[Usage]: vllm infer with 2 * Nvidia-L20, output repeat !!!!

[RoyaltyLJW]

Your current environment

vllm==0.6.1

How would you like to use vllm

Steps to reproduce
This happens to Qwen2.5-32B-Instruct-AWQ
The problem can be reproduced with the following steps:

start vllm service

exec python3 -m vllm.entrypoints.openai.api_server\
   --served-model-name ${model_name}\
   --model ./${model_name}\
   --port 8890 \
   --quantization awq \
   --tensor-parallel-size 2 \
   --enable_auto_tool_choice \
   --tool-call-parser hermes 1>vllm.log 2>&1 &

infer

system_prompt = "你是一个安全专家，你的任务是根据用户的输入，以markdown格式返回结果。"
query = "什么是SSRF漏洞"
messages = [{"role": "system", "content": system_prompt}]
messages.append({"role": "user", "content": query})
 completion = client.chat.completions.create(
      model=model_name,
      messages=messages,
      temperature=0.1,
      top_p=0.9,
      max_tokens=4096,
      tools=[],
      extra_body={
          "repetition_penalty": 1.05,
      },
  )
        
req_id = completion.id
total_token = completion.usage.total_tokens
completion_token = completion.usage.completion_tokens
prompt_tokens = completion.usage.prompt_tokens

output results
The results are expected to be ...
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

with 2 A30, I didn't succeed in reproducing. But with L20，nearly 30% can reproduce
is the problem related to GPU type or cuda driver?
Looking for help

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[jeejeelee]

Maybe similar issue with #10656, cc @mgoin