Skip to content

[https://nvbugs/5791242][fix] workaround for flashinfer.sampling.sampling_from_logits #10713

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
dongfengy merged 1 commit into from
Jan 20, 2026
Merged

[https://nvbugs/5791242][fix] workaround for flashinfer.sampling.sampling_from_logits #10713

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
31 changes: 20 additions & 11 deletions tensorrt_llm/_torch/pyexecutor/sampling_utils_flashinfer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -612,20 +612,29 @@ def sample(
generator: Optional[torch.Generator] = None,
group_metadata: StrategyMetadata | None = None,
) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
logits = self._prepare_logits_with_temperature(
logits, group_logit_indices, self._temperature
)
new_tokens = flashinfer.sampling.sampling_from_logits(
new_tokens, _ = self._sample_with_probs(
logits,
# NB: Leveraging 'indices' would require applying temperature+softmax before batching,
# because 'flashinfer.sampling.softmax' has no 'indices' argument; but that would
# compute unnecessarily softmax also for situations allowing
# flashinfer.sampling...._sampling_from_logits.
# indices=group_logit_indices,
deterministic=True,
group_logit_indices=group_logit_indices,
top_k=None,
top_p=None,
temperature=self._temperature,
generator=generator,
check_nan=self._flashinfer_check_nans(logits),
)
# FIXME: https://nvbugs/5791242
# logits = self._prepare_logits_with_temperature(
# logits, group_logit_indices, self._temperature
# )
# new_tokens = flashinfer.sampling.sampling_from_logits(
# logits,
# # NB: Leveraging 'indices' would require applying temperature+softmax before batching,
# # because 'flashinfer.sampling.softmax' has no 'indices' argument; but that would
# # compute unnecessarily softmax also for situations allowing
# # flashinfer.sampling...._sampling_from_logits.
# # indices=group_logit_indices,
# deterministic=True,
# generator=generator,
# check_nan=self._flashinfer_check_nans(logits),
# )
return new_tokens, None

class BeamSearchSampleOnly(BeamSearchMixin, StrategyImplSampleOnly):
Expand Down