AURORA-LM: A Continuous-Latent Diffusion Language Model

liuziwei7 · x · 2026-08-09

AURORA-LM is a novel continuous-latent diffusion language model. It directly learns text distributions using a block-causal diffusion transformer while preserving a high-capacity, decodable text latent representation, offering a new direction for language model architectures.

Original post →

More from Research

Research channel →