This is the codebase for the SQUASH pipeline which uses pretrained question generation and question answering modules to converting input paragraphs into trees of question-answer pairs.
populate_input.py
- Choose a random instance from QuAC for SQUASHing.extract_answers.py
- Extract individual sentences and entities which will be used for question generation.combine_qa.py
- Module to combine generated questions with outputs from the question answering module.filter.py
- Filter the pool of generated question-answer pairs and carry out binning.squad_eval_utils.py
- Utilities to provide F1 overlap between two answer spans, used for the filtering process.pipeline.sh
- Run the whole SQUASH pipeline.pipeline_custom.sh
- Run the SQUASH pipeline with custom inputs specified as the first command line argument. Make sure you have createsquash/temp/$KEY
andsquash/temp/$KEY/metadata.json
before running this.