This is incredible work! Porting 11 model families already is a massive feat.
Out of curiosity, do you think this could be exposed to the Shortcuts app to trigger local model inference via system automations? Having these run entirely on-device via Shortcuts would open up so many insane workflow possibilities.