discrete-diffusion-faster-inference

Enables diffusion LLMs to achieve 2.5× faster inference than autoregressive models through block-wise generation with parallel inter-block decoding.

Independent GitHub snapshot · Automated classification · Original documentation governs