a-ware/roberta-large-squad-classification

This is roberta-large model finetuned on SQuADv2 dataset for question answering answerability classification . The model is simply an Sequenceclassification model with two inputs (context and question) in a list . The result is either [1] for answerable or [0] if it is not answerable . It can be used to filter out which context is good to give into the QA model to avoid bad answers . It was trained over 4 epochs on squadv2 datasets . The accuracy of the model was 90.48% of the accuracy of QA is now 90% . The data was created with A-ware UG UG by A-,