Exploring Forgetting in Large Language Model Pre-Training

TL;DR

This paper investigates the phenomenon of forgetting during large language model pre-training, introduces new metrics for detection, and proposes simple mitigation methods to improve entity memory retention.

Contribution

It is the first systematic study of forgetting in LLM pre-training, proposing new metrics and low-cost methods to mitigate forgetting.

Findings

New metrics better detect entity memory retention.

Simple mitigation methods reduce forgetting during pre-training.

Insights into learning dynamics and forgetting patterns.

Abstract

Catastrophic forgetting remains a formidable obstacle to building an omniscient model in large language models (LLMs). Despite the pioneering research on task-level forgetting in LLM fine-tuning, there is scant focus on forgetting during pre-training. We systematically explored the existence and measurement of forgetting in pre-training, questioning traditional metrics such as perplexity (PPL) and introducing new metrics to better detect entity memory retention. Based on our revised assessment of forgetting metrics, we explored low-cost, straightforward methods to mitigate forgetting during the pre-training phase. Further, we carefully analyzed the learning curves, offering insights into the dynamics of forgetting. Extensive evaluations and analyses on forgetting of pre-training could facilitate future research on LLMs.