Training a smaller, cheaper model to mimic the output of a larger one. A way to ship AI on devices without losing too much quality.
"The 1B distillation of our 70B model runs fine on a laptop."
No comments yet — say something.
Add your own interpretation of "distillation".