Merge branch 'v0.7.2-dev_wm' into 'v0.7.2-dev'

[fix]修复ds3量化代码引入的报错

See merge request dcutoolkit/deeplearing/vllm!75

vllm/model_executor/model_loader/utils.py

@@ -84,7 +84,7 @@ def get_model_architecture(
                                 'Qwen2MoeForCausalLM', 'ChatGLMModel', 'ChatGLMForConditionalGeneration', 
                                 'BaichuanForCausalLM', 'BloomForCausalLM', 'MedusaModel', 'MixtralForCausalLM', 
                                 'MLPSpeculatorPreTrainedModel', 'FalconForCausalLM', 'DeepseekV2ForCausalLM', 
-                                'DeepseekV3ForCausalLM', 'DeepSeekMTP']  
+                                'DeepseekV3ForCausalLM', 'DeepSeekMTPModel']  
     if any(arch in architectures for arch in support_nn_architectures): 
         if os.getenv('LLAMA_NN') != '0': 
              if (architectures == ['QWenLMHeadModel'] or architectures == ['ChatGLMModel'] ) and visions != []:

vllm/model_executor/models/deepseek_v2.py

@@ -870,7 +870,7 @@ class DeepseekV2ForCausalLM(nn.Module, SupportsPP):
                     
                     weight.data=weight.data.reshape(ori_shape[1],-1)
 
-        if self.config.quantization_config["quant_method"] == "awq" and not envs.VLLM_USE_TRITON_AWQ:
+        if hasattr(self.config, "quantization_config") and self.config.quantization_config["quant_method"] == "awq" and not envs.VLLM_USE_TRITON_AWQ:
             lay_key_words = [
                 "self_attn.q_a_proj.qweight",
                 "self_attn.q_b_proj.qweight",