gszauer/Gab100MFinetune

Gab 100M Finetune 是一个用于监督微调（SFT）的英文指令跟随对话数据集，包含用户与助手之间的对话，旨在用于语言模型的聊天和指令调优。数据涵盖单轮交换（一个用户请求配对一个助手回复）和多轮对话（从简短跟进到较长的主题讨论）。对话主题广泛，包括教育（如阅读、数学、科学）、生活技能（如烹饪、财务、健康）、编程（如Web开发、编码帮助）和常识（如科学解释、历史）。回答风格实用，常使用分步说明、列表、示例和代码片段。数据以纯文本文件存储，使用聊天角色标记（如<|user|>、<|assistant|>）和结束标记（<|end|>）分隔对话轮次。Gab 100M Finetune is a supervised fine-tuning (SFT) dataset of English instruction-following conversations between a user and a helpful assistant, intended for chat and instruction tuning of language models. The data includes single-turn exchanges (one user request paired with one assistant reply) and multi-turn conversations (ranging from short follow-ups to longer topic discussions). Topics cover a broad range of areas such as education (e.g., reading, math, science), life skills (e.g., cooking, finance, health), programming (e.g., web development, coding help), and general knowledge (e.g., science explainers, history). Responses are practical and explanatory, often using step-by-step instructions, lists, examples, and code snippets. Conversations are stored as plain-text files with chat role markers (e.g., <|user|>, <|assistant|>) and an end-of-turn token (<|end|>) to delimit turns.