Kangaroo: Lossless Self-Speculative Decoding via Double Early Exiting

TL;DR

Kangaroo introduces a self-speculative decoding framework that uses a shallow sub-network with early exiting to accelerate large language model inference efficiently, reducing parameters and maintaining accuracy.

Contribution

The paper proposes a novel self-speculative decoding method with a fixed shallow sub-network and early exiting, improving speed and efficiency over traditional draft models.

Findings

Achieves up to 1.68x speedup on Spec-Bench

Uses 88.7% fewer parameters than Medusa-1

Effective in maintaining sampling distribution fidelity

Abstract

Speculative decoding has demonstrated its effectiveness in accelerating the inference of large language models while maintaining a consistent sampling distribution. However, the conventional approach of training a separate draft model to achieve a satisfactory token acceptance rate can be costly. Drawing inspiration from early exiting, we propose a novel self-speculative decoding framework \emph{Kangaroo}, which uses a fixed shallow sub-network as a self-draft model, with the remaining layers serving as the larger target model. We train a lightweight and efficient adapter module on top of the sub-network to bridge the gap between the sub-network and the full model's representation ability. It is noteworthy that the inference latency of the self-draft model may no longer be negligible compared to the large model, necessitating strategies to increase the token acceptance rate while minimizing the drafting steps of the small model. To address this challenge, we introduce an additional early exiting mechanism for generating draft tokens. Specifically, we halt the small model's subsequent prediction during the drafting phase once the confidence level for the current token falls below a certain threshold. Extensive experiments on the Spec-Bench demonstrate the effectiveness of Kangaroo. Under single-sequence verification, Kangaroo achieves speedups up to $1.68\times$ on Spec-Bench, outperforming Medusa-1 with 88.7\% fewer additional parameters (67M compared to 591M). The code for Kangaroo is available at https://github.com/Equationliu/Kangaroo.