arXivZixian Li, Tong Li, Chi Xie, Xiaohui Song, Haonan LuMon, Aug 3, 2026, 5:15 AM PDT
score 16.8
AI researchers speed up language models by drafting several possible next words at once
Original: From Chains to Trees: Parent-Conditioned Drafting for Semi-Autoregressive Speculative Decoding
Source: arxiv.org ↗
Writing ELI5 summary…