Closed radka-j closed 6 months ago
Something like 1-2% of the training data size and/or only focus on adding critical error examples to make the dataset (more) balanced.
This experiment is no longer of interest.
Something like 1-2% of the training data size and/or only focus on adding critical error examples to make the dataset (more) balanced.