Uno: Diffusion-Augmented LLM Uses LoRA Drafts and Ψ-Spec Sampler for Lossless Speedups
burkov · x · 2026-09-08
This paper targets a core LLM inefficiency: next-token prediction yields high-quality output but forces slow sequential generation, wasting accelerator capacity and inflating latency and cost — especially for long traces and concurrent agentic workloads.
Method — Uno (diffusion-augmented LLM): keep the full autoregressive weights trained with standard next-token prediction, then add lightweight LoRA adapters trained in a short Diffusion Distillation phase to propose blocks of tokens in parallel.
Inference: a new Ψ-Spec sampler draws parallel drafts from the diffusion pathway and verifies them against the frozen autoregressive distribution, aiming for lossless speedups — multiple tokens per step while preserving the exact output distribution of the strong base model.
Related event: Uno Uses Discrete Diffusion for Lossless LLM Speedups(3 posts)→
More from Research
- From zero open math problems to a Millennium Prize solve in one semester — FlorianGallwitz · 2026-09-09
- MIT's CrysVCD puts chemistry rules before generation to boost stable AI-designed materials — jiqizhixin · 2026-09-09
- OpenAI claims a solution to the 90-year-old Navier-Stokes Millennium Prize Problem — Swarooprm7 · 2026-09-09
- Bubeck reveals Twitter rumors spurred OpenAI's Navier-Stokes push, proofs differ — burny_tech · 2026-09-09
- OpenAI claims its agents produced a proof of the 90-year-old Navier-Stokes Millennium Prize Problem — sjgadler · 2026-09-09
- SIL-Wheel open-source platform for large-scale video data debuts at ECCV — cindy_x_wu · 2026-09-09