_target_: text_recognizer.models.vqvae.VQVAELitModel interval: step monitor: val/loss latent_loss_weight: 1.0