运行baichuan2报错

(base) ps@ps:/data/model_hub/llm$ CUDA_VISIBLE_DEVICES=1,2 python3 -m fastchat.serve.cli --model-path Baichuan2-13B-Chat --num-gpus 2
[2024-01-31 14:19:32,877] [INFO] [real_accelerator.py:133:get_accelerator] Setting ds_accelerator to cuda (auto detect)
Loading checkpoint shards: 100%|████████████████████████████████████████████████████████████████| 6/6 [00:14<00:00, 2.39s/it]
<reserved_102> : 你好
<reserved_103> : Traceback (most recent call last):
File "", line 198, in _run_module_as_main
File "", line 88, in _run_code
File "/data/FastChat/fastchat/serve/cli.py", line 256, in
main(args)
File "/data/FastChat/fastchat/serve/cli.py", line 191, in main
chat_loop(
File "/data/FastChat/fastchat/serve/inference.py", line 373, in chat_loop
outputs = chatio.stream_output(output_stream)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/data/FastChat/fastchat/serve/cli.py", line 54, in stream_output
for outputs in output_stream:
File "/data/miniconda3/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 56, in generator_context
response = gen.send(request)
^^^^^^^^^^^^^^^^^
File "/data/FastChat/fastchat/serve/inference.py", line 161, in generate_stream
indices = torch.multinomial(probs, num_samples=2)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: probability tensor contains either inf, nan or element < 0

Start with fastchat/serve/inference.py at generate_stream, especially the torch.multinomial call at line 161, and reproduce the provided Baichuan2 command with two GPUs. Trace why the probability tensor contains invalid values and verify the fix by confirming interactive generation completes without the reported RuntimeError.