[Kernel] Enable FusedMoEModularKernel support bias (#27754)

Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
2025-11-01 10:05:12 +08:00
parent 0cdbe7b744
commit bc4486d609
2 changed files with 15 additions and 30 deletions
--- a/vllm/model_executor/layers/fused_moe/layer.py
+++ b/vllm/model_executor/layers/fused_moe/layer.py
@@ -672,8 +672,6 @@ class UnquantizedFusedMoEMethod(FusedMoEMethodBase, CustomOp):
                apply_router_weight_on_input=apply_router_weight_on_input,
            )
        elif self.fused_experts is not None:
-            if self.moe.has_bias:
-                raise ValueError("FusedMoEModularKernel does not support bias.")
            result = self.fused_experts(
                hidden_states=x,
                w1=layer.w13_weight,