Search papers, labs, and topics across Lattice.
1
0
3
Reward modeling inference costs plummet by up to 7x by performing judgments entirely in a compressed latent space, bypassing token-by-token generation.
