frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix][CPU] Skip unsupported custom op register on CPU (#25534)

Browse Source 
Signed-off-by: jiang1.li <jiang1.li@intel.com>
This commit is contained in:
Li, Jiang
2025-09-24 14:21:51 +08:00
committed by GitHub
parent fed8a9b107
commit 1cbcfb94de
1 changed files with 8 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
vllm/model_executor/layers/quantization/utils/fp8_utils.py
View File

@ -130,12 +130,14 @@ def _w8a8_triton_block_scaled_mm_fake(
device=qx.device)
direct_register_custom_op(
"w8a8_triton_block_scaled_mm_func",
_w8a8_triton_block_scaled_mm_func,
fake_impl=_w8a8_triton_block_scaled_mm_fake,
dispatch_key="CUDA",
)
# Note: the check can be removed when CPU torch > 2.7
if not current_platform.is_cpu():
direct_register_custom_op(
"w8a8_triton_block_scaled_mm_func",
_w8a8_triton_block_scaled_mm_func,
fake_impl=_w8a8_triton_block_scaled_mm_fake,
dispatch_key="CUDA",
)
# TODO fix ROCm->Triton custom path: