PleIAs/BSF_Redline
收藏资源简介:
该数据集由Pleias与Bibliothèques Sans Frontières和Dr. Denis Mukwege Foundation合作开发,旨在支持专门的AI助手,用于对抗冲突相关性暴力(CRSV)。数据集核心来源是Red Line Initiative的指南手册,通过语义分块、上下文注入、实体归一化和Markdown结构化等处理,转化为高质量的检索增强生成(RAG)数据源。数据用于训练和优化小语言模型,使其能够在资源受限的环境(如树莓派)中本地运行,提供法律知识辅助,帮助CRSV幸存者及其倡导者。训练过程中使用了来自乌克兰和尼日利亚研讨会的问答数据、法律专家反馈,以及SYNTH管道生成的数据,基于Gemma 3-12B模型进行微调,以提升模型推理和引用能力。
This dataset, developed by Pleias in collaboration with Bibliothèques Sans Frontières and the Dr. Denis Mukwege Foundation, supports specialized AI assistants to combat conflict-related sexual violence (CRSV). The core data source is the Red Line Initiatives Guidebook, which is transformed into a high-quality Retrieval Augmented Generation (RAG) resource through processes such as semantic chunking, context injection, entity normalization, and Markdown structuring. The data is used to train and optimize small language models capable of running locally on hardware like Raspberry Pi, providing legal knowledge assistance to CRSV survivors and their advocates. Training incorporates question-answer data from workshops in Ukraine and Nigeria, expert legal feedback, and data from the SYNTH pipeline, with fine-tuning based on the Gemma 3-12B model to enhance reasoning and citation capabilities.




