Fudan's open-source D3D diffusion drafting speeds up LLM decoding up to 7.48x

Fudan University released D3D, an open-source decoupled diffusion drafting model for asynchronous speculative decoding that removes dependence on target-model hidden states, achieving up to 7.48x decoding speedup on Qwen3.

2026-11-06 ~ 2026-11-07 · 2 related posts

• Fudan's D3D decouples diffusion drafting for asynchronous speculative decoding speedups — Fudan-University · 2026-11-06
• Fudan's D3D diffusion drafter enables async speculative decoding, up to 7.48x speedup on Qwen3 — pmttyji · 2026-11-07