4 ms·
Yep. Or even better, compact after a random number of turns. The model must then learn to preserve useful context at arbitrary context lengths.
by wenhan_zhou 4mo ago
Yep. Or even better, compact after a random number of turns. The model must then learn to preserve useful context at arbitrary context lengths.