3 ms·
A dedicated training/inference card is still more general than an LLama 2 inference card. It's obvious that you will get better efficiency the more you tailor y
by KeplerBoy 2y ago
A dedicated training/inference card is still more general than an LLama 2 inference card.
It's obvious that you will get better efficiency the more you tailor your silicon to the task, with diminishing gains but still.