OpenR1 Math 220K: Verified Math Problems for Reinforcement Learning

qingy2024 provides a cleaned subset of the OpenR1-Math-220k dataset, containing 61,506 mathematics questions where at least one solution was verified by a math_verify process. The dataset is hosted on Hugging Face and was last updated on February 28, 2025. Its primary intended use is for training large language models via reinforcement learning.

Use Cases

• Training reinforcement learning agents for mathematical problem-solving based on verified solution data.
• Benchmarking language model performance on mathematical reasoning tasks.
• Fine-tuning instruction-following models using a curated set of math questions.
• Studying the effectiveness of verification processes in generating training data.

Strengths

• Contains 61,506 verified mathematics questions, providing a substantial corpus.
• Data is cleaned and simplified from the original source, likely improving usability.
• Each question has at least one solution verified by a math_verify process, suggesting a quality filter.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the original source is unknown, limiting comparison of the subset's scale.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Repository: open-r1/OpenR1-Math-220k

Collection Method

Cleaned and simplified subset of the original dataset, filtered for questions with verified solutions.

Time Range

null

Freshness

Last updated 2025-02-28 18:05:18; freshness should be verified.

Geography

null