Adversarial Moment-Matching Distillation of Large Language Models

TL;DR

This paper introduces an imitation learning approach for distilling large language models by matching action-value moments through adversarial training, leading to improved efficiency and state-of-the-art results.

Contribution

It proposes a novel adversarial moment-matching method for knowledge distillation that moves beyond traditional distribution distance minimization.

Findings

Achieves state-of-the-art performance on instruction-following tasks.

Effective in both task-agnostic and task-specific settings.

Outperforms existing knowledge distillation methods.

Abstract

Knowledge distillation (KD) has been shown to be highly effective in guiding a student model with a larger teacher model and achieving practical benefits in improving the computational and memory efficiency for large language models (LLMs). State-of-the-art KD methods for LLMs mostly rely on minimizing explicit distribution distance between teacher and student probability predictions. Instead of optimizing these mandatory behaviour cloning objectives, we explore an imitation learning strategy for KD of LLMs. In particular, we minimize the imitation gap by matching the action-value moments of the teacher's behavior from both on- and off-policy perspectives. To achieve this action-value moment-matching goal, we propose an adversarial training algorithm to jointly estimate the moment-matching distance and optimize the student policy to minimize it. Results from both task-agnostic instruction-following experiments and task-specific experiments demonstrate the effectiveness of our method and achieve new state-of-the-art performance.