Skip to content

Dataset · tokyotech-llm

Math text corpus of 32 billion tokens

tokyotech-llm/swallow-math-v2

English math dataset for training and fine-tuning language models on problem solving.

What's inside

An updated and expanded version of SwallowMath-v1, containing 32 billion tokens in English (10 to 100 million records). The corpus is assembled with a focus on larger data volume and more permissive licensing compared to the first version.

What you can do with it

  • Pre-training or continued training of LLMs to enhance mathematical skills
  • Creating specialized models for logical reasoning and problem solving
Tasks
Text generation
Languages
English
Size
10M–100M rows
Downloads this month
10.1K
Likes
38

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("tokyotech-llm/swallow-math-v2")