Finnish-NLP/dolci-instruct-sft-fi
收藏资源简介:
Dolci Instruct SFT — Finnish (machine-translated) 是一个芬兰语机器翻译数据集,基于Dolci Instruct SFT混合版本(dolci-instruct-sft-filtered-v1,无数学/无LaTeX子集)翻译而成,专为芬兰语大型语言模型(LLMs)的监督微调(SFT)设计。数据集包含234,745行多轮对话(主要为单轮问答),使用translategemma-27b翻译模型从英文原始数据生成。数据格式为聊天消息(包含角色和内容字段),并经过过滤移除了空内容行。数据模式包括id、messages、source_dataset和domain等列,来源数据集包括Aya、FLAN、WildGuardMix等多个子集。该数据集可能存在机器翻译伪影,并继承上游Dolci/Tulu源的ODC-BY许可证。
Dolci Instruct SFT — Finnish (machine-translated) is a Finnish machine-translated dataset based on the Dolci Instruct SFT mixture (dolci-instruct-sft-filtered-v1, no-math/no-latex subset), produced for supervised fine-tuning (SFT) of Finnish LLMs. It contains 234,745 rows of multi-turn chat (mostly single Q→A), translated from the English source using the translategemma-27b model. The format is chat messages with role and content fields, filtered to remove rows with empty content. The schema includes columns for id, messages, source_dataset, and domain, with source datasets comprising Aya, FLAN, WildGuardMix, and others. The dataset may have translation artifacts and inherits the ODC-BY license from upstream Dolci/Tulu sources.




