[Bug]: top_logrpobs generating a WARNING

[FilippoBoni1921]

Your current environment

Using vllm v0.7.3

🐛 Describe the bug

When making a request via the OpenAI API using the top_logprobs atgument, the logging of the request shows a warning that says:

The following fields were present in the request but ignore: {'top_logprobs'}

This behaviour should be reserved to the args in the request that are not expected by the ChatCompletionRequest() class, but top_logprobs is actually part of its args.

This warning is raised by the function __log_extra_fields__() in vllm/entrypoints/openai/protocol.py. The warning started to appear in version v0.7.1 and I think was introduced in the PR #12420 (I would like to tag @maxdebayser for more some 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.

[DarkLight1337]

You made a typo in the argument name. It should be top_logprobs, not top_logrpobs.

[FilippoBoni1921]

You made a typo in the argument name. It should be top_logprobs, not top_logrpobs.

Yes sorry, but just in reporting the issue. The request has the correct name

[maxdebayser]

@FilippoBoni1921, thanks for reporting. The PR that you mentioned was intended to suppress this kind of erroneous message so I was surprised to see this, but I can reproduce the problem. It looks like the message actually refers to the top_logprobs field of the output because I get one warning per output token. I'm going to look into what is going wrong here.

curl http://localhost:8000/v1/chat/completions    -H "Content-Type: application/json"    -d '{
     "model": "meta-llama/Llama-3.2-3B-Instruct",
     "messages": [{"role": "user", "content": "how to make cheesecake"}],
     "top_logprobs": 5,
     "logprobs": true,
     "max_tokens": 1,
     "temperature": 0
}' | jq
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100  1097  100   879  100   218     10      2  0:01:49  0:01:22  0:00:27   220
{
  "id": "chatcmpl-aac8f52c8d3848b6a4d6e242ee5f88de",
  "object": "chat.completion",
  "created": 1740597842,
  "model": "meta-llama/Llama-3.2-3B-Instruct",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "reasoning_content": null,
        "content": "Making",
        "tool_calls": []
      },
      "logprobs": {
        "content": [
          {
            "token": "Making",
            "logprob": -0.1470043957233429,
            "bytes": [
              77,
              97,
              107,
              105,
              110,
              103
            ],
            "top_logprobs": [
              {
                "token": "Making",
                "logprob": -0.1470043957233429,
                "bytes": [
                  77,
                  97,
                  107,
                  105,
                  110,
                  103
                ]
              },
              {
                "token": "Here",
                "logprob": -2.1470043659210205,
                "bytes": [
                  72,
                  101,
                  114,
                  101
                ]
              },
              {
                "token": "A",
                "logprob": -4.6470046043396,
                "bytes": [
                  65
                ]
              },
              {
                "token": "Che",
                "logprob": -5.1470046043396,
                "bytes": [
                  67,
                  104,
                  101
                ]
              },
              {
                "token": "The",
                "logprob": -5.7720046043396,
                "bytes": [
                  84,
                  104,
                  101
                ]
              }
            ]
          }
        ]
      },
      "finish_reason": "length",
      "stop_reason": null
    }
  ],
  "usage": {
    "prompt_tokens": 40,
    "total_tokens": 41,
    "completion_tokens": 1,
    "prompt_tokens_details": null
  },
  "prompt_logprobs": null
}

[maxdebayser]

I've opened a PR. It's a hack but it seems to work.