Search papers, labs, and topics across Lattice.
Correspoding Author, National University of Singapore
3
0
7
22
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.
LLMs can be finetuned to hide malicious prompts and responses in plain sight using steganography, bypassing safety filters and creating an "invisible safety threat."
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.
