arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2406.04770cs.CLcs.AI
收藏

WildBench:利用真实用户在野外场景中的挑战性任务对大语言模型进行基准测试

WildBench: Benchmarking LLMs with Challenging Tasks from Real Users in the Wild

Allen Institute for AI(艾伦人工智能研究所);University of Washington(华盛顿大学)

Bill Yuchen Lin, Yuntian Deng, Khyathi Chandu, Faeze Brahman, Abhilasha Ravichander, Valentina Pyatkin, Nouha Dziri, Ronan Le Bras, Yejin Choi

AI总结:

本研究推出基于真实用户挑战性任务的LLM基准测试框架WildBench,提出WB-Reward和WB-Score两项评估指标,有效缓解长度偏差,与人类评分相关性优于现有基准。

摘要:

我们推出WildBench——一个自动化评估框架,旨在利用具有挑战性的真实世界用户查询对大语言模型(LLM)进行基准测试。WildBench包含1024项任务,这些任务是从超过100万条人类与聊天机器人的对话日志中精心筛选出来的。\n为了借助WildBench实现自动化评估,我们开发了两项指标:WB-Reward和WB-Score,二者均可通过GPT-4-turbo等先进LLM计算得出。WildBench评估采用针对特定任务的检查清单,对模型输出进行系统性评估,并提供结构化解释来佐证评分与比较结果,从而提升自动评判的可靠性与可解释性。WB-Reward采用细粒度的模型响应两两比较机制,生成五种可能结果:优得多、略优、略差、差得多或平局。与以往采用单一基线模型的评估不同,我们选取了三个性能水平各异的基线模型,以确保两两评估的全面性。此外,我们提出了一种缓解长度偏差的简单方法:若获胜响应的字符数比落败响应多出K个以上,则将“略优/略差”的结果转换为“平局”。\nWB-Score则单独评估模型输出的质量,是一种快速且低成本的评估指标。WildBench的测试结果显示,其与Chatbot Arena中针对困难任务的人类投票Elo评分具有强相关性。具体而言,WB-Reward与排名靠前模型的皮尔逊相关系数达到0.98。此外,WB-Score的相关系数达到0.95,在长度控制胜率方面优于ArenaHard的0.91和AlpacaEval2.0的0.89,也优于常规胜率的0.87。

英文摘要:

We introduce WildBench, an automated evaluation framework designed to benchmark large language models (LLMs) using challenging, real-world user queries. WildBench consists of 1,024 tasks carefully selected from over one million human-chatbot conversation logs. For automated evaluation with WildBench, we have developed two metrics, WB-Reward and WB-Score, which are computable using advanced LLMs such as GPT-4-turbo. WildBench evaluation uses task-specific checklists to evaluate model outputs systematically and provides structured explanations that justify the scores and comparisons, resulting in more reliable and interpretable automatic judgments. WB-Reward employs fine-grained pairwise comparisons between model responses, generating five potential outcomes: much better, slightly better, slightly worse, much worse, or a tie. Unlike previous evaluations that employed a single baseline model, we selected three baseline models at varying performance levels to ensure a comprehensive pairwise evaluation. Additionally, we propose a simple method to mitigate length bias, by converting outcomes of ``slightly better/worse'' to ``tie'' if the winner response exceeds the loser one by more than $K$ characters. WB-Score evaluates the quality of model outputs individually, making it a fast and cost-efficient evaluation metric. WildBench results demonstrate a strong correlation with the human-voted Elo ratings from Chatbot Arena on hard tasks. Specifically, WB-Reward achieves a Pearson correlation of 0.98 with top-ranking models. Additionally, WB-Score reaches 0.95, surpassing both ArenaHard's 0.91 and AlpacaEval2.0's 0.89 for length-controlled win rates, as well as the 0.87 for regular win rates.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑