Replace high-dimensional weights with a compact trainable latent vector — 200×–500× fewer trainable params ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible resultsSome results have been hidden because they may be inaccessible to you
Show inaccessible results