some scripts used to process bird text to sql training data
- Get table schemas from sqlite file and table descriptions from csv files.
- Clean and format csv data into json combined with 'create table' schemas.
- Translate descriptions (column and value descriptions) to Chinese using QWen 14B model.
- Translate train data (questions and hints) to Chinese, with QWen 14B.
- Compose the training data for text to sql finetuning.