xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 05:15:42 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Only use triton_kernels for MXFP4 on SM90 and SM100 (#29339)

Browse Source 
Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Michael Goin 2025-11-24 18:22:46 -05:00 committed by GitHub
parent 71df2a57ef
commit c17610e2ba
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 9 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
vllm/model_executor/layers/quantization/mxfp4.py
View File

@ -132,12 +132,15 @@ def get_mxfp4_backend(with_lora_support: bool) -> Mxfp4Backend:
)
# If FlashInfer is not available, try either Marlin or Triton
if (
envs.VLLM_MXFP4_USE_MARLIN
or current_platform.get_device_capability()[0] < 9
or not has_triton_kernels()
or not is_torch_equal_or_newer("2.8.0")
):
triton_kernels_supported = (
has_triton_kernels()
and is_torch_equal_or_newer("2.8.0")
# NOTE: triton_kernels are only confirmed to work on SM90 and SM100
# SM110 fails with this error: https://github.com/vllm-project/vllm/issues/29317
# SM120 needs this fix: https://github.com/triton-lang/triton/pull/8498
and (9, 0) <= current_platform.get_device_capability() < (11, 0)
)
if envs.VLLM_MXFP4_USE_MARLIN or not triton_kernels_supported:
logger.info_once("Using Marlin backend")
return Mxfp4Backend.MARLIN
else: