[–] kp1197 1y ago ↗ Does performing gradient descent on token input embeddings lead to interpretable results? And if not, why?
1 comment
[ 1.9 ms ] story [ 14.1 ms ] thread