Get Started
Home
Topics
Search
Library
Research questionHow can we compare autoregressive and masked-diffusion language models without conflating formulation with architecture?Autoregressive models are commonly decoder-only, whereas masked-diffusion models are often encoder-only. Consequently, differences in speed or perplexity can reflect architectural variation rather than the underlying modeling formulation.
AI
Diffusion Models
Evaluation & Benchmarks
Inference Optimization
Machine Learning
Natural Language Processing
Research Paper
Technology
Latest papersRecent research connected to this question, newest first.Any-Order GPT as Masked Diffusion Model: Decoupling Formulation and ArchitectureThe evidence concerns decoder-only language-model implementations of standard autoregressive and masked-diffusion formulations, with comparisons centered on perplexity and inference speed. It includes masked-diffusion inference using temperature annealing, but does not establish broader deployment or quality effects beyond the reported experiments.research paper · Sep 1, 2026
Related questions
How can masked diffusion language models retain generation quality with only a few denoising steps?How can sparse autoencoder features be shared across language models without per-model retraining?How can continuous diffusion language models reduce denoising steps without sacrificing text-generation quality?How can we tell whether a language diffusion model is memorizing or generalizing?