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

[Log] Optimize startup log (#28948)

Browse Source 
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Signed-off-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
Co-authored-by: Nick Hill <nhill@redhat.com>
This commit is contained in:
Wentao Ye
2025-11-21 11:46:20 -05:00
committed by GitHub
parent b7f1f490a6
commit a42ab317ac
4 changed files with 26 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/model_executor/layers/quantization/fp8.py
View File

@@ -162,9 +162,11 @@ def get_fp8_moe_backend(block_quant: bool) -> Fp8MoeBackend:
# deepGEMM on supported platforms with block-quantized weights
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.")
logger.warning_once(
"DeepGEMM backend requested but not available.", scope="local"
)
elif is_deep_gemm_supported():
logger.info_once("Using DeepGEMM backend for FP8 MoE")
logger.info_once("Using DeepGEMM backend for FP8 MoE", scope="local")
return Fp8MoeBackend.DEEPGEMM
# CUTLASS BlockScaled GroupedGemm on SM100 with block-quantized weights
@@ -173,7 +175,9 @@ def get_fp8_moe_backend(block_quant: bool) -> Fp8MoeBackend:
and current_platform.is_device_capability(100)
and block_quant
):
logger.info_once("Using Cutlass BlockScaled GroupedGemm backend for FP8 MoE")
logger.info_once(
"Using Cutlass BlockScaled GroupedGemm backend for FP8 MoE", scope="local"
)
return Fp8MoeBackend.CUTLASS_BLOCK_SCALED_GROUPED_GEMM
# default to Triton