Break the Sequential Dependency of LLM Inference Using Lookahead Decoding (arXiv:2402.02057)

arXiv:2402.02057. Break the Sequential Dependency of LLM Inference Using Lookahead Decoding. By Yichao Fu et al. (UC Berkeley). Submitted 2024-02-03. Parallel decoding algorithm that accelerates LLM inference without auxiliary models. Up to 1.8x speedup on MT-bench. 2026-10-10: Fetched https://arxiv.org/abs/2402.02057. Title: 'Break the Sequential Dependency of LLM Inference Using Lookahead Decoding'. Peer-reviewed venue confirmed: ICML 2024. Abstract confirms parallel decoding algorithm breaking sequential dependency without auxiliary draft models.