danlou/based-chat-v0.1-Mistral-Nemo-Base-2407

该数据集是作为探索理解监督数据集对于将基础大语言模型（LLMs）微调为对话模型的必要性的一部分而开发的。它是一个基于Mistral-Nemo-Base-2407模型生成的合成数据集，用于微调该模型，产生了relay-v0.1-Mistral-Nemo-2407。数据集的生成使用了facebook/empathetic_dialogues和AlekseyKorshuk/quora-question-pairs两个数据集作为对话启动器，并遵循了CC-BY-NC 4.0许可证。数据集的条目被格式化为类似IRC的提示，并随机采样，但有一些限制，如最小词频。

Based-Chat v0.1 (Mistral Nemo Base 2407) is a synthetic dataset developed to explore the necessity of supervised datasets for fine-tuning base LLMs into conversational models. It is created using the Mistral-Nemo-Base-2407 model and used to fine-tune that model, producing the relay-v0.1-Mistral-Nemo-2407 model. The dataset generation method includes using the facebook/empathetic_dialogues and AlekseyKorshuk/quora-question-pairs datasets as conversation starters and formatting them into IRC-style prompts. The entries in the dataset are randomly sampled, with some restrictions such as minimum word frequency. The dataset is licensed under CC-BY-NC 4.0.