An Apache-2.0 dataset curated by Eric Hartford and Cognitive Computations. The purpose of this dataset is to train R1-style reasoning models.
This is a reformatted version of the Flash subset for ease of use. It adds the model's response to the conversation with the following special tokens: <|begin_of_thought|>, <|end_of_thought|>, <|begin_of_solution|>, <|end_of_solution|>.
Please like the original dataset if you enjoy this reformatted version. Thanks to Eric… See the full description on the dataset page:
https://huggingface.co/datasets/mlabonne/dolphin-r1-flash.
