Skip to content

[BugFix] Stopgap - Flashinfer Autotuner + GPT-OSS + DP/TP #27762

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
merged 4 commits into from
Oct 30, 2025
Merged

[BugFix] Stopgap - Flashinfer Autotuner + GPT-OSS + DP/TP #27762

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
5e54e7e
relax deepep_ht skip
Oct 29, 2025
67718f1
better conditions
Oct 29, 2025
c95b2d7
fix b200
Oct 29, 2025
5ca0376
fixes
Oct 29, 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
20 changes: 13 additions & 7 deletions vllm/model_executor/warmup/kernel_warmup.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,7 +11,7 @@
import torch

import vllm.envs as envs
from vllm.config import VllmConfig
from vllm.config import CUDAGraphMode, VllmConfig
from vllm.logger import init_logger
from vllm.model_executor.warmup.deep_gemm_warmup import deep_gemm_warmup
from vllm.platforms import current_platform
Expand All @@ -30,13 +30,19 @@ def flashinfer_autotune_supported(vllm_config: VllmConfig) -> bool:
Record known issues with vllm + flashinfer autotune here. Return True if
and only if flashinfer autotune will run through without issues.
"""
return not (
vllm_config.parallel_config.data_parallel_size > 1
and (
envs.VLLM_USE_FLASHINFER_MOE_MXFP4_BF16
or envs.VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8
)
is_tp_or_dp = (vllm_config.parallel_config.data_parallel_size > 1) or (
vllm_config.parallel_config.tensor_parallel_size > 1
)
is_fi_mxfp4_backend = (
envs.VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8
or envs.VLLM_USE_FLASHINFER_MOE_MXFP4_BF16
or envs.VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8_CUTLASS
) or (
current_platform.is_cuda() and current_platform.is_device_capability(100)
) # on >=sm100, default mxfp4 backend is flashinfer
is_eager = vllm_config.compilation_config.cudagraph_mode == CUDAGraphMode.NONE

return not (is_tp_or_dp and is_fi_mxfp4_backend and is_eager)


def kernel_warmup(worker: "Worker"):
Expand Down