Skip to content

Dataset · open-r1

350k verified DeepSeek-R1 reasoning traces

open-r1/Mixture-of-Thoughts

350k verified reasoning chains in math, code, and science for training step-by-step reasoning.

What's inside

The dataset contains 350k verified reasoning traces distilled from DeepSeek-R1. Topics include math, programming, and sciences. The data is in English and designed to train language models in step-by-step reasoning. It was used in Open R1 to train the OpenR1-Distill-7B SFT model, replicating the capabilities of DeepSeek-R1-Distill-Qwen-7B on the same base model.

What you can do with it

  • SFT fine-tuning of a 7B model for step-by-step reasoning
  • Testing model capabilities in solving math and coding tasks
  • Replicating DeepSeek-R1 distillation in Open R1
Tasks
Text generation
Languages
English
Size
100K–1M rows
Downloads this month
27.9K
Likes
351

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("open-r1/Mixture-of-Thoughts")