Training Efficient Neural Network Models Via Pruning And Knowledge Distillation