@inproceedings{chen-etal-2025-clasp,
    title = "{CL}a{S}p: In-Context Layer Skip for Self-Speculative Decoding",
    author = "Chen, Longze  and
      Shan, Renke  and
      Wang, Huiming  and
      Wang, Lu  and
      Liu, Ziqiang  and
      Luo, Run  and
      Wang, Jiawei  and
      Alinejad-Rokny, Hamid  and
      Yang, Min",
    editor = "Che, Wanxiang  and
      Nabende, Joyce  and
      Shutova, Ekaterina  and
      Pilehvar, Mohammad Taher",
    booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2025",
    address = "Vienna, Austria",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingestion-acl-25/2025.acl-long.1525/",
    pages = "31608--31618",
    ISBN = "979-8-89176-251-0",
    abstract = "Speculative decoding (SD) is a promising method for accelerating the decoding process of Large Language Models (LLMs). The efficiency of SD primarily hinges on the consistency between the draft model and the verify model. However, existing drafting approaches typically require additional modules to be trained, which can be challenging to implement and ensure compatibility across various LLMs. In this paper, we propose CLaSp, an in-context layer-skipping strategy for self-speculative decoding. Unlike prior methods, CLaSp does not require additional drafting modules or extra training. Instead, it employs a plug-and-play mechanism by skipping intermediate layers of the verify model to construct a compressed draft model. Specifically, we develop a dynamic programming algorithm that optimizes the layer-skipping process by leveraging the complete hidden states from the last verification stage as an objective. This enables CLaSp to dynamically adjust its layer-skipping strategy after each verification stage, without relying on pre-optimized sets of skipped layers. Experimental results across diverse downstream tasks demonstrate that CLaSp achieves a speedup of $1.3\times \sim 1.7\times$ on LLaMA3 series models without altering the original distribution of the generated text."
}