Nemotron-SFT-SWE-v2
收藏资源简介:
Nemotron-SWE-v2 是一个专为提升大型语言模型在软件工程任务(SWE-Bench 风格任务)上的性能而设计的指令调优数据集。该数据集包含通过 OpenHands 框架收集的代理轨迹和针对特定子任务(如代码定位、代码修复和测试生成)的无代理 SWE 子集。数据集总规模约为 256,254 个样本,存储量约为 17GB,采用 JSONL 格式。数据集适用于构建自主软件工程代理和代码辅助工具的 LLM 工程师和研究团队,可用于监督微调、模型蒸馏以及代理策略的基准测试和调试。数据集采用混合自动化和合成方法收集和标注,包含文本和元数据。许可证包括 Creative Commons Attribution 4.0 International (CC-BY 4.0)、Apache 2.0、MIT、BSD-3 和 BSD-2。
Nemotron-SWE-v2 is an instruction-tuning dataset specifically developed to improve the performance of large language models (LLMs) on software engineering tasks (SWE-Bench style tasks). This dataset contains agent trajectories collected through the OpenHands framework, as well as an agent-free SWE subset targeting specific subtasks including code localization, code repair, and test generation. The dataset has a total scale of approximately 256,254 samples, with a storage size of around 17GB, and is stored in JSONL format. It is intended for LLM engineers and research teams that build autonomous software engineering agents and code assistance tools, and can be applied for supervised fine-tuning, model distillation, as well as benchmarking and debugging of agent policies. The dataset is collected and annotated using a hybrid automated and synthetic approach, and includes both text and metadata. Its available licenses include Creative Commons Attribution 4.0 International (CC-BY 4.0), Apache 2.0, MIT, BSD-3, and BSD-2.



