We often encounter situations where we need to preprocess data differently depending on split types such as train, valid, and test.
However, while DatasetDict.map has features to pass rank or index, there's no functionality to pass split types.
Therefore, I propose adding a 'with_splits' parameter to DatasetDict, which would allow passing the split type through fn_kwargs.
Hello datasets!
We often encounter situations where we need to preprocess data differently depending on split types such as train, valid, and test.
However, while DatasetDict.map has features to pass rank or index, there's no functionality to pass split types. Therefore, I propose adding a 'with_splits' parameter to DatasetDict, which would allow passing the split type through fn_kwargs.