farbodtavakkoli/OTel-Safety

OTel-Safety是一个专门用于训练大型语言模型的数据集，旨在在RAG（检索增强生成）流程中，当检索到的上下文不足或不相关时，模型能够选择不回答。该数据集是Open Telco AI（OTel）项目的一部分，该项目是电信领域最大的开源AI计划，由100多名行业和学术领域的专家共同策划。数据集包含多个字段，如anchor、prompt、completion等，用于训练模型识别上下文不足的情况并产生拒绝回答。数据来源包括3GPP规范、GSMA PRDs、O-RAN文档等权威电信资料。数据收集和处理过程包括四个阶段的清理流程，最终得到326K个样本。数据集的用途包括微调大型语言模型以改善电信领域RAG流程中的拒绝行为，但也存在一些局限性，如领域范围、语言、拒绝边界和时间覆盖范围等。

OTel-Safety is a specialized dataset for training large language models to abstain from answering when the retrieved context in a RAG pipeline is insufficient or irrelevant. It is part of the Open Telco AI (OTel) project, the largest open-source AI initiative in telecommunications, curated by over 100 domain experts from industry and academia. The dataset includes fields such as anchor, prompt, completion, etc., to train models to recognize context insufficiency and produce a refusal response. Data sources include authoritative telecommunications materials like 3GPP specifications, GSMA PRDs, and O-RAN documentation. The data collection and processing involve a four-stage cleaning pipeline, resulting in 326K samples. The dataset is intended for fine-tuning large language models to improve abstention behavior in telecom-domain RAG pipelines, but it has limitations such as domain scope, language, abstention boundary, and temporal coverage.