Use GitHub Copilot locally on your Macbook with one-click!
demo.mov
This video is not sped up or slowed down.
- First, open VS Code Settings and add the following to your settings.json file:
"github.copilot.advanced": {
"debug.testOverrideProxyUrl": "https://localhost:5001",
"debug.overrideProxyUrl": "https://localhost:5001"
}
- Create a virtualenv to run this Python process, install the requirements, and download the models.
virtualenv venv
source venv/bin/activate
pip install -r requirements.txt
# First setup run. This will download several models to your ~/models folder.
python app.py --setup
- Run it!
python app.py
Enjoy your on-device Copilot!
Is the code as good as GitHub Copilot?
For simple line completions yes. For simple function completions, mostly. For complex functions... maybe.
Is it as fast as GitHub Copilot?
On my Macbook Pro with an Apple M2 Max, the 7b models are roughly as fast. The 34b models are not. Please consider this repo a demonstration of a very inefficient implementation. I'm sure we can make it faster; please do submit a pull request if you'd like to help. For example, I think we need debouncer because sometimes llama.cpp/GGML isn't fast at interrupting itself when a newer request comes in.
Can this be packaged as a simple Mac app?
Yes!, I'm sure it can be, I just haven't had the time. Please do submit a pull request if you're into that sort of thing!
Should there be a meta-model that routes to a 1b for autocomplete, 7b for more complex autocomplete, and a 34b for program completion?
Hmm, that seems like an interesting idea.
OK, but in summary, is it good?
Only if your network is bad. I don't think it's competitive if you have fast Internet. But it sure is awesome on airplanes and while tethering!