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

[Feature] Add env var VLLM_MOE_USE_DEEP_GEMM (#28422)

Browse Source 
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye
2025-11-10 21:29:48 -05:00
committed by GitHub
parent 39029d5192
commit de540c0354
4 changed files with 19 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe.py
View File

@@ -966,10 +966,18 @@ class CompressedTensorsW8A8Fp8MoEMethod(CompressedTensorsMoEMethod):
max_num_tokens=max_num_tokens_per_rank,
num_dispatchers=prepare_finalize.num_dispatchers(),
quant_config=self.moe_quant_config,
allow_deep_gemm=(
envs.VLLM_USE_DEEP_GEMM and envs.VLLM_MOE_USE_DEEP_GEMM
),
)
else:
logger.debug("TritonOrDeepGemmExperts(%s)", self.__class__.__name__)
return TritonOrDeepGemmExperts(self.moe_quant_config, allow_deep_gemm=True)
return TritonOrDeepGemmExperts(
self.moe_quant_config,
allow_deep_gemm=(
envs.VLLM_USE_DEEP_GEMM and envs.VLLM_MOE_USE_DEEP_GEMM
),
)
def get_fused_moe_quant_config(
self, layer: torch.nn.Module

2
vllm/model_executor/layers/quantization/fp8.py
View File

@@ -158,7 +158,7 @@ def get_fp8_moe_backend(block_quant: bool) -> Fp8MoeBackend:
return Fp8MoeBackend.MARLIN
# deepGEMM on supported platforms with block-quantized weights
if envs.VLLM_USE_DEEP_GEMM and block_quant:
if envs.VLLM_USE_DEEP_GEMM and envs.VLLM_MOE_USE_DEEP_GEMM and block_quant:
if not has_deep_gemm():
logger.warning_once("DeepGEMM backend requested but not available.")
elif is_deep_gemm_supported():