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

[Kernel] Delegate construction of FusedMoEQuantConfig to FusedMoEMethodBase subclasses (#22537)

Browse Source 
Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
bnellnm
2025-09-17 19:43:31 -04:00
committed by GitHub
parent e6585ddb45
commit 5963b98b46
68 changed files with 2698 additions and 2526 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
vllm/model_executor/warmup/deep_gemm_warmup.py
View File

@@ -81,9 +81,14 @@ def _fp8_linear_may_use_deep_gemm(module: torch.nn.Module) -> bool:
def _fused_moe_grouped_gemm_may_use_deep_gemm(module: torch.nn.Module) -> bool:
if not (isinstance(module, FusedMoE)
and module.moe_config.quant_dtype == torch.float8_e4m3fn
and module.moe_config.block_shape == deep_gemm_block_shape()):
if not isinstance(module, FusedMoE):
return False
moe_quant_config = module.quant_method.get_fused_moe_quant_config(module)
if (moe_quant_config is None
or moe_quant_config.quant_dtype != torch.float8_e4m3fn
or moe_quant_config.block_shape != deep_gemm_block_shape()):
return False
if not isinstance(module.quant_method.fused_experts,