As large language models (LLMs) scale, their training-token budgets must also increase to maintain an appropriate tokens-per-parameter ratio (TPP). However, high-quality domain data is much harder to scale than general web data. As model size and the training-token budget increase, the fraction of high-quality data in the training mixture tends to decrease.
Repeating the available high-quality data provides an effective way to counteract this dilution, but excessive repetition may lead to overfitting. A new paper from Hugging Face studies this trade-off under practical LLM scaling, where the training-token budget grows proportionally with model size.
For a fixed domain, the researchers found that, surprisingly, at a fixed TPP, the optimal repetition count mildly increases with model size. Across different domains, the optimal repetition count is strongly negatively correlated with the final validation loss of a domain: domains with lower loss can generally benefit from more repetitions. In contrast, the amount of unique domain data is only weakly related to the optimal repetition count.
These findings suggest that repetition counts tuned on smaller proxy models with the same TPP can provide a practical estimate for larger models, offering a cost-effective strategy for optimizing pretraining data mixtures.