feat: add response_id from API response to prediction results

[xuchi-0808]

Summary

Add the id field from OpenAI-compatible API responses (e.g., chatcmpl-xxx from /v1/chat/completions) to the prediction result JSONL files.

Currently, AISBench discards the id field from API responses. This PR captures it and stores it as response_id in the prediction output, making it possible to trace inference results back to specific API calls.

Changes

1. ais_bench/benchmark/models/output.py — Added response_id field to Output class
2. ais_bench/benchmark/models/api_models/vllm_custom_api_chat.py — Parse id from /v1/chat/completions responses (both text and stream modes)
3. ais_bench/benchmark/models/api_models/vllm_custom_api.py — Parse id from /v1/completions responses (both text and stream modes)
4. ais_bench/benchmark/openicl/icl_inferencer/output_handler/gen_inferencer_output_handler.py — Include response_id in the prediction result dict written to predictions/{dataset}.jsonl

Example output

Each line in predictions/{model_abbr}/{dataset_abbr}.jsonl will now include:

{
  "success": true,
  "uuid": "abc12345",
  "response_id": "chatcmpl-9a8b7c6d5e",
  "origin_prompt": "...",
  "prediction": "...",
  "gold": "..."
}

Design notes

• For streaming responses, the id is captured from the first SSE chunk and preserved
• The response_id field defaults to empty string when not available (backward compatible)
• Both chat API (/v1/chat/completions) and legacy completion API (/v1/completions) are covered

[gemini-code-assist]

Code Review

This pull request introduces a response_id field to track API response identifiers across text and stream responses. It updates the Output class, the VLLM custom API models, and the output handler to capture and propagate this ID. The feedback suggests using getattr to retrieve response_id in the output handler, which provides a more robust and Pythonic approach by avoiding explicit type checks and potential attribute errors.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

Using getattr(output, "response_id", "") is a more robust and Pythonic way to retrieve the response_id attribute. It avoids potential AttributeError if output is a mock object or a custom subclass that does not have the response_id attribute initialized, and it eliminates the need for explicit type checking since getattr safely handles any object type (including str).

            "response_id": getattr(output, "response_id", ""),