biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Remove redundant mutates_args and dispatch_key for direct_register_custom_op (#25512)

Browse Source 
Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Michael Goin
2025-09-23 18:48:40 -04:00
committed by GitHub
parent 95bc60e4cb
commit 7361ab379f
28 changed files with 9 additions and 66 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/model_executor/layers/quantization/utils/fp8_utils.py
View File

@@ -91,9 +91,7 @@ if current_platform.is_rocm():
direct_register_custom_op(
op_name="rocm_aiter_gemm_w8a8_blockscale",
op_func=rocm_aiter_gemm_w8a8_blockscale_impl,
mutates_args=[],
fake_impl=rocm_aiter_gemm_w8a8_blockscale_fake,
dispatch_key=current_platform.dispatch_key,
)
if (envs.VLLM_ROCM_USE_AITER and envs.VLLM_ROCM_USE_AITER_LINEAR
and current_platform.is_fp8_fnuz()):
@@ -135,7 +133,6 @@ def _w8a8_triton_block_scaled_mm_fake(
direct_register_custom_op(
"w8a8_triton_block_scaled_mm_func",
_w8a8_triton_block_scaled_mm_func,
mutates_args=[],
fake_impl=_w8a8_triton_block_scaled_mm_fake,
dispatch_key="CUDA",
)