修改awq相关printf提示，在cmakelist中添加ck so路径

lmdeploy/turbomind/turbomind.py

@@ -381,7 +381,7 @@ class TurboMind:
         self.config = cfg
         self.model_name = cfg.model_name
         self.data_type = cfg.weight_type
-        print("from_workspace_cfg:",cfg)
+        #print("from_workspace_cfg:",cfg)
 
         # create model
         logger.warning(f'model_config:\n\n{cfg.toini()}')

src/turbomind/models/llama/CMakeLists.txt

@@ -25,6 +25,7 @@ set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fPIC")
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -fPIC")
 #set_property(TARGET Llama PROPERTY POSITION_INDEPENDENT_CODE  ON)
 #set_property(TARGET Llama PROPERTY CUDA_RESOLVE_DEVICE_SYMBOLS  ON)
+target_link_directories(Llama PUBLIC ../../../../3rdparty/)
 target_link_libraries(Llama PUBLIC cudart
         gemm_s4_f16
         cublasMMWrapper

src/turbomind/models/llama/LlamaLinear.h

@@ -165,11 +165,6 @@ private:
                }
             
             }
-            if(cublas_wrapper_->m_dump_switch==2)
-            {
-                std::cout<<" m: "<<batch_size<<" n: "<<weight.output_dims<<" k: "<<weight.input_dims<<std::endl;
-                PrintScale<T>(stream_,output_data,36,0,0,0);
-            }
             sync_check_cuda_error();
         }
         else {
@@ -249,10 +244,6 @@ private:
             else if(weight.w4_weight_layout==2) //TN 模式padding ck
             {
                 //检查ck workspace 的空间是否足够
-                if(batch_size*weight.output_dims>M_max*N_max)
-                {
-                    FT_CHECK_WITH_INFO(0, "error! ck workspace is not enough");
-                }
 
                 if(weight.input_dims%4096==0)
                 {
@@ -265,11 +256,6 @@ private:
                }
             }
             addFusedSiluActivation(stream_,output_data,output_tmp,batch_size,weight.output_dims,1);
-            if(cublas_wrapper_->m_dump_switch==2)
-            {
-                std::cout<<" m: "<<batch_size<<" n: "<<weight.output_dims<<" k: "<<weight.input_dims<<std::endl;
-                PrintScale<T>(stream_,output_data,36,0,0,0);
-            }
             sync_check_cuda_error();
         }
         else {

src/turbomind/models/llama/LlamaWeight.cc

@@ -69,14 +69,14 @@ LlamaWeight<T>::LlamaWeight(size_t     head_num,
         std::string str_w4_weight_layout = std::to_string(w4_weight_layout);
         const char* env_value = str_w4_weight_layout.c_str();
         setenv(env_name,env_value , 1);
-        printf("set LMDEPLOY_WEIGHTLAYOUT_SWITCH env: %d \n",w4_weight_layout);
+        //printf("set LMDEPLOY_WEIGHTLAYOUT_SWITCH env: %d \n",w4_weight_layout);
     }
     else
     {
         std::string str_w4_weight_layout = std::to_string(-1);
         const char* env_value = str_w4_weight_layout.c_str();
         setenv(env_name,env_value , 1);
-        printf("set LMDEPLOY_WEIGHTLAYOUT_SWITCH env: %d \n",w4_weight_layout);
+        //printf("set LMDEPLOY_WEIGHTLAYOUT_SWITCH env: %d \n",-1);
     }
     mallocWeights();
 }