Diffusion LMs now match autoregressive quality
A new diffusion language model allocates more computation to harder tokens, closing the quality gap with standard autoregressive models.
A new diffusion language model allocates more computation to harder tokens, closing the quality gap with standard autoregressive models.