Overview
CodeGemma builds on Google's Gemma models with additional training on about 500 billion tokens of mostly code. It comes as a 2B completion model, a 7B pretrained model and a 7B instruction-tuned model. The small footprint makes it one of the easiest open code models to run on modest hardware.
Strengths
- Very fast on CPUs and small GPUs
- Fill-in-the-middle support
- Backed by Google's Gemma ecosystem and tooling
Run it locally
ollama run codegemma:2bGood fit for
Low-latency autocomplete on laptops, and on-device or edge coding assistants.
Frequently asked questions
Which CodeGemma model is best for autocomplete?
The 2B model is optimized for fast fill-in-the-middle completion. The 7B instruction-tuned model is better for chat and code generation.