Best approach to fine-tune on custom dataset

[haimat]

Hello, we want to fine-tune your model(s) on our custom dataset. But doing so we don't want to loose the information already trained into your models. Therefore, when fine-tuning a docTR model with custom data, is it recommended to include the official datasets you have trained the models on also in the training data, or should we simply use our custom dataset for the fine-tuning?

Also, if you recommend to include the "official" training data into the fine-tuning dataset, where would I get that from including all the annotations?