lstm network loss does not drop

You may want to have a look at this post. I use RNNs for classification tasks regularly, and never require view(), reshape() or resize() which can be very harmful when not used carefully.