Track the model's learning progress during training


I am training a model to predict verb inflections, e.g.:
input “miscalibrate⦅V PST⦆”
target “miscalibrated”
(The data is tokenized using pyonmttok)

What I would like to do is track the model’s learning during the training process - what incorrect/correct verb forms are generated as the data is iterated over.
I have seen this done in papers (for example here: Recurrent Neural Networks in Linguistic Theory: Revisiting Pinker and Prince (1988) and the Past Tense Debate - ACL Anthology) but I cannot find an answer as to how this is done.

Thank you in advance for any tips!