Skip to content

Remove unnecessary FlashInfer top-k fallback (environment issue, not a CUDA 13 bug) #2

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
oceanplexian merged 1 commit into from
Aug 19, 2026
Merged

Remove unnecessary FlashInfer top-k fallback (environment issue, not a CUDA 13 bug) #2

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
20 changes: 5 additions & 15 deletions vllm/model_executor/models/qwen3_dflash2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,7 @@
import torch
import torch.nn.functional as F
from torch import nn
from transformers import Qwen3Config

from vllm.compilation.decorators import support_torch_compile
from vllm.config import CacheConfig, VllmConfig
Expand Down Expand Up @@ -56,22 +57,11 @@ def _flashinfer_topk() -> Callable[..., tuple[torch.Tensor, torch.Tensor]] | Non
return top_k


_flashinfer_topk_failed = False


def _topk(scores: torch.Tensor, k: int) -> tuple[torch.Tensor, torch.Tensor]:
global _flashinfer_topk_failed
impl = _flashinfer_topk()
if impl is not None and not _flashinfer_topk_failed and scores.is_cuda:
try:
return impl(scores, k, sorted=True, deterministic=True)
except Exception:
_flashinfer_topk_failed = True
logger.warning_once(
"FlashInfer radix top-k unavailable; falling back to torch.topk "
"at roughly half the speed."
)
return torch.topk(scores, k, dim=-1)
if impl is None or not scores.is_cuda:
return torch.topk(scores, k, dim=-1)
return impl(scores, k, sorted=True, deterministic=True)


def _grouped_conv(
Expand Down Expand Up @@ -160,7 +150,7 @@ def __init__(
self,
vllm_config: VllmConfig,
*,
config,
config: Qwen3Config,
layer_idx: int,
cache_config: CacheConfig | None = None,
quant_config: QuantizationConfig | None = None,
Expand Down