Skeleton-of-Thought: Prompting LLMs for Efficient Parallel Generation
This work addresses inference efficiency for LLM users, but it is an initial attempt and incremental in optimizing existing methods.
The paper tackles the high generation latency of large language models (LLMs) caused by sequential decoding, proposing Skeleton-of-Thought (SoT) to first generate answer skeletons and then complete them in parallel, achieving considerable speed-ups across 12 LLMs and potentially improving answer quality in some categories.
This work aims at decreasing the end-to-end generation latency of large language models (LLMs). One of the major causes of the high generation latency is the sequential decoding approach adopted by almost all state-of-the-art LLMs. In this work, motivated by the thinking and writing process of humans, we propose Skeleton-of-Thought (SoT), which first guides LLMs to generate the skeleton of the answer, and then conducts parallel API calls or batched decoding to complete the contents of each skeleton point in parallel. Not only does SoT provide considerable speed-ups across 12 LLMs, but it can also potentially improve the answer quality on several question categories. SoT is an initial attempt at data-centric optimization for inference efficiency, and showcases the potential of eliciting high-quality answers by explicitly planning the answer structure in language.