set llama and qwen tn layout

vllm/model_executor/model_loader/utils.py

@@ -80,7 +80,7 @@ def get_model_architecture(
     architectures = getattr(model_config.hf_config, "architectures", [])
     visions = getattr(model_config.hf_config, "visual", []) or getattr(model_config.hf_config, "vision_config", [])
     # TODO: support deepseek distillation series models ( 'LlamaForCausalLM', 'Qwen2ForCausalLM' )
-    support_nn_architectures = ['LlamaForCausalLM', 'Qwen2ForCausalLM', 'QWenLMHeadModel', 'Qwen2VLForConditionalGeneration', 'Qwen2_5_VLForConditionalGeneration', 
+    support_nn_architectures = ['QWenLMHeadModel', 'Qwen2VLForConditionalGeneration', 'Qwen2_5_VLForConditionalGeneration', 
                                 'Qwen2MoeForCausalLM', 'ChatGLMModel', 'ChatGLMForConditionalGeneration', 
                                 'BaichuanForCausalLM', 'BloomForCausalLM', 'MedusaModel', 'MixtralForCausalLM', 
                                 'MLPSpeculatorPreTrainedModel', 'FalconForCausalLM', 'DeepseekV2ForCausalLM',