Model Compression And Efficient Inference