This repository contains code for training and evaluating a text-to-text transformer model for question generation.
The models are trained on the deepset/germanquad dataset.
The model is based on the mT5 model.
These models are fine-tuned on the aforementioned datasets for the amount of epochs specified in the script_config.json file.
The training is done using the run-qg.py script. The script can be run with the following command:
python run-qg.py script_config.jsonThe script can be configured using the script_config.json file.
The evaluation is done using the run-qg.py script as well.
For this, change the configuration to include the attribute "do_eval": true and run the script.
Feel free to set "do_train": false as well, if you don't want to train the model.
The inference is done using the run-qg.py script as well.
For this, change the configuration to include the attribute "do_predict": true and run the script.
Feel free to set "do_train": false and "do_eval": false as well, if you don't want to train or evaluate the model.
This project is licensed under the Apache License 2.0 - see the LICENSE file for details
- Huggingface for providing the datasets, transformers and evaluate libraries
- deepset for providing the German Question Answering Dataset (GermanQuAD)
If you have any questions or suggestions, feel free to raise an issue.