Choose dataset split to use in data_config #645
Description
Is your feature request related to a problem? Please describe.
Currently we can only choose train split for our dataset using fms-hf..
Would request ability to load other splits of our dataset as main split.
Describe the solution you'd like
key in data_config which holds value to the split name to load for the dataset.
Describe alternatives you've considered
Additional context
tuning-config-recommender has ability to decide and check data object keys from other splits to decide on various fields like conversation_columns but the output of the tuning-config-recommender is not useful as fms-hf does not yet support loading dataset from other splits.