Skip to content

Fix hybrid_linear_attn_backend crash with ngram speculation #20739

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
hnyls2002 merged 5 commits into from
Apr 8, 2026
+4 −3
Merged

Fix hybrid_linear_attn_backend crash with ngram speculation #20739

Changes from 2 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
9bb2bb9
Fix NgramVerifyInput missing topk attribute
he-yufeng Mar 17, 2026
2e73812
Merge branch 'main' into fix/ngram-missing-topk
he-yufeng Mar 17, 2026
8a744e5
Fix hybrid_linear_attn_backend reading topk from spec_info instead of…
kpham-sgl Mar 26, 2026
cea0aaf
Merge branch 'main' into fix/ngram-missing-topk
kpham-sgl Apr 7, 2026
ea7cdec
Merge branch 'main' into fix/ngram-missing-topk
kpham-sgl Apr 8, 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
3 changes: 3 additions & 0 deletions python/sglang/srt/speculative/ngram_info.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -70,6 +70,9 @@ def __init__(
self.draft_token_num = draft_token_num
self.device = self.custom_mask.device
self.grammar = grammar
# Ngram speculative decoding doesn't use tree attention (topk=1),
# but some attention backends read spec_info.topk unconditionally.
self.topk = 1

def get_spec_adjust_token_coefficient(self) -> Tuple[int, int]:
return self.draft_token_num, self.draft_token_num
Expand Down
Loading