Llemma 7B

About

Llemma 7B is an innovative open-source large language model tailored for mathematical tasks, featuring 7 billion parameters. It builds upon Code Llama 7B and has been enhanced with the Proof-Pile-2 dataset, comprising 200 billion tokens of scientific papers and mathematical content. Renowned for its advanced chain-of-thought reasoning, Llemma 7B significantly surpasses other models like Llama-2 and Code Llama. It excels in tool use, such as Python interpreters and theorem proving, without additional fine-tuning, and is openly accessible, driving further research. The model performs exceptionally in mathematical benchmarks like MATH and GSM8k, providing a robust base for future advancements.

Capabilities