Train Single LSTM Using Multiple Training Sets Collected on Separate Instances

I’m a little confused on how to deal with training a single network using data from separate occasions. In my case, I collected experimental data in 30-second intervals for 23 tests. This means that I want to train, validate, and test my network using these 23 separate time series.

Is there a good way to organize my data so that the torch library recognizes separate sequences during training? Concatenating everything would not be a good idea since this would result in discontinuities in the data.