Do i use "h_n" or "output" of PyTorchs LSTM module in a fully-connected layer for a multi-step time series forecasting problem?

Question

My situation is that i have univariate time series data and want to make a multi-step prediction based on previous observations using PyTorchs LSTM module.

I want to use a LSTM in combination with a fully-connected layer to directly predict the next n time-steps as an output vector.

Here i learned the meaning of the outputs of a LSTM in PyTorch: What's the difference between "hidden" and "output" in PyTorch LSTM?.

But now im struggling with the input of the fully-connected layer. Do i use "h_n" or "output" and whats the difference?