SEEKR: Selective Attention-Guided Knowledge Retention for Continual Learning of Large Language Models

TL;DR

SEEKR introduces a selective attention-guided approach for continual learning in large language models, significantly reducing data replay needs while maintaining or improving performance by focusing on important attention heads.

Contribution

The paper proposes SEEKR, a novel method that uses attention head selection for efficient knowledge retention in continual learning of LLMs, reducing data requirements.

Findings

SEEKR outperforms existing methods in benchmarks.

Achieves comparable or better results with only 10% of replay data.

Reduces replayed data to 1%, improving efficiency.

Abstract

Continual learning (CL) is crucial for language models to dynamically adapt to the evolving real-world demands. To mitigate the catastrophic forgetting problem in CL, data replay has been proven a simple and effective strategy, and the subsequent data-replay-based distillation can further enhance the performance. However, existing methods fail to fully exploit the knowledge embedded in models from previous tasks, resulting in the need for a relatively large number of replay samples to achieve good results. In this work, we first explore and emphasize the importance of attention weights in knowledge retention, and then propose a SElective attEntion-guided Knowledge Retention method (SEEKR) for data-efficient replay-based continual learning of large language models (LLMs). Specifically, SEEKR performs attention distillation on the selected attention heads for finer-grained knowledge retention, where the proposed forgettability-based and task-sensitivity-based measures are used to identify the most valuable attention heads. Experimental results on two continual learning benchmarks for LLMs demonstrate the superiority of SEEKR over the existing methods on both performance and efficiency. Explicitly, SEEKR achieves comparable or even better performance with only 1/10 of the replayed data used by other methods, and reduces the proportion of replayed data to 1%.