Train CNN-LSTM model in Pytorch

I have a model consisting of 5 steps of LSTM, where the outputs at each step are one-dimensional and the input at each stage is a 3-dimensional vector which is the output of CNN. The CNN corresponding to each step consists of 4 convolutional layers and one pooling, and the CNNs at each step are different (so I have a total of 5 independent CNNs, one for each LSTM step). I tried with no success to train to model and to get a good loss in the training and test sets. I am looking for somebody who can train the model and give a good loss for the train and the test set. It is important to give figures of the loss over iterations and how it converges and other figures showing the good results for training and testing and the code to generate them.
Attached is the training code "modified_code.py" and the testing code "TestingCNNLSTM.py" and the needed data for:
training: input "Weather-200.pkl", "Irrigation-200.pkl" and output "Y-200.pkl"
testing: input "Weather-TEST", "Irrigation-TEST.pkl" and output "Y-YEST.pkl"