[Bugfix] reasoning_parser parameter handling in run_batch.py

[inc-jeong]

Purpose

This PR fixes the issue where the reasoning_parser parameter was not being properly handled in run_batch.py, causing reasoning parser functionality to not work correctly during batch processing.
related issue : #26224

Test Plan

python3 -m vllm.entrypoints.openai.run_batch \
  -i data/input.txt \
  -o data/qwen3_output.txt \
  --model ./model/qwen3-8b \
  --trust_remote_code \
  --reasoning-parser qwen3

Test Result

As-is

{
  "id": "vllm-66cc876c8bde44aea116c5a785caa38d",
  "custom_id": "_2e61202f13bcfee31903706fe5fa92b8",
  "response": {
    "status_code": 200,
    "request_id": "vllm-batch-1531509d67354d59a9c2d2ac4758c5e9",
    "body": {
      "id": "chatcmpl-da3824d262604e729d7926235dfb44fc",
      "object": "chat.completion",
      "created": 1759564798,
      "model": "vllm",
      "choices": [
        {
          "index": 0,
          "message": {
            "role": "assistant",
            "content": "<think>\nOkay, let's tackle this. ... </think>\n\nmain_content: ...",
            "refusal": null,
            "annotations": null,
            "audio": null,
            "function_call": null,
            "tool_calls": [],
            "reasoning_content": null
          },
          "logprobs": null,
          "finish_reason": "stop",
          "stop_reason": null,
          "token_ids": null
        }
      ],
      "service_tier": null,
      "system_fingerprint": null,
      "usage": {
        "prompt_tokens": 8264,
        "total_tokens": 8835,
        "completion_tokens": 571,
        "prompt_tokens_details": null
      },
      "prompt_logprobs": null,
      "prompt_token_ids": null,
      "kv_transfer_params": null
    }
  },
  "error": null
}

To-be

{
  "id": "vllm-9696dc47ef3b42f1a09411f759be19c3",
  "custom_id": "_2e61202f13bcfee31903706fe5fa92b8",
  "response": {
    "status_code": 200,
    "request_id": "vllm-batch-9690b9a5318940aa9a70260bb1143ed0",
    "body": {
      "id": "chatcmpl-48c82ff10f10495eb6560f720efbdc1c",
      "object": "chat.completion",
      "created": 1759564955,
      "model": "vllm",
      "choices": [
        {
          "index": 0,
          "message": {
            "role": "assistant",
            "content": "\n\nmain_content: ...",
            "refusal": null,
            "annotations": null,
            "audio": null,
            "function_call": null,
            "tool_calls": [],
            "reasoning_content": "\nOkay, let's tackle this. ..."
          },
          "logprobs": null,
          "finish_reason": "stop",
          "stop_reason": null,
          "token_ids": null
        }
      ],
      "service_tier": null,
      "system_fingerprint": null,
      "usage": {
        "prompt_tokens": 8264,
        "total_tokens": 8835,
        "completion_tokens": 571,
        "prompt_tokens_details": null
      },
      "prompt_logprobs": null,
      "prompt_token_ids": null,
      "kv_transfer_params": null
    }
  },
  "error": null
}

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors.

You ask your reviewers to trigger select CI tests on top of fastcheck CI.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

If you have any questions, please reach out to us on Slack at https://slack.vllm.ai.

🚀

[gemini-code-assist]

Code Review

This pull request aims to fix an issue with reasoning_parser handling in run_batch.py. The intention is correct, adding validation and passing the parameter to OpenAIServingChat. However, the implementation incorrectly assumes a nested structure for the parsed command-line arguments, which will lead to an AttributeError. The args object is a flat namespace, and the reasoning_parser value should be accessed directly from it for validation. For initializing OpenAIServingChat, it's best to use the vllm_config object, which holds the definitive engine configuration. I've provided suggestions to correct these paths.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

[inc-jeong]

Hello @aarnphm, @chaunceyjiang, could you please take a look when you have a moment?

[chaunceyjiang]

LGTM. Could you write an e2e for this?

[inc-jeong]

@chaunceyjiang , thank you for review.

Could you write an e2e for this?

Do you mean to write the test code in the https://github.com/vllm-project/vllm/tree/main/tests/entrypoints/openai directory?

[chaunceyjiang]

Do you mean to write the test code in the https://github.com/vllm-project/vllm/tree/main/tests/entrypoints/openai directory?

Yes.

[inc-jeong]

@chaunceyjiang ,
add test code in test_run_batch.py

[chaunceyjiang]

Thanks~

[inc-jeong]

@chaunceyjiang , It seems the failure is occurring during the CI stage — could you help me look into it?

[chaunceyjiang]

@chaunceyjiang , It seems the failure is occurring during the CI stage — could you help me look into it?

I retried it once — let’s see if it still fails.