What question did this study set out to answer?

To enhance the complex mathematical reasoning capabilities of open-source large language models by creating a specialized dataset through question bootstrapping.

September 21, 2023Open Access

MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models

View Full Paper

Ask AI

Bookmark

To enhance the complex mathematical reasoning capabilities of open-source large language models by creating a specialized dataset through question bootstrapping.
Bootstrapped mathematical problems by rewriting questions from multiple perspectives without external knowledge, generating the MetaMathQA dataset.
Fine-tuned LLaMA-2 models of multiple parameter scales on MetaMathQA.
Evaluated reasoning accuracy across the GSM8K and MATH mathematical reasoning benchmark suites.
MetaMath-7B achieved 66.4% on GSM8K and 19.4% on MATH, exceeding same-size state-of-the-art models by 11.5% and 8.7%, respectively.
MetaMath-70B achieved an accuracy of 82.3% on GSM8K, outperforming GPT-3.5-Turbo.

Discussion

No takes yet. Share an insight, caveat, or question.

Implication

Experimental study demonstrates enhanced mathematical reasoning in fine-tuned open-source models, highlighting the power of question bootstrapping without external data.

Cite This Study

Yu et al. (2023) studied this question.

synapsesocial.com/papers/6a10a83f8090e499da61830d https://doi.org/10.48550/arxiv.2309.12284

View Full Paper

Ask AI

Bookmark

Implication

Experimental study demonstrates enhanced mathematical reasoning in fine-tuned open-source models, highlighting the power of question bootstrapping without external data.

MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models

Key Points

Discussion

Implication

Cite This Study

MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models

Key Points

Discussion

Implication

Cite This Study

Also Consider

Also Consider