arXiv NLP
techCenter
TreeGraft: Adaptive Multi-Drafter Grafting for Tree-Based Speculative Decodingtranslating…
1 min readUnknownarXiv Press Group
arXiv:2608.26112v1 Announce Type: new
Abstract: Speculative decoding accelerates large language model inference through a draft-then-verify paradigm. Building on this, tree-structured methods improve inference by organizing proposals into multiple candidate paths, increasing…