| This is the HuDS (https://huds.uni-goettingen.de) warmed-up model trained on the mixed dataset of music, cellular automata, and PCFGs for the continued training on natural language for the BabyLM 2026 Challenge. The respective trained model is supplied separately. |