How to initialize hidden layers of RNN with given sequences

I trained a simple 2-layer RNN on 1000 sequences, each with length = 100, and 3 features each. I used batch_size = 32, hidden_size = 8, num_layers = 2. After training, I need to initialize the hidden layers of the RNN with another given set of 1000 sequences, each with length = 50, and 3 features each to make predictions on test data. How can I initialize the 2 hidden layers, given the sequences are of different length (100 and 50) and different number of features in hidden_size during training = 8 and in initialized sequences = 3? Should I just resize the sequences to match the shape of the hidden state that I have from training the model?