This respository contains the code used for my thesis Using Language Models to Generate Text-to-SQL Training Data An Approach to Improve Performance of a Text-to-SQL Parser. This code will generate SQL question/query pairs using GPT-3's API. Running the script src/gen_sql.py will call the API according to the config values specified in src/configs/gpt.yaml. The default value for the configuation can be found in src/config.py.
The code used to train and evaluate the model was from the PICARD repo.
The generated text-to-SQL data can be downloaded here.