Can you share the method for training my own scenario table?

Great job! My scene table is quite complex, containing Chinese, English, Japanese, Spanish, and so on. It has no borders and merged cells... So I might need to train it myself for better results? Could you share how to train it by myself? If there are more detailed hardware specifications and code, or reference links or learning articles, I would be very grateful!