In processing datasets, there is a little difference in training dataset and eval dataset. I want to use a parameter like mode=train/eval
to implement the different processing method in a function, otherwise I have to write two functions which are mostly equal but different in some details, and pass them in train_dataset.map and eval_dataset.map seperately.
Hi! Yes, passing two different functions separately is one option. Another is to have one function (but still two map
calls, one for each split) and pass different fn_kwargs
in map
to change the mode.
Thanks a lot! I missed the fn_kwargs
and that is just what I want.