Post

i kind of wonder how possible it is to fine-tune an either transcription model or very small OLM that does cleanup only, like S1-mini, and to make that fine-tuned on all of your typing styles for various things (like emailing coworkers and messaging friends as two separate models)

that specific model I linked is only a fine-tune of.. I think it's 0.6B qwen? so it'd probably be pretty possible to do that on device, even with the CPU, pretty fast. like, I'm even guessing you could fine-tune on a phone, and that'd be perfectly fine

oh fuck, you could totally do some kind of sync thing type thing where you move the model between devices oh fuck, you could even do parametric memory systems too... I don't know if that's really applicable to a transcription model, but it'd be funny if you could do that