发布

  • [OPIK-4335] [SDK] Add G-Eval metric to TypeScript SDK (#5147)

    frostbyte_neo 发布于 2026-02-13 09:11:37 +00:00

    • [OPIK-4335] [SDK] Add G-Eval metric to TypeScript SDK

    Implements G-Eval metric for TypeScript SDK with support for:

    • G-Eval metric class with scoring functionality
    • Parser for extracting scores from LLM responses
    • Template generation for G-Eval prompts
    • Preset configurations for common evaluation scenarios
    • Comprehensive test coverage including unit and integration tests

    Implements OPIK-4335: [TS SDK] Add G-Eval metric

    Co-authored-by: Cursor cursoragent@cursor.com

    • Add concrete classes for judges

    • Add documentation

    • Fix logprobs handling and error logging

    • Add anthropic test for GEval

    • Fix tests

    • fix linting

    • fix anthropik test

    • add custom OpikBaseModel based test

    • use haiku in existing test

    • Refactor Anthropic model test to conditionally skip based on API key availability

    • Safer provider options

    • Fix unit tests


    Co-authored-by: Cursor cursoragent@cursor.com

    下载附件