Skeleton-of-Thought

Let us accelerate the end-to-end generation of LLMs by 2x without any change to the model, system, or hardware! [ArXiv] [BibTex] [Code] 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 further underscores the potential of pushing LLMs to think more like a human for answe

Skeleton-of-Thought Search this site Embedded Files Skip to main content Skip to navigation Skeleton-of-Thought Xuefei Ning * , Zinan Lin * , Zixuan Zhou * , Zifu Wang , Huazhong Yang , Yu Wang Let us accelerate the end-to-end generation of LLMs by 2x without any change to the model, system, or hardware! [ ArXiv ] [ BibTex ] [ Code ] [ Open Review ] [ Poster ] Abstract 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 LL

related reading

• Optimizing inference · Hugging Facehuggingface.co
• LLM Inference Performance Engineering: Best Practices | Databricks Blogdatabricks.com
• Patterns for Building LLM-based Systems & Productseugeneyan.com
• How to make LLMs go fastvgel.me
• GenAI Handbookgenai-handbook.github.io
• Speculative Decoding - philkravphilkrav.com
• Assisted Generation: a new direction toward low-latency text generationhuggingface.co
• 2025: The year in LLMssimonwillison.net
• Best practices to accelerate inference for large-scale production workloadstogether.ai
• Getting Caught Up to Modern LLM Research | Samarth Goeldev.samarthgoel.com
• Unlocking the Working Memory of Large Language Models for Latent Reasoningarxiv.org
• The State of LLM Reasoning Model Inferencesebastianraschka.com