LST: Ladder Side-Tuning for Parameter and Memory Efficient Transfer Learning

TL;DR

Ladder Side-Tuning (LST) is a novel parameter-efficient transfer learning method that significantly reduces training memory by using a small side network with shortcut connections, outperforming existing PETL techniques in accuracy and memory savings.

Contribution

LST introduces a separate ladder side network with shortcut connections, eliminating the need for backpropagation through the entire backbone, thus greatly reducing memory usage during fine-tuning.

Findings

LST saves 69% of memory costs compared to full fine-tuning.

LST achieves higher accuracy than Adapter and LoRA in low-memory settings.

LST outperforms other PETL methods on NLP and vision-language tasks.

Abstract

Fine-tuning large pre-trained models on downstream tasks has been adopted in a variety of domains recently. However, it is costly to update the entire parameter set of large pre-trained models. Although recently proposed parameter-efficient transfer learning (PETL) techniques allow updating a small subset of parameters (e.g. only using 2% of parameters) inside a pre-trained backbone network for a new task, they only reduce the training memory requirement by up to 30%. This is because the gradient computation for the trainable parameters still requires backpropagation through the large pre-trained backbone model. To address this, we propose Ladder Side-Tuning (LST), a new PETL technique that can reduce training memory requirements by more substantial amounts. Unlike existing parameter-efficient methods that insert additional parameters inside backbone networks, we train a ladder side network, a small and separate network that takes intermediate activations as input via shortcut connections (called ladders) from backbone networks and makes predictions. LST has significantly lower memory requirements than previous methods, because it does not require backpropagation through the backbone network, but instead only through the side network and ladder connections. We evaluate our method with various models (T5 and CLIP-T5) on both NLP (GLUE) and vision-and-language (VQA, GQA, NLVR2 , MSCOCO) tasks. LST saves 69% of the memory costs to fine-tune the whole network, while other methods only save 26% of that in similar parameter usages (hence, 2.7x more memory savings). Moreover, LST achieves higher accuracy than Adapter and LoRA in a low-memory regime. To further show the advantage of this better memory efficiency, we also apply LST to larger T5 models, attaining better GLUE performance than full fine-tuning and other PETL methods. The accuracy-efficiency trade-off also holds on VL tasks.