JudgeLM: Fine-tuned Large Language Models are Scalable Judges

📄 arXiv: 2310.17631v2 📥 PDF

作者: Lianghui Zhu, Xinggang Wang, Xinlong Wang

分类: cs.CL, cs.AI

发布日期: 2023-10-26 (更新: 2025-03-01)

备注: JudgeLM is accepted by ICLR2025. Code is available at https://github.com/baaivision/JudgeLM

🔗 代码/项目: GITHUB

💡 一句话要点

提出JudgeLM以高效评估大语言模型在开放场景中的表现

🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 大语言模型 评估方法 微调技术 数据集构建 性能提升 偏差处理 开放式场景 多模态评估

📋 核心要点

  1. 现有方法在评估大语言模型时面临基准和指标不足的问题，无法全面反映其能力。
  2. 本文提出JudgeLM，通过微调大语言模型作为评估者，利用高质量数据集和新基准进行高效评估。
  3. 实验结果显示，JudgeLM在评估性能上超过90%的教师评估者一致性，且在多种评估场景中表现优异。

📝 摘要（中文）

评估大语言模型（LLMs）在开放式场景中的表现具有挑战性，因为现有基准和指标无法全面衡量其能力。为了解决这一问题，本文提出了JudgeLM，通过对LLMs进行微调，使其成为可扩展的评估者。我们构建了一个全面的大规模高质量数据集，包含任务种子、LLMs生成的答案和GPT-4生成的评判，用于微调高性能评估者，并提出了新的评估基准。实验表明，JudgeLM在多个参数规模下均表现出色，且在现有的PandaLM基准和新基准上均取得了最先进的评估性能。

🔬 方法详解

问题定义：本文旨在解决现有评估大语言模型的基准和指标不足的问题，导致无法全面评估其在开放式场景中的表现。

核心思路：通过微调大语言模型，使其成为可扩展的评估者JudgeLM，利用高质量数据集和新评估基准来提升评估效率和效果。

技术框架：JudgeLM的整体架构包括数据集构建、模型微调和评估基准设计三个主要模块。首先构建包含任务种子和生成答案的数据集，然后对不同规模的模型进行微调，最后在新基准上进行评估。

关键创新：JudgeLM的核心创新在于引入了多种技术来解决微调过程中的偏差问题，包括位置偏差、知识偏差和格式偏差，显著提升了评估性能。

关键设计：在模型微调过程中，采用了交换增强、参考支持和参考丢弃等技术，确保JudgeLM在评估时能够更准确地反映生成答案的质量。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

实验结果显示，JudgeLM在PandaLM基准和新提出的基准上均取得了最先进的评估性能。JudgeLM-7B模型在8个A100 GPU上仅需3分钟即可评估5000个样本，与教师评估者的一致性超过90%，甚至超越了人类评估者之间的一致性。

🎯 应用场景

JudgeLM在多个领域具有广泛的应用潜力，包括教育评估、内容生成质量控制和人机交互系统等。其高效的评估能力能够帮助开发者和研究人员更好地理解和优化大语言模型的表现，推动相关技术的进步和应用。

📄 摘要（原文）

Evaluating Large Language Models (LLMs) in open-ended scenarios is challenging because existing benchmarks and metrics can not measure them comprehensively. To address this problem, we propose to fine-tune LLMs as scalable judges (JudgeLM) to evaluate LLMs efficiently and effectively in open-ended benchmarks. We first propose a comprehensive, large-scale, high-quality dataset containing task seeds, LLMs-generated answers, and GPT-4-generated judgments for fine-tuning high-performance judges, as well as a new benchmark for evaluating the judges. We train JudgeLM at different scales from 7B, 13B, to 33B parameters, and conduct a systematic analysis of its capabilities and behaviors. We then analyze the key biases in fine-tuning LLM as a judge and consider them as position bias, knowledge bias, and format bias. To address these issues, JudgeLM introduces a bag of techniques including swap augmentation, reference support, and reference drop, which clearly enhance the judge's performance. JudgeLM obtains the state-of-the-art judge performance on both the existing PandaLM benchmark and our proposed new benchmark. Our JudgeLM is efficient and the JudgeLM-7B only needs 3 minutes to judge 5K samples with 8 A100 GPUs. JudgeLM obtains high agreement with the teacher judge, achieving an agreement exceeding 90% that even surpasses human-to-human agreement. JudgeLM also demonstrates extended capabilities in being judges of the single answer, multimodal models, multiple answers, multi-turn chat, etc. Code is available at https://github.com/baaivision/JudgeLM.