LangSmith 支持分类和数值指标,在编写自定义评估器时可以返回任一种。 要将评估器结果记录为数值指标,必须返回为:
  • (仅 Python)intfloatbool
  • 形式为 {"key": "metric_name", "score": int | float | bool} 的字典
要将评估器结果记录为分类指标,必须返回为:
  • (仅 Python)str
  • 形式为 {"key": "metric_name", "value": str | int | float | bool} 的字典
以下是一些示例:
  • Python:需要 langsmith>=0.2.0
  • TypeScript:langsmith@0.1.32 及更高版本支持多个分数
def numerical_metric(inputs: dict, outputs: dict, reference_outputs: dict) -> float:
    # Evaluation logic...
    return 0.8
    # Equivalently
    # return {"score": 0.8}
    # Or
    # return {"key": "numerical_metric", "score": 0.8}

def categorical_metric(inputs: dict, outputs: dict, reference_outputs: dict) -> str:
    # Evaluation logic...
    return "english"
    # Equivalently
    # return {"key": "categorical_metric", "score": "english"}
    # Or
    # return {"score": "english"}

相关


Connect these docs programmatically to Claude, VSCode, and more via MCP for real-time answers.