一作ACL 2026Poster· Natural Language Processing
L ong S pec: Long-Context Lossless Speculative Decoding with Efficient Drafting and Verification
Penghui Yang · Cunxiao Du · Fengzhuo Zhang · Haonan Wang · Tianyu Pang · Chao Du · Bo An
摘要
As Large Language Models (LLMs) can now process extremely long contexts, efficient inference over these extended inputs has become increasingly important, especially for emerging applications like LLM agents that highly depend on this capability. Speculative decoding (SD) offers a promising lossl...