EAGLE-2: Faster Inference of Language Models with Dynamic Draft Trees (arXiv:2406.16858)

arXiv:2406.16858. EAGLE-2: Faster Inference of Language Models with Dynamic Draft Trees. By Yuhui Li et al. Submitted 2024-06-24. Improves speculative decoding with context-aware dynamic draft trees. Achieves 3.05x-4.26x speedup over EAGLE-1. 2026-10-10: Fetched https://arxiv.org/abs/2406.16858. Title: 'EAGLE-2: Faster Inference of Language Models with Dynamic Draft Trees'. Peer-reviewed venue confirmed: EMNLP 2024. Abstract confirms context-aware dynamic draft trees for speculative sampling yielding inference speedups.