Search papers, labs, and topics across Lattice.
3
0
5
Reasoning in base LLMs can be unlocked by simply prepending specific tokens, matching RL-tuned performance on math and coding.
Recovering lost visual details in long-horizon video generation is now possible by compositing non-contiguous spatio-temporal memory, even with a frozen generator.
A novel loss function that boosts convergence speed by up to 40% while improving image quality by addressing the spectral imbalance in pixel-space training.
