去掉调试信息

vllm/model_executor/layers/linear.py

@@ -108,19 +108,14 @@ class UnquantizedLinearMethod(LinearMethodBase):
               x: torch.Tensor,
               bias: Optional[torch.Tensor] = None) -> torch.Tensor:
         weight = layer.weight
-        #print("**************matmul weight.shape:",weight.shape)
-        #print("self.use_llama_nn:",self.use_llama_nn)
         
         if self.separate_bias_add:
-            #print("********self.separate_bias_add")
             
             if bias is not None:
                 return F.linear(x, weight) + bias
             return F.linear(x, weight)
         
         if self.use_llama_nn:
-            # print("**************matmul input.shape:",x.shape)
-            # print("**************matmul weight.shape:",weight.shape)
           
             if bias is not None:
                 return torch.matmul(x, weight) +bias

vllm/model_executor/models/llama.py

@@ -459,14 +459,10 @@ class LlamaForCausalLM(nn.Module):
                 #print("key:\n",key)
                 matches = re.findall(combined_words, layername)
                 if matches:                    
-                    #print(layername)
-                    # print(weight.data)
                     #创建一个跟value一样大的tensor
                     _weight = torch.zeros_like(weight.data)
                     ori_shape =_weight.shape
                     
-                    # if layername=="model.layers.0.self_attn.qkv_proj.weight":
-                    #     print("weight.data[0:5][0:5]:",weight.data[0:5][0:5])
                     ops.trans_w16_gemm(_weight,weight.data,_weight.shape[0],_weight.shape[1])
                     weight.data.copy_(_weight)