def play_a_match_single()

in fastchat/llm_judge/common.py [0:0]

• 37 lines of code
• 4 McCabe index (conditional complexity)

def play_a_match_single(match: MatchPair, output_file: str):
    question, model, answer, judge, ref_answer, multi_turn = (
        match.question,
        match.model,
        match.answer,
        match.judge,
        match.ref_answer,
        match.multi_turn,
    )

    if judge.prompt_template["type"] == "single":
        score, user_prompt, judgment = run_judge_single(
            question, answer, judge, ref_answer, multi_turn=multi_turn
        )

        question_id = question["question_id"]
        turn = 1 if not multi_turn else 2
        result = {
            "question_id": question_id,
            "model": model,
            "judge": (judge.model_name, judge.prompt_template["name"]),
            "user_prompt": user_prompt,
            "judgment": judgment,
            "score": score,
            "turn": turn,
            "tstamp": time.time(),
        }
        print(
            f"question: {question_id}, turn: {turn}, model: {model}, "
            f"score: {score}, "
            f"judge: {(judge.model_name, judge.prompt_template['name'])}"
        )
    else:
        raise ValueError(f"invalid judge type: {judge['type']}")

    if output_file:
        os.makedirs(os.path.dirname(output_file), exist_ok=True)
        with open(output_file, "a") as fout:
            fout.write(json.dumps(result) + "\n")

    return result