FuseAI/FuseChat-3.0-SFT-Data
收藏Hugging Face2025-02-07 更新2025-02-15 收录
下载链接：
https://hf-mirror.com/datasets/FuseAI/FuseChat-3.0-SFT-Data
下载链接
链接失效反馈官方服务：
资源简介：
FuseChat-3.0是一个通过隐式模型融合技术提升大型语言模型性能的数据集。它包含了指令遵循、一般对话、数学、编码和中文语言等类别的数据，用于训练和优化目标LLM。数据集的构建基于多个开源社区数据集，经过筛选和预处理。训练过程采用监督微调（SFT）和直接偏好优化（DPO）两个阶段。该数据集展示了在不同任务上的性能提升效果。
FuseChat-3.0 is a dataset designed to enhance the performance of large language models (LLMs) through implicit model fusion technology. It includes data for instruction following, general conversation, mathematics, coding, and Chinese language, which are used to train and optimize the target LLM. The dataset is constructed from multiple open-source community datasets through filtering and preprocessing. The training process involves two stages: Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO). The dataset demonstrates performance improvements on various tasks.
提供机构：
FuseAI
