Nemotron-Terminal-Synthetic-Tasks
收藏资源简介:
Terminal-Corpus 是一个基于 Linux 终端环境的技能导向型合成任务数据集,专为评估和训练自主智能体而设计。数据集采用 cc-by-4.0 许可,包含英语编写的代码相关任务,规模在10万至100万样本之间。每个任务遵循严格的四组件架构:1) 自然语言描述的指令文件(instruction.md);2) 基于9个预构建领域专用Docker镜像的环境上下文;3) 使用pytest框架的自动化验证套件;4) 包含黄金执行路径的解决方案轨迹(solution/)。数据集通过标准化容器环境、程序化验证机制和结构化任务设计,支持大规模终端操作任务的自动化测试与学习。
Terminal-Corpus is a skill-oriented synthetic task dataset based on the Linux terminal environment, specifically designed for evaluating and training autonomous AI agents. This dataset is licensed under CC-BY-4.0, contains code-related tasks written in English, and has between 100,000 and 1,000,000 samples. Each task follows a strict four-component architecture: 1) Instruction file (instruction.md) described in natural language; 2) Environmental context based on 9 pre-built domain-specific Docker images; 3) Automated validation suite using the pytest framework; 4) Solution trajectory (solution/) containing the golden execution path. The dataset supports automated testing and learning for large-scale terminal operation tasks through standardized container environments, programmatic validation mechanisms, and structured task design.



