Lite-Coder/LiteCoder-SFT-Terminal-preview

LiteCoder-SFT-Terminal-preview 是一个包含940条合成轨迹的数据集，旨在增强模型的代理能力，是我们开发中小型代码代理模型的最新努力的一部分。通过完全依赖合成管道（不转换任何现有数据集），我们能够在具有挑战性的 Terminal Bench 上取得显著成果，以极高的数据效率匹配领先开源模型的性能。数据集构建流程包括任务采样、可行性检查、环境准备和轨迹生成。

LiteCoder-SFT-Terminal-preview is a dataset consisting of 940 synthetic trajectories for enhancing models agentic abilities and is part of our recent efforts to develop capable small and medium-sized code agent models. By relying entirely on a fully synthetic pipeline—without converting any existing datasets—we were able to secure significant gains on the challenging Terminal Bench, matching the performance of leading open-source models with extreme data efficiency. The data construction pipeline involves Task Sampling, Feasibility Check, Environment Preparation, and Trajectory Generation.