lthn/livebench-data_analysis
收藏资源简介:
--- dataset_info: features: - name: question_id dtype: string - name: category dtype: string - name: turns sequence: string - name: ground_truth dtype: string - name: task dtype: string - name: livebench_release_date dtype: timestamp[s] - name: livebench_removal_date dtype: string splits: - name: test num_bytes: 305848 num_examples: 150 download_size: 144796 dataset_size: 305848 configs: - config_name: default data_files: - split: test path: data/test-* arxiv: 2406.19314 --- # Dataset Card for "livebench/data_analysis" LiveBench is a benchmark for LLMs designed with test set contamination and objective evaluation in mind. It has the following properties: - LiveBench is designed to limit potential contamination by releasing new questions monthly, as well as having questions based on recently-released datasets, arXiv papers, news articles, and IMDb movie synopses. - Each question has verifiable, objective ground-truth answers, allowing hard questions to be scored accurately and automatically, without the use of an LLM judge. - LiveBench currently contains a set of 18 diverse tasks across 6 categories, and we will release new, harder tasks over time. This is the instruction_following category of livebench. See more in our [paper](https://arxiv.org/abs/2406.19314), [leaderboard](https://livebench.ai/), and [datasheet](https://github.com/LiveBench/LiveBench/blob/main/docs/DATASHEET.md).
--- 数据集信息: 特征: - 名称:问题ID(question_id),数据类型:字符串 - 名称:分类(category),数据类型:字符串 - 名称:轮次(turns),数据类型:字符串序列 - 名称:标准答案(ground_truth),数据类型:字符串 - 名称:任务(task),数据类型:字符串 - 名称:LiveBench发布日期(livebench_release_date),数据类型:秒级时间戳 - 名称:LiveBench移除日期(livebench_removal_date),数据类型:字符串 划分: - 名称:测试集(test),字节数:305848,样本数:150 下载大小:144796 字节 数据集总大小:305848 字节 配置项: - 配置名称:default,数据文件: - 划分:test,路径:data/test-* arXiv编号:2406.19314 --- # "livebench/data_analysis"数据集卡片 LiveBench是一款专为大语言模型(Large Language Model)设计的基准测试集,其研发初衷为规避测试集污染并实现客观评估。该基准具备以下特性: - 本基准通过每月更新全新测试题目、并基于最新发布的数据集、arXiv论文、新闻文章以及IMDb电影剧情简介生成试题,以最大程度降低测试集污染风险。 - 每道题目均配备可验证的客观标准答案,无需借助大语言模型作为评判器,即可实现复杂题目的精准自动评分。 - 当前LiveBench包含覆盖6个分类的18项多样化任务,后续还将陆续发布难度更高的全新任务。 本数据集为LiveBench的指令遵循(instruction_following)分类子集。 更多详情可参阅我们的[研究论文](https://arxiv.org/abs/2406.19314)、[排行榜页面](https://livebench.ai/)以及[数据集说明文档](https://github.com/LiveBench/LiveBench/blob/main/docs/DATASHEET.md)。



