Diffusion language models

tags

LLM, Diffusion models, Language modeling, Transformers

Language model architecture that use diffusion instead of autoregression. They generate text by iteratively denoising masked or noised tokens, which enables parallel decoding.