`do_sample` is always set to `FALSE` when using Hugging Face API

The Hugging Face API sample script contains the following lines:

output_ids = model.generate(
        **inputs,
        do_sample=True if args.temperature > 1e-5 else False,
...

and

parser.add_argument("--temperature", type=float, default=0.7)

So clearly do_sample should be on by default, but actually running the script produce the following message:

/envs/pytorch/lib/python3.10/site-packages/transformers/generation/configuration_utils.py:362: 
UserWarning: `do_sample` is set to `False`. However, `temperature` is set to `0.9` -- this flag is only used in sample-based generation modes. 
You should set `do_sample=True` or unset `temperature`. This was detected when initializing the generation config instance, which means the corresponding file may hold incorrect parameterization and should be fixed.

In fact, this message would appear even if I modify the sample script and set do_sample=True in the model.generate() call. I have verified that this is true for fschat=0.2.29-0.2.32 and the newest version of transformers. I have also verified that the Hugging Face version Llama-2 does not show this warning.

Start with fastchat/serve/huggingface_api.py and its model.generate call, then reproduce the warning using the script's default --temperature. Compare the effective generation settings with the reported warning; the issue is done when the sample no longer reports do_sample=False while temperature is nonzero.