OpenR1-Math-Raw

Dataset description

OpenR1-Math-Raw is a large-scale dataset for mathematical reasoning. It consists of 516k math problems sourced from AI-MO/NuminaMath-1.5 with 1 to 8 reasoning traces generated by DeepSeek R1. The traces were verified using Math Verify and LLM-as-Judge based verifier (Llama-3.3-70B-Instruct)

The dataset contains:

• 516,499 problems
• 1,209,403 R1-generated solutions, with 2.3 solutions per problem on average
• re-parsed answers (reparsed_answers) using Llama-3.3-70B-Instruct

With following distribution of correct answers:

You can load the dataset as follows:

from datasets import load_dataset

ds = load_dataset("open-r1/OpenR1-Math-Raw", split="train")

Dataset curation

We only keep the solutions that fit in the 16k-token budget, and follow the <think>...</think> reasoning format. Only the non-synthetic problems from NuminaMath-1.5 were used.

For a more curated sample of this dataset and more details please see open-r1/OpenR1-Math-220k.

Changelog

[1.1]

• Added reparsed_answers column, created by meta-llama/Meta-Llama-3.3-70B-Instruct prompted to extract answer from $solution$ column.
• Added correctness column, which contains verification results computed by math-verify on the reparsed_answers/answer column, as well as llama-based verification using meta-llama/Meta-Llama-3.3-70B-Instruct

License

The dataset is licensed under Apache 2.0