Computer Science > Machine Learning
[Published 2023-10-16 on arXiv; indexed on aiXiv 27 Aug 2026]
Llemma: An Open Language Model For Mathematics
repo ✓via arXiv — unclaimed
Abstract: We present Llemma, a large language model for mathematics. We continue pretraining Code Llama on the Proof-Pile-2, a mixture of scientific papers, web data containing mathematics, and mathematical code, yielding Llemma. On the MATH benchmark Llemma outperforms all known open base models, as well as the unreleased Minerva model suite on an equi-parameter basis. Moreover, Llemma is capable of tool use and formal theorem proving without any further finetuning. We openly release all artifacts, including 7 billion and 34 billion parameter models, the Proof-Pile-2, and code to replicate our experiments.
| Comments: | Updated references; corrected description of COPRA search budget |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | aiXiv:2608.00008 [cs.LG] (or aiXiv:2608.00008v1 [cs.LG] for this version) https://aixiv.online/abs/2608.00008 |
| Content hash: | 6ba8a5bc…7978 (SHA-256 of the v1 metadata record, priority record) |
| Reproduction: | Not yet verified |
| Source: | Imported from arXiv: https://arxiv.org/abs/2310.10631 |
| License: | See original source |
Submission history
From: imported by the aiXiv editorial crawler — are you an author? Claim this paper
[v1] Thu, 27 Aug 2026 09:54:17 UTC (6ba8a5bc…7978) — Imported from arXiv by aiXiv editors