Reproduce “Factorization-Guided Training and Inference for Diffusion-Drafted Speculative Decoding” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiakun Fan, Min Xiao, Qiang Wang, Neiwen Ling, Zhichao Liu, Zhihao Jia, Dimitrios S. Nikolopoulos, Zijun Wei
DScout improves diffusion-drafted speculative decoding by aligning supervision with prefix acceptance and configuring inference for throughput. A drafter proposes several tokens for verification, but only the prefix before the first rejection is accepted. On Qwen3-8B, DScout reports 4.14× mean speedup across six tasks at temperature zero and concurrency one, versus its matched autoregressive baseline in evaluation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.