Search papers, labs, and topics across Lattice.
University of Massachusetts Amherst
1
0
3
Discrete diffusion language models can now achieve higher accuracy without retraining the entire backbone, thanks to a lightweight recurrent memory module that bridges denoising steps.
