1 paper · 1 filter
Kejing Xia, Mingzhe Li, Lixuan Wei +5
Discrete diffusion language models (dLLMs) generate text by iteratively denoising a masked sequence. However, standard dLLMs condition each denoising step solely on the current har…