"finish_reason": "length" --> how to increase max_new_tokens

Nobody has claimed this yet.

Hello,

I am using the fastchat openai-like api to serve model but sometimes the model stops because of output length limit, even though the max size (4096) is not reached yet.

```
JSON: {
"id": "chatcmpl-aMCSDBC6eBQx74i9XWuCBt",
"object": "chat.completion",
"created": 1692929522,
"model": "vicuna-7b-v1.5",
"choices": [
{
"index": 0,
"message": {
"role": "assistant",
"content": "some long text"
},
*"finish_reason": "length"*
}
],
"usage": {
"prompt_tokens": 2623,
"total_tokens": 3134,
"completion_tokens": 511
}
}

```

I tried adding `max_new_tokens` to the request but it didn't work.

```
openai.api_key = "EMPTY"
openai.api_base = "http://localhost:8000/v1"
response = openai.ChatCompletion.create(
model='vicuna-7b-v1.5',
messages=messages,
max_new_tokens=2000
)

```

No file or test is named. Start by tracing the FastChat OpenAI-like API's handling of token-limit request parameters and compare the reported max_new_tokens request with the returned usage. Done means determining whether the behavior needs a code fix or clearer documentation of the supported parameter.

Written by the indexing model from the issue text.