Skip to content

[DSA] Fix the pooled-indexer breakable prefill bridge under DP attention #41311

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
kpham-sgl merged 2 commits into from
Sep 26, 2026
Merged

[DSA] Fix the pooled-indexer breakable prefill bridge under DP attention #41311

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ab5a769
Fix the pooled-indexer breakable prefill bridge under DP attention
Sep 26, 2026
6de44c6
Bound the pooled-indexer bridge rows and cover DP + breakable prefill…
Sep 26, 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
18 changes: 14 additions & 4 deletions python/sglang/srt/layers/attention/dsa/kpool_prefill_cuda_graph.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,11 +37,21 @@ def _kpool_indexer_prefill_with_output(
)
if not return_indices:
return
if result is None or result.shape != (n, output.shape[1]):
raise ValueError("Pooled-indexer prefill returned an unexpected top-k shape")
num_logical = sum(forward_batch.extend_seq_lens_cpu)
if (
result is None
or result.ndim != 2
or not num_logical <= result.shape[0] <= n
or result.shape[1] != output.shape[1]
):
raise ValueError(
"Pooled-indexer prefill returned an unexpected top-k shape: got "
f"{None if result is None else tuple(result.shape)}, expected "
f"between {num_logical} and {n} rows of width {output.shape[1]}"
)
# The following captured attention segment reads this stable padded buffer.
output[:n].copy_(result)
output[n:].fill_(-1)
output[:num_logical].copy_(result[:num_logical])
output[num_logical:].fill_(-1)


def _kpool_indexer_prefill_capture_stub(
Expand Down
2 changes: 2 additions & 0 deletions test/registered/e2e/models/test_glm53_flash_b200.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -108,6 +108,8 @@ class TestGLM53FlashB200HighThroughput(
"--enable-dp-attention",
"--dp-size",
"4",
"--cuda-graph-backend-prefill",
"breakable",
Comment on lines +111 to +112

@kpham-sgl kpham-sgl Sep 26, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

why add this? I think GLM 5.3 Flash hasn't supported BCG yet?

@mmangkad mmangkad Sep 26, 2026

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

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

See #38522

]


Expand Down
Loading