Possibility for custom split between training and validation set in CVs

I would be nice to have the possibility to provide a custom split between the training and validation set (similar to the built in keras option validation_split) so that e.g. the following cross-validation is possible:

# fold 0: |         Tr (40%)              | Val (10%) |                   Test (50%)              |
# fold 1: |         Test (50%)                        | Val (10%) |         Train (40%)           |