4 ms·

Doesn't it entirely depend on how specialized the training data for a given fine tuned model might be?