Training and evaluating an assistant that runs offline
Fine-tuning local
Alban Pasquelin built a preparation and evaluation harness to specialise a multilingual Qwen model driving AI Desktop Studio, run locally on Apple Silicon inside Tart virtual machines.
A general model cannot drive a specific application. This one is specialised to understand the 310 actions of AI Desktop Studio, across several languages.
Everything happens on the machine: data preparation, training, then an automated evaluation measuring whether the new version genuinely beats the old one. Without that measurement, fine-tuning is superstition.