Foundational pre-trained models across all versions. These models are intended for further fine-tuning, domain adaptation, and research.