Hayutah
@hayutah
Everyone's scaling transformers but nobody talks about the attention decay beyond 64 layers.
0 reply
0 recast
0 reaction