Mitigating Catastrophic Forgetting in Large Language Models with Self-Synthesized Rehearsal
This addresses the problem of data unavailability in real-world continual learning for LLM users, though it is an incremental improvement over existing rehearsal-based approaches.
The paper tackles catastrophic forgetting in large language models during continual learning by proposing a Self-Synthesized Rehearsal framework that generates synthetic instances for rehearsal, achieving superior or comparable performance to conventional methods while being more data-efficient.
Large language models (LLMs) suffer from catastrophic forgetting during continual learning. Conventional rehearsal-based methods rely on previous training data to retain the model's ability, which may not be feasible in real-world applications. When conducting continual learning based on a publicly-released LLM checkpoint, the availability of the original training data may be non-existent. To address this challenge, we propose a framework called Self-Synthesized Rehearsal (SSR) that uses the LLM to generate synthetic instances for rehearsal. Concretely, we first employ the base LLM for in-context learning to generate synthetic instances. Subsequently, we utilize the latest LLM to refine the instance outputs based on the synthetic inputs, preserving its acquired ability. Finally, we select diverse high-quality synthetic instances for rehearsal in future stages. Experimental results demonstrate that SSR achieves superior or comparable performance compared to conventional rehearsal-based approaches while being more data-efficient. Besides, SSR effectively preserves the generalization capabilities of LLMs in general domains.