Skip to content

[None][fix] SpecDec: Sampling seed fix #11081

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
MartinMarciniszyn merged 1 commit into from
Feb 20, 2026
Merged

[None][fix] SpecDec: Sampling seed fix #11081

Show file tree
Hide file tree
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
2 changes: 1 addition & 1 deletion ATTRIBUTIONS-Python.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5261,7 +5261,7 @@ For more information, please refer to <http://unlicense.org>
- `Tracker`: https://github.com/tox-dev/py-filelock/issues


## flashinfer-python (0.6.3)
## flashinfer-python (0.6.4)

### Licenses
License: `Apache-2.0`
Expand Down
2 changes: 1 addition & 1 deletion requirements.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,7 +54,7 @@ ordered-set
peft
patchelf
einops
flashinfer-python==0.6.3
flashinfer-python==0.6.4
opencv-python-headless
xgrammar==0.1.25
llguidance==0.7.29
Expand Down
2 changes: 1 addition & 1 deletion security_scanning/pyproject.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,7 +56,7 @@ dependencies = [
"peft (>=0.18.1,<0.19.0)",
"patchelf (>=0.17.2.4,<0.18.0.0)",
"einops (>=0.8.2,<0.9.0)",
"flashinfer-python (==0.6.3)",
"flashinfer-python (==0.6.4)",
"xgrammar (==0.1.25)",
"llguidance (==0.7.29)",
"jsonschema (>=4.26.0,<5.0.0)",
Expand Down
15 changes: 12 additions & 3 deletions tensorrt_llm/_torch/speculative/interface.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -388,9 +388,9 @@ def __init__(self, use_separate_draft_kv_cache: bool = False):
super().__init__()
self.guided_decoder: Optional["CapturableGuidedDecoder"] = None
self.force_num_accepted_tokens = get_force_num_accepted_tokens()
self.use_flashinfer = IS_FLASHINFER_AVAILABLE and flashinfer.__version__ >= "0.6.0"
self.seed = 0
self.offset = 0
self.use_flashinfer = IS_FLASHINFER_AVAILABLE and flashinfer.__version__ >= "0.6.4"
self.seed: Optional[torch.Tensor] = None
self.offset: Optional[torch.Tensor] = None
self.use_separate_draft_kv_cache = use_separate_draft_kv_cache

@property
Expand Down Expand Up @@ -688,7 +688,16 @@ def _sample_tokens_for_batch(
top_ps = spec_metadata.top_ps[:num_tokens]

if self.use_flashinfer:
# Lazily initialize seed/offset tensors on correct device
if self.seed is None:
self.seed = torch.tensor([0],
dtype=torch.int64,
device=logits.device)
self.offset = torch.tensor([0],
dtype=torch.int64,
device=logits.device)
self.seed += 1
self.seed %= (2**31)

sampled_tokens = sampling_batch_spec_dec_one_model(
logits,
Expand Down