LMSerg/iola-history-sft
收藏资源简介:
IOLA History SFT 是一个用于监督微调(SFT)的数据集,专门设计用于基于 `google/gemma-3-1b-it` 模型的独立历史LoRA适配器,名为IOLA。该数据集旨在训练和评估模型处理历史相关内容的能力,包括稳定历史事实的自然文本回答、在用户请求来源时返回 `rag_search` JSON 响应,以及对模糊重命名问题进行澄清询问。数据集包含多个版本(v1、v2、v3),每个版本有训练和评估文件,总行数较少(小于1000行),涉及俄语(ru)语言,并适用于文本生成任务。
IOLA History SFT is a supervised fine-tuning (SFT) dataset specifically developed for an independent historical LoRA adapter named IOLA built upon the `google/gemma-3-1b-it` model. The core objectives of this dataset are to train and evaluate the model's capabilities in handling historical-related content, including generating natural textual responses that uphold accurate historical facts, returning a `rag_search` JSON-formatted response when users request source information, and posing clarifying questions for ambiguous renaming-related issues. The dataset includes multiple versions (v1, v2, v3), each with dedicated training and evaluation files. It has a relatively small total size with fewer than 1000 rows, supports the Russian (ru) language, and is tailored for text generation tasks.



