Anyone use huggingface with Amazon SDK?
huggingface_estimator.fit({‘train’: training_input_path, ‘test’: test_input_path})
– The input paths above are always CSVs in the sdk documentation. I’m creating a custom NER dataset to train a Bert model, so I was wondering if those input paths could be a pickle if I used the dataset.save_to_disk feature hugging face offers.
If that sounds like non-sense, please don’t hesitate to inform of a better flow.
Thank you so much.