Diffusion-Based Language Model

A language model that generates text by iteratively predicting and refining masked (hidden) tokens across the entire output, rather than predicting one token at a time from left to right.