Announcement
PoolDINO is claimed to use 4x–16x fewer tokens than RAE, with similar generation quality
A post claims cheaper training and sampling, showing 8x compression against RAE on an M1 Pro CPU without an accelerator.
TLDR
A post presents PoolDINO as using 4x–16x fewer tokens than RAE while delivering similar generation quality, and claims cheaper training and sampling. It also shares a comparison of RAE with 8x compression on an M1 Pro CPU without an accelerator.
Combined views
12K
3 Sources, first seen ago
