Skip to content

Dataset · gretelai

Synthetic school math problems with step-by-step reasoning

gretelai/gretel-math-gsm8k-v1

GSM8K-style math problems generated by Llama-3.1-405B with detailed thoughts and answers.

What's inside

The dataset contains English text-based problems at the elementary and middle school level requiring multi-step reasoning, complete with step-by-step thoughts and solutions. The dataset size ranges from 10,000 to 100,000 examples. The data is fully synthesized based on the GSM8K benchmark using the Gretel Navigator tool and the Meta-Llama-3.1-405B model.

What you can do with it

  • LLM SFT fine-tuning to improve logical and mathematical skills
  • Training models to build reasoning chains
  • Evaluating language models' ability to solve multi-step problems
Tasks
Question answering
Languages
English
Size
10K–100K rows
Downloads this month
2.1K
Likes
39

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("gretelai/gretel-math-gsm8k-v1")