_target_: text_recognizer.models.transformer.TransformerLitModel
interval: step
monitor: val/loss
ignore_tokens: [ <s>, <e>, <p> ]
start_token: <s>
end_token: <e>
pad_token: <p>