Skip to content

Dataset · IFM

Math problem-solving and reasoning data

IFM/Math-Reasoning

Large corpus of math problems, rewrites, and dialogues for training reasoning language models.

What's inside

Part of the K2 Horizon series, this repository contains text data for training models on reasoning tasks: math problem solving, text reformulation, and dialogues. The volume ranges from 1 to 10 billion records, divided into several subsets. All parts feature a parquet train split with streaming and viewing support.

What you can do with it

  • Training and SFT of language models for step-by-step math problem solving
  • Developing logical reasoning skills in open LLMs
Tasks
Text generation
Size
1B–10B rows
Downloads this month
117.2K
Likes
39

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("IFM/Math-Reasoning")