So what about self-hosted models?

[pelag0s]

The documentation mentions qwen2.5-coder:32b and from my personal experience that is currently one of the best coding models, also for coding assistant tasks.

Did anyone try it out with PR-Agent? Or did you find an even better model which can be self-hosted locally?

And if yes, what context size is needed for the review functionality to work correctly?

(I also found a page which says for self-hosted models, fine tuning is needed. But I also saw that most models tested there were quite old. Is this still needed?)