currently in the training notebook, we first create the dataset, and then load it again because we only pass the ConfigHolder to train_model. There should be an option to pass the dataset, and then we check the dataset config for correctness (or, if overridden, modify the ConfigHolder)
currently in the training notebook, we first create the dataset, and then load it again because we only pass the
ConfigHolder
totrain_model
. There should be an option to pass the dataset, and then we check the dataset config for correctness (or, if overridden, modify theConfigHolder
)