MSE loss function decreasing very slowly

The reason for your model converging so slowly is because of your leaning rate (1e-5 == 0.000001), play around with your learning rate. I find default works fine for most cases.

try: 1e-2

or you can use a learning rate that changes over time as discussed here