qwen3_moe (#3084)

2025-10-05 08:37:06 +08:00 · 2025-08-06 14:45:27 +08:00
parent 91dc87f1c5
commit 20839abccf
30 changed files with 1361 additions and 1087 deletions
--- a/fastdeploy/model_executor/layers/quantization/block_wise_fp8.py
+++ b/fastdeploy/model_executor/layers/quantization/block_wise_fp8.py
@@ -81,8 +81,16 @@ class BlockWiseFP8LinearMethod(QuantMethodBase):
        super().__init__()
        self.quant_config = quant_config

-    def create_weights(self, layer):
+    def create_weights(self, layer, **extra_weight_attrs):
        layer.weight_shape.reverse()
+
+        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=[
                (layer.output_size + self.quant_config.weight_block_size[0] - 1)