Open access dataset for machine-generated text detection in Russian

Open access dataset for machine-generated text detection in Russian. The dataset consists of a zip-archive with 450k documents from Russian Wikipedia corpus as a real part for detection task and 450k machine-generated documents. Generated documents contains 150k samples for each language model (SberAI-small, SberAI-large, Facebook-XGLM) with 3 sampling methods (top-k, top-p, pure) and 2 priming sequence strategy (word, sentence). We provide you with access both to the data with the combined generated samples and to the individual parts of the documents generated by the different language models.

面向俄语机器生成文本检测的开源数据集。本数据集以压缩包形式提供，包含来自俄语维基百科语料库的45万篇真实文档（作为检测任务的真实样本），以及45万篇机器生成文档。生成文档针对每款语言模型（SberAI-small、SberAI-large、Facebook-XGLM）均提供15万条样本，涵盖3种采样策略（top-k、top-p、pure）与2种提示序列策略（word、sentence）。我们同时提供合并后的机器生成样本数据集，以及不同语言模型各自生成的单类别文档子集的访问权限。