I haven’t found a good way to train a model, I referencing docs is fairly easy, but from my understanding you would need access to the same amount of compute power that was used to train the original model; in order to update its weights and train your own model