Looking for LSTM seq2seq example

I’m experimenting with LSTM models, but can’t find a simple guide/example for a time series (values not text) model predicting n steps into the future (like horizon in fb prophet). Ideal case would be with another “regressor” (like temperature) as an additional dimension for value at timestep x.

My main problem is I’m not sure whether I’m keeping stateful “the right way” and would like to learn from a proper implementation.
My series predicted on 8->1 and cyclically shifted with a new prediction (last7+newpred) - >1 etc. cycles too fast or gets instable very fast.
Any links / examples would be great.
Thanks in advance, Andy