Search papers, labs, and topics across Lattice.
2
0
4
9
DMax unlocks faster diffusion language model decoding by reframing the process as iterative self-correction in embedding space, achieving up to 2x speedup without sacrificing accuracy.
dVoting unlocks significant reasoning gains for diffusion LMs at test time by iteratively refining only the most uncertain tokens, sidestepping the computational bottleneck of full re-sampling.
