This repository is about using BERT (Bidirectional Encoder Representations from Transformers) for the dataset Spokensquad, which includes questions, answers and paragraphs where the model is supposed to extract the answers for the questions posed. "Bert-base-uncased" model by hugging face is used. The dataset is trained and validated using Palmetto cluster since the the training set has 37,111 question answer pairs and testing set has 5,351 question answer pairs. Accuracy and Training loss is plotted by training and validating data.