Sparse autoencoder interpretability for the Borzoi model - View it on GitHub