Skip to content

Fix Illegal Instruction/IMA errors when using DP attention -- num_tokens_for_logprob calculation #12115

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
ch-wan merged 2 commits into from
Oct 25, 2025
Merged

Fix Illegal Instruction/IMA errors when using DP attention -- num_tokens_for_logprob calculation #12115

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
fff4454
fix num_tokens_for_logprob calculation when return_logprob=False
YAMY1234 Oct 25, 2025
c2203c4
fix format & gemini suggestion
YAMY1234 Oct 25, 2025
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
13 changes: 8 additions & 5 deletions python/sglang/srt/managers/scheduler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2073,15 +2073,18 @@ def prepare_mlp_sync_batch_raw(
num_tokens_for_logprob = num_tokens
else:
num_tokens = local_batch.extend_num_tokens
num_tokens_for_logprob = sum(
[
if local_batch.return_logprob:
num_tokens_for_logprob = sum(
# We should have at least 1 token for sample in every case.
max(extend_len - logprob_start_len, 1)
for logprob_start_len, extend_len in zip(
local_batch.extend_logprob_start_lens, local_batch.extend_lens
local_batch.extend_logprob_start_lens,
local_batch.extend_lens,
)
]
)
)
else:
# When return_logprob = False, only need last token per request
num_tokens_for_logprob = local_batch.batch_size()

if local_batch is None or local_batch.forward_mode.is_decode_or_idle():
can_cuda_graph = 1
Expand Down
Loading