MEFT: Memory-Efficient Fine-Tuning through Sparse Adapter

TL;DR

MEFT introduces a memory-efficient fine-tuning method for large language models that leverages activation sparsity and CPU memory to enable larger adapters without requiring extensive GPU resources.

Contribution

The paper proposes a novel approach to fine-tune LLMs with larger adapters by exploiting activation sparsity and CPU memory, improving performance under limited GPU resources.

Findings

Achieves comparable fine-tuning results with limited GPU memory

Utilizes CPU memory and activation sparsity for efficient adapter training

Reduces communication overhead with a Mixture of Experts architecture

Abstract

Parameter-Efficient Fine-tuning (PEFT) facilitates the fine-tuning of Large Language Models (LLMs) under limited resources. However, the fine-tuning performance with PEFT on complex, knowledge-intensive tasks is limited due to the constrained model capacity, which originates from the limited number of additional trainable parameters. To overcome this limitation, we introduce a novel mechanism that fine-tunes LLMs with adapters of larger size yet memory-efficient. This is achieved by leveraging the inherent activation sparsity in the Feed-Forward Networks (FFNs) of LLMs and utilizing the larger capacity of Central Processing Unit (CPU) memory compared to Graphics Processing Unit (GPU). We store and update the parameters of larger adapters on the CPU. Moreover, we employ a Mixture of Experts (MoE)-like architecture to mitigate unnecessary CPU computations and reduce the communication volume between the GPU and CPU. This is particularly beneficial over the limited bandwidth of PCI Express (PCIe). Our method can achieve fine-tuning results comparable to those obtained with larger memory capacities, even when operating under more limited resources such as a 24GB memory single GPU setup, with acceptable loss in training efficiency. Our codes are available at https://github.com/CURRENTF/MEFT.