1 paper · 1 filter
Jingyu Liu, Xin Dong, Zhifan Ye +6
Diffusion language models hold the promise of fast parallel generation, while autoregressive (AR) models typically excel in quality due to their causal structure aligning naturally…