| engine_path: "data" | |
| hf_tokenizer: "TinyLlama/TinyLlama-1.1B-Chat-v1.0" | |
| kv_cache_free_gpu_mem_fraction: 0.9 | |
| enable_kv_cache_reuse: true | |
| enable_chunked_context: False | |
| port: 50051 | |
| fsm_cache_dir: "/cache/model/fsm_cache" | |
| max_num_tokens: 8192 | |
| added_tokens { | |
| tokens { | |
| content: "<unk>" | |
| single_word: False | |
| lstrip: False | |
| rstrip: False | |
| normalized: False | |
| special: True | |
| } | |
| tokens { | |
| content: "<s>" | |
| single_word: False | |
| lstrip: False | |
| rstrip: False | |
| normalized: False | |
| special: True | |
| } | |
| tokens { | |
| content: "</s>" | |
| single_word: False | |
| lstrip: False | |
| rstrip: False | |
| normalized: False | |
| special: True | |
| } | |
| } |