File tree Expand file tree Collapse file tree
projects/llm_framework/main_llm Expand file tree Collapse file tree Original file line number Diff line number Diff line change 1+ {
2+ "mode" :" qwen2.5-1.5B-ax630c" ,
3+ "type" :" llm" ,
4+ "capabilities" :[
5+ " text_generation" ,
6+ " chat"
7+ ],
8+ "input_type" :[
9+ " llm.utf-8" ,
10+ " llm.utf-8.stream" ,
11+ " llm.chat_completion" ,
12+ " llm.chat_completion.stream"
13+ ],
14+ "output_type" :[
15+ " llm.utf-8" ,
16+ " llm.utf-8.stream"
17+ ],
18+ "mode_param" :{
19+ "tokenizer_type" :2 ,
20+ "filename_tokenizer_model" :" http://localhost:8080" ,
21+ "filename_tokens_embed" :" model.embed_tokens.weight.bfloat16.bin" ,
22+ "filename_post_axmodel" :" qwen2_post.axmodel" ,
23+ "template_filename_axmodel" :" qwen2_p128_l%d_together.axmodel" ,
24+ "b_use_topk" :false ,
25+ "b_bos" :false ,
26+ "b_eos" :false ,
27+ "axmodel_num" :28 ,
28+ "tokens_embed_num" :151936 ,
29+ "tokens_embed_size" :1536 ,
30+ "b_use_mmap_load_embed" :true ,
31+ "b_dynamic_load_axmodel_layer" :false
32+ }
33+ }
You can’t perform that action at this time.
0 commit comments