Who cares if the code is open source, or pre-training weights are released? Virtually every Masters in CS student in 2024 is building this from scratch. The differentiator is the training dataset, or at worst, the weights after fine tuning the model.