fix: 修复W8A8INT8读config问题

vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe.py

@@ -1516,6 +1516,7 @@ class CompressedTensorsW8A8Int8MoEMethod(CompressedTensorsMoEMethod):
                 f"{self.weight_quant}, {self.input_quant}")
 
         self.static_input_scales = not self.input_quant.dynamic
+        self.tritonsingleton= W8a8GetCacheJSON()
 
 
     def create_weights(self, layer: torch.nn.Module, num_experts: int,
@@ -1570,6 +1571,27 @@ class CompressedTensorsW8A8Int8MoEMethod(CompressedTensorsMoEMethod):
             layer.w13_input_scale = None
             layer.w2_input_scale = None
 
+    def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
+        E=layer.w13_weight.shape[0]
+        N1=layer.w13_weight.shape[1]
+        N2=layer.w2_weight.shape[1]
+        K=layer.w2_weight.shape[2]
+        if [E,N1,N2,K] not in self.tritonsingleton.moe_weight_shapes:
+            self.tritonsingleton.moe_weight_shapes.append([E,N1,N2,K])
+
+        TOPK= self.tritonsingleton.topk
+
+        json_file=self.tritonsingleton.get_moeint8json_name(E,N1,N2,K,TOPK)
+        configs_dict=self.tritonsingleton.get_moeint8_triton_cache(json_file,E,N1,N2,K,TOPK)
+
+        #warmup
+        if configs_dict:
+            self.tritonsingleton.triton_moejson_dict.update(configs_dict)
+
+        #生成模型配置文件
+        #self.tritonsingleton.gen_model_json(block_size)
+        return
+
 
     def apply(
         self,