Breaking AR's Sampling Bottleneck: Provable Acceleration via Diffusion Language Models

Open in new window