meta-math/MetaMathQA

Dataset Overview

Dataset Name

• Name: MetaMathQA
• Data Augmentation Source: Enhanced from the training sets of GSM8K and MATH.
• Test Set Usage: No test‑set data are included in the augmentation.

Model Training

• Model Name: MetaMath‑Mistral‑7B
• Base Model: Mistral‑7B
• Training Dataset: MetaMathQA
• Performance Improvement: Using MetaMathQA and upgrading the base model from llama‑2‑7B to Mistral‑7B raises GSM8K performance from 66.5 to 77.7.

Experimental Results

• Model Performance Comparison:
  • MetaMath‑Mistral‑7B: GSM8K Pass@1 = 77.7, MATH Pass@1 = 28.2.
  • Other Models: Includes MPT‑7B, Falcon‑7B, LLaMA‑1‑7B, etc. Detailed metrics are shown in the experiment table.

Citation

@article{yu2023metamath,
  title={MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models},
  author={Yu, Longhui and Jiang, Weisen and Shi, Han and Yu, Jincheng and Liu, Zhengying and Zhang, Yu and Kwok, James T and Li, Zhenguo and Weller, Adrian and Liu, Weiyang},
  journal={arXiv preprint arXiv:2309.12284},
  year={2023}
}