bug: Vicuna performance is not great

[gaocegege]

us-central1-docker.pkg.dev/nth-guide-378813/modelzai/llm-vicuna-7b:23.06.12

Output:

{"id":"30921e14-f339-403b-bc9a-f5a621013d60","object":"chat","created":"2023-06-06T12:08:11.786163","model":"lmsys/vicuna-7b-delta-v1.1","usage":{"prompt_tokens":40,"
completion_tokens":99,"total_tokens":139},"choices":[{"message":{"content":"\"\r宇馬{[ assum��language  \ruwe\"\r��language  \ruwe\"\r��language  \ruwe\"\r��language 
 \ruwe\"\r��language  \ruwe\"\r��language  \ruwe\"\r��language  \ruwe\"\r��language  \ruwe\"\r��language  \ruwe\"\r��language  \ruwe\"\r��language  \ruwe\"\r��languag
e  \ruwe\"\r��language  \ruwe\"\r��language  \ruwe\"\r��language  \ruwe\"\r��language  \ruwe","role":"assistant","name":""},"index":0,"finish_reason":"length"}]} 

[kemingy]

https://github.com/lm-sys/FastChat/blob/main/docs/vicuna_weights_version.md

[gaocegege]

We are using https://huggingface.co/lmsys/vicuna-13b-delta-v1.1, the format may be wrong in LLMSpec.