Use the pretrained BERT based sentiment classification.

a$)$ Measure train speed when running in CPU and GPU. Measure how long does it take to train one iteration?

b$)$ Change the classification head by using a linear transformation that transforms features to $200,$ use RELU, use a linear transformation that transforms features to num$\_$labels, use Softmax, run training for $5$ epochs with two different learning rates and measure the accuracy for each.