The distilbert-base-uncased model has been trained to treat spaces as part of the token. As a result, the first word of the sentence is encoded differently if it is not preceded by a white space. To ensure the first word includes a space, we set add_prefix_space=True. You can check both the model and its paper here: distilbert/distilbert-base-uncased

Unfortunately you have to read through the model's technical report or research around to see how they are trained. Happy learning! :)