Search papers, labs, and topics across Lattice.
Correspoding Author
1
0
3
3
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.