mbart_qg

use mbart model to train a chinese question generating model

install requirements.txt

pip install -r requirements.txt

download training data

download train data from Delta Reading Comprehension Dataset and put the json files in "drcd" folder

mbart_qg
│───DRCD
│   │   DRCD_dev.json
│   │   DRCD_test.json
│   │   DRCD_training.json
│   README.md
│   config.py
│   ...

set training parameters

set Hyperparameter in config.py

training

run

python train.py

train.py will read data from DRCD_dev.json and DRCD_training.json and concatenate the context and answer into model input label is the question in dataset. the model for every epoch will save in "model" folder

inference

run

python predict.py

predict.py will load the "best_model" set in then input context and question

Evaluation

run

python rouge_test.py

rouge_test.py will read DRCD_test.json and generate question. Then calculate ROUGE-L score of generate question and origi