teapotai/teapot-chat

茶壶聊天数据集是一个高质量的数据集，包含约1000个真实世界的用户聊天互动，旨在帮助训练和评估基于上下文、抗幻觉的语言模型。每个对话都是从我们的网站上的用户与TeapotLLM聊天中获取的，TeapotLLM是一个开源的语言模型，经过微调，严格使用提供的上下文来回答。所有用户查询都与从Brave Search API检索的相关上下文配对，并由TeapotLLM回答，TeapotLLM经过训练，依靠提供的上下文来避免幻觉。这个数据集中的每个条目都经过人工审查和 curated，以确保相关性、清晰性和事实一致性。这个数据集是一个持续努力的一部分，旨在通过提供透明、高质量的资源来回馈社区，这些资源有助于构建更安全、更可靠的AI系统，特别是对于那些在本地优先或资源受限部署上工作的开发人员。

Teapot Chat is a high-quality dataset of ~1k real-world user chat interactions designed to help train and evaluate context-grounded, hallucination-resistant language models. Each conversation was sourced from our website where users chat with TeapotLLM — an open-source language model fine-tuned to answer strictly using provided context. All user queries were paired with relevant context retrieved from the Brave Search API and answered by TeapotLLM, which was trained to avoid hallucinations by relying exclusively on the given context. Every entry in this dataset has been manually reviewed and curated by human reviewers to ensure relevance, clarity, and factual alignment. This dataset is part of an ongoing effort to give back to the community by providing transparent, high-quality resources for building safer, more reliable AI systems — especially for developers working on local-first or resource-constrained deployments.