cognitivecomputations/dolphin-r1

Dolphin R1是一个由Eric Hartford和Cognitive Computations团队创建的Apache-2.0许可的数据集。这个数据集包含了800k个样本，其组成与用于训练DeepSeek-R1 Distill模型的训练集相似，包括来自DeepSeek-R1的30万个推理样本，来自Gemini 2.0的30万个快速思考推理样本，以及20万个Dolphin聊天样本。该数据集的目的是用于训练R1风格的推理模型。

Dolphin R1 is an Apache-2.0 licensed dataset curated by Eric Hartford and the Cognitive Computations team. This dataset contains 800k samples similar in composition to the training set used for training DeepSeek-R1 Distill models, including 300k reasoning samples from DeepSeek-R1, 300k reasoning samples from Gemini 2.0 flash thinking, and 200k Dolphin chat samples. The purpose of this dataset is to train R1-style reasoning models.