Search papers, labs, and topics across Lattice.
1
0
3
Diffusion language models can achieve up to 3.4x faster inference without sacrificing quality by committing stable token prefixes instead of scattering accepted tokens.
