Use of batch_size argument, passed through training_args
batch_size
argument is given by the researcher as a training_args
, nevertheless in practice this is not used in the experiment: in all examples provided, the batch_size
is further fixed as a default argument of the traning_data
method when the training plan is customized by the researcher. This mean, that if one want to change the batch size, he finally has to modify training_data
rather than simply update the training_args
.