Code Llama
Llama 2 specialised for code, in 7B/13B/34B (70B added 2024-01-29), with Python and Instruct variants and up to 100K context.
Continued pretraining of Llama 2 on code with fill-in-the-middle for the small sizes; the 34B reached 53.7% HumanEval, the best open result at release, same license as Llama 2.
- Date
- Thursday, 24 August 2023
- Lab
- Meta
- Kind
- open-weights
- Access
- open weights (restricted license)
Figures
| Measure | Value | Measured by |
|---|---|---|
| HumanEval (34B) | 53.7% MBPP 56.2% | company |
Sources
This record was checked against its sources on 6 October 2026. How we check