To train the base QG-Net model, run python qg_net_base.py To train the BERT based QG-Net model, run python qg_net_bert.py

An example of running inference to compute the metrics has been shown in inference.py

The dataset used along with the trained weights can be found at - https://drive.google.com/drive/folders/1Jlo4wUBKDapEhR6D2-uaR3N8cctCy8lF?usp=sharing

RNNEncoder: This class encode the input using a bidirectional input and produces representations for each word in the context

RNNDecoder: This class decode the output of Encoder in sequential manner with attention and pointer network (used for calculating probabilities from source side).
QGNet: This class combine the encoder and decoder. the forward pass uses teacher forcing if a target is available. else special token '' is used as input for the first decoder step.

QADataset: Dataset used for converting data to form usable by Huggingface transformers

BertEncoder: We used pre-trained encoder from bert 'bert-base-uncased' which assume capital and small alphabets as same.

BertDecoder: The decoder will decode the encoder output and calculate probability distribution using attention and pointer network. Uses a BertLMHeadModel for decoding.

BertPointerNetwork-: This class combine the encoder and decoder. the forward pass uses teacher forcing if a target is available. else special token '[CLS]' is used as input for the first decoder step.

train_step: this function will check our model on train data and return how much loss we obtain for this data.
eval_step:this function will evalute our model on test data and test how much loss we obtain for this data.
train_model:The function will train our model for certain epochs which are 10 in our case.also this function is used to print losses and to save model at pt format.