KOALA: Enhancing Speculative Decoding for LLM via Multi-Layer Draft Heads with Adversarial Learning

TL;DR

KOALA introduces a multi-layer adversarial training approach to improve speculative decoding in LLMs, significantly enhancing accuracy and reducing inference latency compared to traditional draft heads.

Contribution

It proposes a novel multi-layer draft head architecture with adversarial learning, substantially improving speculative decoding performance in large language models.

Findings

Latency speedup ratio improved by 0.24x-0.41x

Draft head accuracy significantly increased

Outperforms original draft heads by 10.57%-14.09% in speed

Abstract

Large Language Models (LLMs) exhibit high inference latency due to their autoregressive decoding nature. While the draft head in speculative decoding mitigates this issue, its full potential remains unexplored. In this paper, we introduce KOALA (K-layer Optimized Adversarial Learning Architecture), an orthogonal approach to the draft head. By transforming the conventional single-layer draft head into a multi-layer architecture and incorporating adversarial learning into the traditional supervised training, KOALA significantly improves the accuracy of the draft head in predicting subsequent tokens, thus more closely mirroring the functionality of LLMs. Although this improvement comes at the cost of slightly increased drafting overhead, KOALA substantially unlocks the draft head's potential, greatly enhancing speculative decoding. We conducted comprehensive evaluations of KOALA, including both autoregressive and non-autoregressive draft heads across various tasks, demonstrating a latency speedup ratio improvement of 0.24x-0.41x, which is 10.57%-14.09% faster than the original draft heads.