Hi, I am wondering if it is possible to create a custom component that only augments training data (with synonyms, random characters, etc) so I can train the model using the appended augmentations of the original training data. I am confused because all discussions about this seem to involve training and processing new incoming user messages, which are irrelevant in this case.
I think it would probably better to do this as part of a custom Training Data Importers
I have successfully implemented the custom training data importers and it works fine when training. However my custom training data importer is not called during cross-validation. Have I missed something or is the feature not yet available to use when cross-validating?