Diffusion-augmented LLM Uno: 8B beats 26B DiffusionGemma with 3x lossless speedup
iScienceLuvr · x · 2026-09-04
A new paper (UIUC, Cornell, Cerebras; incl. Eric Xing) introduces diffusion-augmented LLMs: two weight sets — standard NTP-trained AR weights for quality, plus lightweight diffusion weights to draft multiple tokens in parallel via a Ψ-Spec sampler, giving lossless speedups.
Key results:
- The 8B Uno beats DFlash and Eagle3 speculative decoding on throughput at every batch size, without a separate draft model, with up to 3x speedup over the base AR model.
- 8B Uno outperforms the 26B open d-LLM DiffusionGemma and proprietary Mercury 2 across agentic tool use, coding, and long-context reasoning.
Paper, code, and HuggingFace weights are public.
More from Infra
- Developer Moves AI Agents to a Dedicated Linux Workstation Using Incus System Containers — kevinkern · 2026-10-21
- Nvidia guides 70% growth citing unmet AI demand, while media panics over a $500M Korean neocloud IPO — firstadopter · 2026-10-21
- Pichai: everyone is massively short on AI compute — 'everyone wishes you had two times what you had planned' — firstadopter · 2026-10-21
- Stanford CS336 Lecture 6: fused Triton kernels cut softmax memory traffic ~4x — stanfordnlp · 2026-10-21
- General Compute partners with Positron AI: Atlas memory-centric hardware ships Q1 2027 — aftahi_ai · 2026-10-21
- AI bond issuance hits $419B in 2026, up 90% from all of 2025 — SumitGup · 2026-10-21