Skip to content

Dataset · OpenDataArena

ODA-Math-460k: math for training

OpenDataArena/ODA-Math-460k

460k cleaned English math reasoning examples for training and evaluating reasoning models.

What's inside

The dataset is compiled from top open math corpora ranked by OpenDataArena. It was cleaned of duplicates and benchmark overlaps, filtered using LLMs, and prepared with answers verified by a verifier. It contains English tabular and text examples, numbering from 100k to 1M, with a focus on 'curriculum-complex' tasks.

What you can do with it

  • Fine-tune a mathematical reasoning model (SFT)
  • Evaluate model robustness on tasks cleaned of benchmarks
  • Gather RL data for verifiable answers
Tasks
Question answering, Text generation
Languages
English
Size
100K–1M rows
Downloads this month
539
Likes
106

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("OpenDataArena/ODA-Math-460k")