Now I have been looking RNN's for a while. And the vanilla version seems pretty obvious to me, a model trains by running each word in a sequence through the network one at a time. Here in the rolled and unrolled version:
Now i went on to look at lstm networks, where this model is used for a LSTM "cell"
The way it is decribed to me in many articles and videos, this cell is a model of what it looks like inside of a single neuron of the lstm network. This makes sense to me. However, I then find a model like this: