How to train LSTM with variable time-stemp?

Question

Example of data:

x = np.array([
      [[1.5, 0, 3], [2, 5, 1], [3, 5, 6]],
      [[3, 2, 3], [4, 5, 0], [4, 5, 0], [4, 5, 0], [6, 5, 0]],
      [[1, 2, 3], [2