support qwen3 nn layout

README.md

@@ -74,7 +74,7 @@ cd dist
 pip install vllm*
 
 2. 源码编译安装
-python3 setup.py install （若调试，可使用VLLM_INSTALL_PUNICA_KERNELS=1 python3 setup.py develop）
+python3 setup.py install （若调试，可使用python3 setup.py develop）
 ```
 若需要添加git号，设置环境变量: export ADD_GIT_VERSION=1
 

vllm/model_executor/model_loader/utils.py

@@ -89,7 +89,8 @@ def get_model_architecture(
         model_config: ModelConfig) -> Tuple[Type[nn.Module], str]:
     architectures = getattr(model_config.hf_config, "architectures", [])
     visions = getattr(model_config.hf_config, "visual", []) or getattr(model_config.hf_config, "vision_config", [])
-    support_nn_architectures = ['LlamaForCausalLM', 'Qwen2ForCausalLM', 'QWenLMHeadModel', 'Qwen2VLForConditionalGeneration', 'Qwen2_5_VLForConditionalGeneration', 'Qwen2MoeForCausalLM', 'ChatGLMModel', 'ChatGLMForConditionalGeneration', 
+    support_nn_architectures = ['LlamaForCausalLM', 'Qwen2ForCausalLM', 'QWenLMHeadModel', 'Qwen2VLForConditionalGeneration', 'Qwen2_5_VLForConditionalGeneration', 'Qwen2MoeForCausalLM', 'Qwen3ForCausalLM',
+                                'ChatGLMModel', 'ChatGLMForConditionalGeneration', 
                                 'BaichuanForCausalLM', 'BloomForCausalLM', 'MedusaModel', 'MixtralForCausalLM', 'MLPSpeculatorPreTrainedModel', 'FalconForCausalLM', 'DeepseekV2ForCausalLM', 'DeepseekV3ForCausalLM', 'DeepSeekMTPModel']  
     if any(arch in architectures for arch in support_nn_architectures): 
         if os.getenv('LLAMA_NN') != '0':