lthn/livebench-data_analysis

1---2dataset_info:3  features:4  - name: question_id5    dtype: string6  - name: category7    dtype: string8  - name: turns9    sequence: string10  - name: ground_truth11    dtype: string12  - name: task13    dtype: string14  - name: livebench_release_date15    dtype: timestamp[s]16  - name: livebench_removal_date17    dtype: string18  splits:19  - name: test20    num_bytes: 30584821    num_examples: 15022  download_size: 14479623  dataset_size: 30584824configs:25- config_name: default26  data_files:27  - split: test28    path: data/test-*29arxiv: 2406.1931430---31# Dataset Card for "livebench/data_analysis"32LiveBench is a benchmark for LLMs designed with test set contamination and objective evaluation in mind. It has the following properties:33- 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.34- Each question has verifiable, objective ground-truth answers, allowing hard questions to be scored accurately and automatically, without the use of an LLM judge.35- LiveBench currently contains a set of 18 diverse tasks across 6 categories, and we will release new, harder tasks over time.36This is the instruction_following category of livebench.37 38See 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).39