Search papers, labs, and topics across Lattice.
New York University, NVIDIA
1
0
3
10
Latent recurrent models now match non-recurrent baselines at one loop, unlocking compute-for-accuracy scaling without the prior "one-loop tax."
