Overview
Code Llama brought Llama 2 to code in August 2023 and helped kick off the open coding model ecosystem. It ships in three flavors: base, Python (further trained on Python) and Instruct (tuned for chat). The 70B models followed in January 2024. Models were trained on 16K-token sequences and remain stable on inputs up to about 100K tokens.
Strengths
- Mature tooling and broad support across inference engines
- Python-specialized variant
- Fill-in-the-middle on the 7B and 13B models
Run it locally
ollama run codellama:13bGood fit for
Baselines, education and existing pipelines that already depend on Llama 2-era tooling.
Frequently asked questions
Is Code Llama still worth using?
Newer models such as Qwen2.5-Coder outperform it at the same size. Code Llama is still widely supported and useful as a baseline.
Can Code Llama be used commercially?
Yes, under the Llama 2 Community License. Companies with more than 700 million monthly active users need a separate license from Meta.