Meta · Code Models

Code Llama

Name: Code Llama
Author: Meta

Meta's code-specialized Llama 2 models from 7B to 70B, with base, Python and instruction-tuned versions and code infilling support.

Official page

Run it locally

With Ollama installed, download and start the default version with:

ollama run codellama:7b

Variant	Download
codellama:7b	3.8GB
codellama:13b	7.4GB
codellama:34b	19GB
codellama:70b	39GB

Memory: About 3.8GB for the default codellama:7b version (4-bit)
Formats: Safetensors (Hugging Face), GGUF (Ollama, llama.cpp, LM Studio)

The download size is roughly the memory (RAM or GPU VRAM) needed to load a variant; allow extra for long contexts. See our hardware guide.

Overview

Code Llama continued training Llama 2 on code-heavy data. It came in base, Python-specialized and instruct variants, and the smaller models support fill-in-the-middle for editor completion.

Highlights

7B, 13B, 34B and 70B sizes
Base, Python and Instruct versions
Fill-in-the-middle code completion
Trained with 16K sequences, stable up to 100K tokens

Typical uses

Local code completion and explanation, research on code models.

Things to consider

Superseded by newer code models such as Qwen2.5-Coder.
Download sizes refer to the default quantized versions in the Ollama library; full-precision weights are larger.

Specifications are taken from the model card on Hugging Face, the Ollama library and the provider’s announcement. Benchmark results depend on evaluation settings; see how AI benchmarks work.

#Code Models #Open Weights #Local AI #Meta