GenDistiller: Distilling Pre-trained Language Models based on an Autoregressive Generative Model

TL;DR

GenDistiller is a novel autoregressive knowledge distillation framework that significantly reduces the size of pre-trained speech models like WavLM while maintaining or improving performance on downstream tasks.

Contribution

It introduces an autoregressive layer-by-layer distillation method that generates hidden representations, enabling substantial model compression with minimal performance loss.

Findings

Achieves 82% reduction in WavLM size.

Outperforms baseline distillation methods on SUPERB tasks.

Uses 33% fewer parameters with similar time consumption.

Abstract

Pre-trained speech language models such as HuBERT and WavLM leverage unlabeled speech data for self-supervised learning and offer powerful representations for numerous downstream tasks. Despite the success of these models, their high requirements for memory and computing resource hinder their application on resource restricted devices. Therefore, this paper introduces GenDistiller, a novel knowledge distillation framework which generates the hidden representations of the pre-trained teacher model directly by a much smaller student network. The proposed method takes the previous hidden layer as history and implements a layer-by-layer prediction of the teacher model autoregressively. Experiments on SUPERB reveal the advantage of GenDistiller over the baseline distilling method without an autoregressive framework, with 33% fewer parameters, similar time consumption and better performance on most of the SUPERB tasks. Ultimately, the proposed GenDistiller reduces the size of WavLM by 82%.