Can Scale Save Us From Plasticity Loss in Large Language Models?
For researchers in continual learning and large language models, this work provides empirical evidence that plasticity loss remains a challenge in modern LLMs, with scaling laws suggesting that simply increasing model size is insufficient to overcome it.
This paper investigates plasticity loss in GPT-style Transformer models (5M-314M parameters) trained on multilingual continual learning, finding that plasticity loss persists and follows a predictable scaling law where larger models delay but do not prevent it. The phenomenon also occurs under stationary training, challenging its exclusivity to continual learning.
The loss of plasticity - the ability of a network to learn new information after having already learned older information - is a fundamental challenge in creating artificial neural networks capable of continual learning. Although this phenomenon has been known for decades, it has mostly been studied in older, relatively small architectures and rarely in natural-language domains. To determine whether loss of plasticity remains a problem in the modern transformer-based LLM paradigm, we study plasticity loss in GPT-style Transformer models trained on a multilingual continual learning problem. Consistent with prior work, we find evidence of plasticity loss across models ranging from 5M to 314M non-embedding parameters, as measured by deterioration on a held-out Vietnamese probing task. We further find that the onset of plasticity loss follows a predictable scaling law, growing sublinearly with model size. These results suggest that larger models may delay the measurable effects of plasticity loss, but that increasing parameter count alone is likely to be insufficient to completely prevent it. We also find evidence of plasticity loss under stationary multilingual training, challenging the view that the phenomenon is exclusive to continual learning with abrupt task changes. Overall, our results suggest that even large Transformer language models trained on natural-language will eventually lose the ability to efficiently adapt to new data after sufficiently long training, in both continual and stationary settings.