Tag
ARMOR-DP is a modular framework for multilingual BabyLM that dynamically adjusts language sampling based on per-language validation loss, showing that the weakest language shifts during training and that dynamic protection outperforms fixed Chinese-protection baselines.