camel-ai/SETA-Env
收藏资源简介:
SETA-Env 是一个开源的可验证强化学习终端环境数据集,专为社区训练和评估而设计。该数据集包含两个主要子集:SETA_Synth(合成任务)和SETA_Evolve(终端代理任务的演化变体)。当前版本共包含4567个环境,其中SETA_Synth有3255个任务,SETA_Evolve有1312个任务。每个任务都打包为一个自包含的Harbor风格任务目录,包括运行任务、构建环境、执行参考解决方案和运行测试所需的文件。数据集结构清晰,任务文件夹包含task.toml、instruction.md、environment、solution和tests等子目录。数据来源多样,包括ask_ubuntu、kaggle_notebook、nl2bash、stack_overflow和unix_linux_se等,并提供了详细的来源细分统计。数据集支持混合许可证,具体取决于数据来源。
SETA-Env is an open-source verifiable RL terminal environment dataset for community training and evaluation. This release contains two top-level subsets: SETA_Synth (synthesized tasks) and SETA_Evolve (evolved variants of terminal-agent tasks). The current release includes 4567 environments: 3255 in SETA_Synth and 1312 in SETA_Evolve. Each task is packaged as a self-contained Harbor-style task directory with the files needed to run the task, build its environment, execute a reference solution, and run tests. The dataset structure includes task folders with components such as task.toml, instruction.md, environment, solution, and tests. Sources include ask_ubuntu, kaggle_notebook, nl2bash, stack_overflow, and unix_linux_se, with detailed breakdowns provided. The dataset has mixed licensing based on the source.




