[XPU] enable is_act_and_mul for xpu (#37481)

Signed-off-by: Chendi Xue <chendi.xue@intel.com>
Co-authored-by: Kunshang Ji <kunshang.ji@intel.com>

vllm/model_executor/layers/fused_moe/experts/xpu_moe.py

@@ -62,7 +62,7 @@ class XPUExperts(mk.FusedMoEExpertsModular):
 
     @staticmethod
     def _supports_no_act_and_mul() -> bool:
-        return False
+        return True
 
     @staticmethod
     def _supports_activation(activation: MoEActivation) -> bool:
@@ -70,6 +70,7 @@ class XPUExperts(mk.FusedMoEExpertsModular):
             MoEActivation.SILU,
             MoEActivation.GELU,
             MoEActivation.SWIGLUOAI,
+            MoEActivation.RELU2_NO_MUL,
         ]
 
     @staticmethod

vllm/model_executor/layers/fused_moe/layer.py

@@ -538,9 +538,11 @@ class FusedMoE(PluggableLayer):
         # for heuristic purposes, so it must be initialized first.
         self.quant_method: FusedMoEMethodBase = _get_quant_method()
 
-        if not self.moe_config.is_act_and_mul and not current_platform.is_cuda_alike():
+        if not self.moe_config.is_act_and_mul and not (
+            current_platform.is_cuda_alike() or current_platform.is_xpu()
+        ):
             raise NotImplementedError(
-                "is_act_and_mul=False is supported only for CUDA and ROCm for now"
+                "is_act_and_mul=False is supported only for CUDA and XPU for now"
             )
 
         if self.enable_eplb and not self.quant_method.supports_eplb: