open-thoughts/TaskTrove

TaskTrove是一个开源的任务数据集集合，由OpenThoughts-Agent团队发布。它包含来自100多个任务源的超过750,000个独特任务，涵盖了流行的RL和SFT训练目标，如SWE-Smith、R2EGym和SWE-Re-Bench等。数据集中的任务分为带有验证器和不带验证器两种类型，分别适用于RL训练和评估以及SFT/数据生成。TaskTrove是AgentTrove的任务补充，AgentTrove中的代理轨迹是通过使用Harbor框架对这些任务数据集运行模型生成的。数据集的结构保留了原始HuggingFace仓库的文件和目录结构，每个源数据集都存储为一个子目录。

TaskTrove is an open-source collection of agentic task datasets, released by the OpenThoughts-Agent team. It contains over 750,000 unique tasks drawn from over 100 task sources, including popular RL and SFT training targets such as SWE-Smith, R2EGym, and SWE-Re-Bench. Tasks in TaskTrove are categorized into those with verifiers (for RL training and evaluation) and those without (for SFT/datagen). TaskTrove serves as the task complement to AgentTrove, where agent traces in AgentTrove were generated by running models against these task datasets using the Harbor framework. The repository structure preserves the original HuggingFace repo files and directories, with each source dataset stored as a subdirectory.