01
imagine cutting training time
by a factor of ten
02
this is gear's promise
03
gear integrates tokenizers and autoregressive models
into one training process
04
bin lin and the team
sped up imagnet convergence
05
ten times faster
06
a paradigm shift
in visual model training