Don't Half-listen: Capturing Key-part Information in Continual Instruction Tuning
This addresses the problem of maintaining performance across tasks during continual learning for LLMs, representing an incremental improvement over existing methods.
The paper tackles catastrophic forgetting in continual instruction tuning for large language models by proposing a method based on Key-part Information Gain to dynamically replay data and refine training objectives, achieving superior performance on both seen and held-out tasks.
Instruction tuning for large language models (LLMs) can drive them to produce results consistent with human goals in specific downstream tasks. However, the process of continual instruction tuning (CIT) for LLMs may bring about the catastrophic forgetting (CF) problem, where previously learned abilities are degraded. Recent methods try to alleviate the CF problem by modifying models or replaying data, which may only remember the surface-level pattern of instructions and get confused on held-out tasks. In this paper, we propose a novel continual instruction tuning method based on Key-part Information Gain (KPIG). Our method computes the information gain on masked parts to dynamically replay data and refine the training objective, which enables LLMs to capture task-aware information relevant to the correct response and alleviate overfitting to general descriptions in instructions. In addition, we propose two metrics, P-score and V-score, to measure the generalization and instruction-following abilities of LLMs. Experiments demonstrate our method achieves superior performance on both seen and held-out tasks.