Track the frontier, from condensed matter to cosmology.
Search papers, labs, and topics across Lattice.
2
0
0
Distilling causal LLMs into parallel diffusion models fundamentally handicaps students when the teacher ignores visible future tokens—aligning this context yields up to 4-point accuracy gains at 1.5× faster training.
Asynchronous RL for LLMs can be sped up 2.5x by explicitly controlling policy-gradient variance, without sacrificing synchronous performance.