Exploring Forgetting in Large Language Model Pre-Training
This work addresses forgetting in LLM pre-training, a domain-specific issue that could enhance model stability and knowledge retention for AI researchers and practitioners.
The paper tackled the problem of catastrophic forgetting during large language model pre-training, introducing new metrics to better detect entity memory retention and exploring low-cost methods to mitigate it, with results showing improved retention compared to traditional perplexity-based approaches.
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.