[ROCm][Quantization] GPT_OSS in amd-quark format model loading and emulations (#29008)

Signed-off-by: xuebwang-amd <xuebwang@amd.com>
Signed-off-by: Robert Shaw <robshaw@redhat.com>
Co-authored-by: Robert Shaw <robshaw@redhat.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>

vllm/model_executor/layers/quantization/mxfp4.py

@@ -229,10 +229,15 @@ class Mxfp4Config(QuantizationConfig):
             )
         return None
 
+    def is_mxfp4_quant(self, prefix: str, layer: torch.nn.Module) -> bool:
+        """MXFP4 config always uses MXFP4 quantization."""
+        return True
+
 
 class Mxfp4MoEMethod(FusedMoEMethodBase):
     def __init__(self, moe: FusedMoEConfig):
         super().__init__(moe)
+        self.weight_dtype = "mxfp4"
         self.mxfp4_backend = get_mxfp4_backend(moe.is_lora_enabled)
 
         self.marlin_input_dtype = None