Data Scarcity and Model Sparsity: Mixtures-of-Experts Overfit More to Repeated Data
Atindra Jha, Margaret Li, Jure Leskovec, Percy Liang and Luke Zettlemoyer (Stanford University and University of Washington) run the first controlled study of data repetition for Mixture-of-Experts language models and find that sparse models overfit to repeated data faster than dense models of matched active size.
Ask this paper
Setup: Repetition rates are varied across single- and multi-domain data mixes and across MoE configurations, including expert count and granularity, for models from 80M to 1B active parameters (8.5B total).
Headline result: 80M dense models repeat data more than 8x with minimal degradation. MoEs start to degrade at 4x and fall below dense models after 32x, losing the advantage they hold on all-unique data.
Total parameters set the risk: The overfitting effect grows with sparsity and tracks total rather than active parameter count, so adding experts raises exposure to repetition.
Regularization partly fixes it: Dropout helps, and strong masking-based regularization lets MoEs beat dense models even at more than 64x repetition. No method matches training on all-unique data.
Mechanism: MoE routing stabilizes early in training in every setting, and expert specialization correlates with overfitting to the repeated data.
Abstract
As the supply of human-written text is exhausted, it has become standard practice to repeat language model training data. Prior work has studied data repetition for densely activated Transformers, but the effects of data repetition remains largely unexplored for recently dominant sparse architectures such as Mixture-of-Experts (MoE), despite their increased compute efficiency. We vary data repetition rates across single- and multi-domain data mixes, and across MoE settings, including expert count and granularity. We consistently find, for models ranging from 80M to 1B active (8.5B total) parameters, that MoEs degrade more rapidly under data repetition. This effect increases with sparsity, dictated by total rather than active parameters. While 80M dense models can repeat data over 8x with minimal degradation, MoEs instead begin to suffer at 4x, and deteriorate rapidly, ceding their performance benefits in all-unique data settings to underperform dense models after 32x. We experiment with existing regularization methods as a potential remedy. We find that some methods, such as dropout, can mitigate overfitting. In particular, with strong masking-based regularization, MoEs are able to outperform dense models even when data is repeated more than 64 times. However, no method fully matches the performance of all-unique training data. Finally, we analyze internal mechanisms correlated with MoE overfitting in high repetition regimes, and find that MoE routing universally stabilizes early in training, and that expert specialization correlates with overfitting to repeated data. In sum, our work addresses the adverse interactions between sparsity and data repetition: we present evidence for the core mechanisms of overfitting and its potential remediation, and suggest promising avenues for future methods to reduce over-specialization in model parameters by disrupting memorization patterns.