jescy525/nexus-sft-v1-mix

nexus-sft-v1-mix是一个用于监督微调（SFT）的数据集，属于AETHER家族，具体为mix组。数据格式为JSONL ChatML消息，每个条目包含消息列表（角色包括system、user或assistant，内容为文本）、任务类型标签（如function_calling、code、reasoning_cot等）、来源数据集ID、语言（如英语en、法语fr）和系统来源信息。数据集应用了MinHash去重技术，阈值为0.85，以降低重复内容。该数据集由prepare_sft.py管道于2026年5月25日生成，适用于文本生成和对话任务，支持英语和法语，规模在10万到100万条之间。

nexus-sft-v1-mix is a supervised fine-tuning (SFT) dataset from the AETHER family, specifically the mix group. The data is formatted as JSONL ChatML messages, with each entry containing a list of messages (roles: system, user, or assistant, with text content), task type tags (e.g., function_calling, code, reasoning_cot), source dataset ID, language (e.g., en for English, fr for French), and system source information. MinHash deduplication is applied with a threshold of 0.85 to reduce redundancy. Generated by the prepare_sft.py pipeline on May 25, 2026, it is suitable for text-generation and conversational tasks, supports English and French, and has a size between 100K and 1M entries.