caskcsg/LongMagpie_singledoc_longcontext_dataset

LongMagpie是一个自合成框架，用于自动生成大规模长上下文指令数据。该框架利用已对齐的长上下文大型语言模型（LLMs）自动生成上下文相关的查询，通过收集这些文档-查询对和模型的响应，LongMagpie能够产生高质量的指令，无需人工干预。实验表明，LongMagpie在长上下文任务上取得了领先性能，同时在短上下文任务上也保持了竞争力。

LongMagpie is a self-synthesis framework for automatically generating large-scale long-context instruction data. The framework utilizes aligned long-context large language models (LLMs) to auto-generate contextually relevant queries. By harvesting these document-query pairs and the models responses, LongMagpie produces high-quality instructions without human effort. Experiments demonstrate that LongMagpie achieves leading performance on long-context tasks while maintaining competitive performance on short-context tasks.