Hacker News new | ask | show | jobs
by agold97 446 days ago
yeah — sounds about right. retraining the whole model just to add one jargon-y term isn’t super efficient. this approach lets you plug in a vocab list at runtime instead, which feels a lot more scalable.