LM-Polygraph/triviaqa

这是triviaqa数据集的预处理版本，用于LM-Polygraph的基准测试。数据集包含多个子集，每个子集包含训练和测试两个分割，每个分割包含两个字符串列：input和output。数据集的创建目的是将数据集创建代码与基准测试代码分离。数据集来源于https://huggingface.co/datasets/trivia_qa，并通过build_dataset.py脚本进行处理。数据集包含与其源数据集相同的偏见、风险和限制。

This is a preprocessed version of the triviaqa dataset, specifically tailored for benchmarks in LM-Polygraph. The dataset includes multiple configurations (subsets) such as continuation, empirical_baselines, ling_1s, verb_1s_top1, verb_1s_topk, verb_2s_cot, verb_2s_top1, and verb_2s_topk. Each configuration has its own train and test splits, with corresponding number of examples and bytes. The dataset features include input and output columns, both of which are strings. The dataset is curated by LM-Polygraph and is intended for use in performing benchmarks, not for further preprocessing. The source data is collected from the trivia_qa dataset on Hugging Face and processed using a script from the LM-Polygraph repository. The dataset inherits the biases, risks, and limitations of its source dataset.