
Expand Down
Loading