Fine-tuning Flan-T5 for question answering using scraped Quora data

Follow the full discussion on Reddit.
Recently I scraped 56,400 question/answer pairs off Quora, and trained Flan-T5 on the resulting dataset. I released the dataset and model on HuggingFace, which you can find in the comments. I plan to continually add to the dataset, but proxy costs are pretty expensive since Quora is hella bloated.