The model used in human tutorial 'catlas-enformer-release-model_2.pth' has been fine-tuned to visual embedding, right?