Meet LLEMMA, a math-focused open-source AI outperforming rivals like Google's Minerva.

LLEMMA's models have 7 billion and 34 billion parameters, fine-tuned on Proof-Pile-2 dataset

LLEMMA adapts to tasks via fine-tuning and exhibits superior math problem-solving abilities

LLEMMA can use tools and prove formal theorems without additional fine-tuning

LLEMMA surpasses Google's Minerva on an "equi-parameter basis

Researchers release LLEMMA's assets, enabling further research and improvement

LLEMMA is the first open-source model matching closed-source model performance

LLEMMA contributes to domain-specific language models and enhanced reasoning capabilities

LLEMMA's achievements benefit various fields by specializing language models

LLEMMA inspires new research topics, including reward modeling and algorithmic reasoning

Open-Source AI LLEMMA Outperforms Google's Minerva