Inference & Servingadvanced12 min · 9 steps
Diffusion Language Models
Instead of writing left to right, start from a fully masked sequence and denoise every position in parallel — fast, good at code and infill, still behind on hard reasoning.
forward passes8
tokens / pass1
token
metrics are illustrativeProblemStep 1 of 9
Autoregressive: one token per forward pass
Almost every LLM you use is autoregressive (AR): it predicts each token from the past only, left to right, one forward pass per token.
Eight tokens means eight sequential passes, no matter how predictable the text is.
Lesson 8 of 9 in Inference & Serving