匠数科技大模型sft数据集

匠数大模型SFT数据集由匠数科技整理，包含1138万条中文和276万条英文高质量SFT数据，涵盖50类任务。数据经统一格式化、清洗及严格内容审核，确保安全可靠。提供中英文类别关键词，适用于大模型指令微调、多任务学习与内容安全研究，支持开放学术与产业应用。

Jiangshu Large Model SFT Dataset, curated by Jiangshu Technology, contains 11.38 million high-quality Chinese and 2.76 million high-quality English SFT data entries, covering 50 task categories. The dataset has undergone unified formatting, data cleaning and strict content auditing to ensure its security and reliability. It provides category keywords in both Chinese and English, and is suitable for large language model instruction fine-tuning, multi-task learning and content security research, supporting open academic and industrial applications.

搜集汇总

背景与挑战

背景概述

该数据集是一个高质量的大模型SFT数据集，由匠数科技整理，包含约1138万条中文和276万条英文数据，覆盖50类任务。数据经过统一格式化、清洗和严格内容审核，确保安全可靠，适用于大模型指令微调、多任务学习和内容安全研究。