# Hugging Face — Code Llama: Llama 2 learns to code

- Company: Hugging Face (huggingface.co)
- Announced: 2023-08-25
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/codellama
- Record: https://forck.live/items/2005-code-llama-llama-2-learns-to-code
- Subject: Platform
- Models affected: Code Llama
- License: same permissive community license as Llama 2

Meta released Code Llama, a family of open-access code-specialized models based on Llama 2, with sizes 7B, 13B, and 34B parameters, including base, Python specialist, and instruct variants, trained on 500B tokens of code data, with a 16k context window (extrapolatable to 100k), and integrated into the Hugging Face ecosystem.

## Evidence

Verbatim from https://huggingface.co/blog/codellama:

> Code Llama is a family of state-of-the-art, open-access versions of Llama 2 specialized on code tasks

---

Record: https://forck.live/items/2005-code-llama-llama-2-learns-to-code
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md