[Refactor] Create a function util and cache the results for has_deepgemm, has_deepep, has_pplx (#20187)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye
2025-06-28 22:06:38 +00:00
committed by GitHub
parent daec9dea6e
commit 4d36693687
12 changed files with 61 additions and 58 deletions
@@ -2,7 +2,6 @@
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import functools
import importlib.util
from typing import Any, Callable, Optional, Union
import torch
@@ -38,13 +37,12 @@ from vllm.model_executor.parameter import (BlockQuantScaleParameter,
from vllm.model_executor.utils import set_weight_attrs
from vllm.platforms import current_platform
from vllm.scalar_type import scalar_types
from vllm.utils import has_deep_gemm
ACTIVATION_SCHEMES = ["static", "dynamic"]
logger = init_logger(__name__)
has_deep_gemm = importlib.util.find_spec("deep_gemm") is not None
def _is_col_major(x: torch.Tensor) -> bool:
assert x.dim() == 3
@@ -451,7 +449,7 @@ class Fp8MoEMethod(FusedMoEMethodBase):
# Check for DeepGemm support.
self.allow_deep_gemm = False
if envs.VLLM_USE_DEEP_GEMM:
if not has_deep_gemm:
if not has_deep_gemm():
logger.warning_once("Failed to import DeepGemm kernels.")
elif not self.block_quant:
logger.warning_once("Model is not block quantized. Not using "