[ ] Tutorial for fine-tuning on SQuAD 1.1. The QA dataset pre-processing is more complicated than the classification example and deserves more description.
[ ] improve SQuAD 1.1 with extra fine-tuning epochs on TriviaQA dataset
[ ] reproduce result on SQuAD 2.0
[ ] example for finetuning BERT for other QA datasets
Thanks to @fierceX we now have an fine-tuning example of BERT for SQuAD 1.1. https://github.com/dmlc/gluon-nlp/pull/493
Followup works:
[ ] Tutorial for fine-tuning on SQuAD 1.1. The QA dataset pre-processing is more complicated than the classification example and deserves more description.
[ ] improve SQuAD 1.1 with extra fine-tuning epochs on TriviaQA dataset
[ ] reproduce result on SQuAD 2.0
[ ] example for finetuning BERT for other QA datasets