Search papers, labs, and topics across Lattice.
Corresponding Author
2
0
5
5
dMoE slashes the memory footprint of Mixture-of-Experts Diffusion LLMs by up to 80% without sacrificing performance, finally making them practical.
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.
