Reducing Fine-Tuning Memory Overhead by Approximate and Memory-Sharing Backpropagation

TL;DR

This paper introduces a novel approach combining approximate backpropagation and memory-sharing strategies to significantly reduce memory overhead during fine-tuning of large pretrained models without sacrificing training efficiency.

Contribution

It proposes the Approximate Backpropagation theory and a memory-sharing strategy, enabling memory-efficient training by decoupling forward and backward passes and sharing activation memory.

Findings

Reduces peak memory usage by up to 30% in experiments.

Maintains training efficiency without extra computation.

Applicable to vision and language models.

Abstract

Fine-tuning pretrained large models to downstream tasks is an important problem, which however suffers from huge memory overhead due to large-scale parameters. This work strives to reduce memory overhead in fine-tuning from perspectives of activation function and layer normalization. To this end, we propose the Approximate Backpropagation (Approx-BP) theory, which provides the theoretical feasibility of decoupling the forward and backward passes. We apply our Approx-BP theory to backpropagation training and derive memory-efficient alternatives of GELU and SiLU activation functions, which use derivative functions of ReLUs in the backward pass while keeping their forward pass unchanged. In addition, we introduce a Memory-Sharing Backpropagation strategy, which enables the activation memory to be shared by two adjacent layers, thereby removing activation memory usage redundancy. Our method neither induces extra computation nor reduces training efficiency. We conduct extensive experiments with pretrained vision and language models, and the results demonstrate that our proposal can reduce up to $\sim$$30\%$ of the peak memory usage. Our code is released at https://github.com/yyyyychen/LowMemoryBP.