Uno: lossless LLM speedups via diffusion-augmented autoregression
MBZUAI
Diffusion-augmented LLMs split parameters into standard next-token weights plus lightweight diffusion heads trained to emit several tokens in parallel, added via a distillation phase with minimal pipeline overhead. The accompanying Psi-Spec samplers give lossless speedup without a draft model, and the 8B Uno beats speculative decoding at every batch size with up to 3x speedups, outperforming the 26B DiffusionGemma and Mercury 2. Code and checkpoints released.
Why it matters
112 upvotes on HF Daily Papers; a credible drop-in alternative to speculative decoding that keeps AR quality while decoding in parallel.
Importance: 4/5
Notable paper with 112 upvotes on HF Daily Papers (bump applied); code and checkpoints released