qwen3_moe (#3084)

fastdeploy/model_executor/layers/quantization/weight_only.py

@@ -168,7 +168,7 @@ class WeightOnlyLinearMethod(QuantMethodBase):
         super().__init__()
         self.quant_config = quant_config
 
-    def create_weights(self, layer):
+    def create_weights(self, layer, **extra_weight_attrs):
 
         # The scale shape should be equal to the output dim of weight using Per-Channel Quantization.
         weight_scale_shape = [layer.weight_shape[1]]
@@ -177,6 +177,14 @@ class WeightOnlyLinearMethod(QuantMethodBase):
         if self.quant_config.name() == "wint4":
             layer.weight_shape[0] //= 2
         layer.weight_dtype = "int8"
+
+        layer.weight = layer.create_parameter(
+            shape=layer.weight_shape,
+            dtype=layer.weight_dtype,
+            is_bias=False,
+            default_initializer=paddle.nn.initializer.Constant(0),
+        )
+
         layer.weight_scale = layer.create_parameter(
             shape=weight_scale_shape,
             dtype=layer._dtype,