clean nvfp4 related code (#6644)

fastdeploy/model_executor/layers/quantization/nvfp4.py

@@ -167,12 +167,17 @@ class ModelOptNvFp4LinearMethod(QuantMethodBase):
         layer,
         **extra_weight_attrs,
     ):
+        # 因为模型存储是列存储的，所以这里需要not一下！
         extra_weight_attrs["output_dim"] = not extra_weight_attrs["output_dim"]
-        weight_shape = layer.weight_shape[::-1]
-        weight_shape[1] = weight_shape[1] // 2
+        K = layer.weight_shape[0]
+        N = layer.weight_shape[1]
+        # 因为模型的存储时候权重是[N,K//2]
+        # 所以这里创建的权重是为了契合模型存储的权重！
+        weight_shape = [N, K // 2]
         layer.weight_dtype = "uint8"
+
         input_scale_shape = [1]
-        weight_scale_shape = [layer.weight_shape[::-1][0], layer.weight_shape[::-1][1] // self.quant_config.group_size]
+        weight_scale_shape = [N, K // self.quant_config.group_size]
         weight_scale_2_shape = [1]
 
         self._create_main_weight(layer, weight_shape, extra_weight_attrs)
@@ -332,7 +337,9 @@ class ModelOptNvFp4LinearMethod(QuantMethodBase):
         else:
             raise ValueError(f"Unsupported backend: {self.backend}.")
 
+        # shape 恢复到[K//2,N]
         w = layer.weight.T
+        # shape 恢复到[K//group_size, N]
         w_scale_interleaved = layer.weight_scale_interleaved.T
 
         if backend == "cutlass":
@@ -343,7 +350,8 @@ class ModelOptNvFp4LinearMethod(QuantMethodBase):
         out = fp4_gemm(x_fp4, w, x_scale_interleaved, w_scale_interleaved, layer.alpha, output_dtype, backend=backend)
         if layer.with_bias:
             out = paddle.add(out, layer.bias)
-        return out.view(*output_shape)
+        assert out.shape == output_shape
+        return out
 
 
 class ModelOptNvFp4FusedMoE(QuantMethodBase):
@@ -385,7 +393,7 @@ class ModelOptNvFp4FusedMoE(QuantMethodBase):
 
     def create_weights(self, layer, **extra_weight_attrs):
         """
-        Triton MoE create weight process.
+        NVFP4 MoE create weight.
         """
         self.up_gate_proj_weight_shape = [
             layer.num_local_experts,