update qwen3_moe.py

README.md

@@ -33,6 +33,8 @@ vLLM是一个快速且易于使用的LLM推理和服务库,使用PageAttention
 | Qwen2MoeForCausalLM                 | Qwen2-57B-A14B,Qwen2-57B-A14B-Instruct        | Yes | No  | -   |
 | LlavaForConditionalGeneration       | LLaMA,LLaMA-2,LLaMA-3                         | Yes | No  | -   |
 | Qwen2VLForConditionalGeneration     | Qwen2-VL                                      | Yes | No  | Yes |
+| Qwen2_5_VLForConditionalGeneration  | Qwen.5-VL                                     | Yes | No  | Yes |
+| Gemma3ForConditionalGeneration      | Gemma 3                                       | Yes | -   | -   |
 | MiniCPMV                            | MiniCPM-V                                     | Yes | No  | -   |
 | Phi3VForCausalLM                    | Phi-3.5-vision                                | Yes | No  | -   |
 | BertModel                           | bge-large-zh-v1.5                             | Yes | No  | -   |

vllm/model_executor/models/qwen3_moe.py

@@ -414,9 +414,6 @@ class Qwen3MoeModel(nn.Module):
         params_dict = dict(self.named_parameters())
         loaded_params: Set[str] = set()
         for name, loaded_weight in weights:
-            if self.use_llama_nn:
-                current_count = loaded_weight.current_count 
-                total_count = loaded_weight.total_count
             for (param_name, weight_name, shard_id) in stacked_params_mapping:
                 # Skip non-stacked layers and experts (experts handled below).
                 if weight_name not in name:
@@ -493,7 +490,7 @@ class Qwen3MoeModel(nn.Module):
                     weight_loader(param, loaded_weight)
             loaded_params.add(name)
         
-        if self.use_llama_nn and self.quant_method is None and current_count==total_count:
+        if self.use_llama_nn and self.quant_method is None:
             lay_key_words = [
                 "gate_up_proj.weight",
                 "down_proj.weight",