Villanova LLM 2B Family

Villanova LLMs is a family of language models built through a progressive pipeline ranging from basic pre‑training to multimodal capabilities. The project introduces a transparent multilingual model, starting with an experimental 2‑billion‑parameter version and extending to the final 14‑billion‑parameter model. The model is trained exclusively on public, multilingual data across five European languages, with complete documentation of data processing and training.

Overall, the Villanova LLMs family provides a modular and extensible foundation for building advanced AI systems, where the robustness of the base model plays a crucial role in ensuring the quality and effectiveness of all instruction‑tuned and multimodal variants derived from it.