trollek/ThoughtfulAssistant-v01
收藏Hugging Face2024-07-18 更新2024-07-22 收录
下载链接：
https://hf-mirror.com/datasets/trollek/ThoughtfulAssistant-v01
下载链接
链接失效反馈官方服务：
资源简介：
该数据集的创建目的是教导语言模型在回答问题之前进行思考。通过特定的提示模板生成响应背后的推理过程，并将这些推理过程格式化。数据集的内容来源于多个已存在的数据集，并经过过滤和整合。生成推理过程所使用的模型包括Mistral-7B-Instruct-v0.3和Nous-Hermes-2-Mixtral-8x7B-DPO。数据集的格式为JSON，包含用户指令和模型响应的结构，其中模型响应包括推理过程和原始响应。
This dataset is designed to train language models to think before answering questions. By using a specific prompt template to generate responses that include the thought process, the dataset format includes human instructions and GPT responses, where the GPT responses contain thought start and end markers, as well as the original response. The construction of the dataset involves filtering multi-turn conversations from multiple source datasets and using specific models to generate thought content.
提供机构：
trollek
