CLaSp: In-Context Layer Skip for Self-Speculative Decoding
This addresses the problem of computational inefficiency in LLM inference for users needing faster deployment, offering a plug-and-play solution that is incremental over existing speculative decoding methods.
The paper tackles the challenge of accelerating large language model decoding without additional training by proposing CLaSp, an in-context layer-skipping strategy for self-speculative decoding, achieving speedups of 1.3x to 1.7x on LLaMA3 models while preserving text distribution.
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 1.3x ~ 1.7x on LLaMA3 series models without altering the original distribution of the generated text.