@@ -139,7 +139,7 @@ struct LLM::Impl {
139139 llama_layers.resize (attr.axmodel_num );
140140 auto dev_assign = distributeModels ((int )_attr.dev_ids .size (), attr.axmodel_num );
141141 std::vector<int > rets (attr.axmodel_num , 0 );
142- std::atomic<int > process_idx (2 );
142+ std::atomic<int > process_idx (1 );
143143#pragma omp parallel for if (_attr.dev_ids.size() > 1)
144144 for (int i = 0 ; i < attr.axmodel_num ; i++)
145145 {
@@ -159,7 +159,7 @@ struct LLM::Impl {
159159 if (ret != 0 ) { ALOGE (" init post axmodel(%s) failed" , attr.filename_post_axmodel .c_str ()); return false ; }
160160 char path[1024 ];
161161 sprintf (path, " init post axmodel ok,remain_cmm(%d MB)" , axcl_GetCMMRemain (post_devid));
162- update_cqdm (&cqdm, attr.axmodel_num + 2 , " count" , path);
162+ update_cqdm (&cqdm, attr.axmodel_num + 1 , " count" , path);
163163 }
164164#else
165165 llama_layers.resize (attr.axmodel_num );
@@ -172,17 +172,17 @@ struct LLM::Impl {
172172 if (ret != 0 ) { ALOGE (" init axmodel(%s) failed" , llama_layers[i].filename .c_str ()); return false ; }
173173 int remain_cmm = get_remaining_cmm_size ();
174174 sprintf (axmodel_path, " init %d axmodel ok,remain_cmm(%d MB)" , i, remain_cmm);
175- update_cqdm (&cqdm, i + 2 , " count" , axmodel_path);
175+ update_cqdm (&cqdm, i + 1 , " count" , axmodel_path);
176176 }
177177 {
178178 int ret = llama_post.init (attr.filename_post_axmodel .c_str (), -1 );
179179 if (ret != 0 ) { ALOGE (" init post axmodel(%s) failed" , attr.filename_post_axmodel .c_str ()); return false ; }
180180 int remain_cmm = get_remaining_cmm_size ();
181181 sprintf (axmodel_path, " init post axmodel ok,remain_cmm(%d MB)" , remain_cmm);
182- update_cqdm (&cqdm, attr.axmodel_num + 2 , " count" , axmodel_path);
182+ update_cqdm (&cqdm, attr.axmodel_num + 1 , " count" , axmodel_path);
183183 }
184184#endif
185-
185+ printf ( " \n " );
186186 {
187187 _attr.max_token_len = llama_layers[0 ].layer .get_input (" mask" ).nSize / sizeof (unsigned short ) - 1 ;
188188 ALOGI (" max_token_len : %d" , _attr.max_token_len );
@@ -217,9 +217,9 @@ struct LLM::Impl {
217217 ALOGE (" embed_selector.Init(%s, %d, %d) failed" , _attr.filename_tokens_embed .c_str (), _attr.tokens_embed_num , _attr.tokens_embed_size );
218218 return false ;
219219 }
220- update_cqdm (&cqdm, 1 , " count" , " embed_selector init ok" );
220+ update_cqdm (&cqdm, attr. axmodel_num + 2 , " count" , " embed_selector init ok" );
221221 }
222-
222+ printf ( " \n " );
223223 if (!postprocess.load_config (attr.post_config_path )) { ALOGW (" load postprocess config(%s) failed" , attr.post_config_path .c_str ()); }
224224 ALOGI (" LLM init ok" );
225225 return true ;
0 commit comments