具有3x8矩阵特征的张量流LSTM

I am trying to design a python 3 Tensor flow LSTM to work on a data set. My problem is that the feature set is a matrix - 8 sets of numbers, where each set is 1,2 or 3 numbers. Currently I'm storing the data as a 3x8 matrix where sets that only had 1 or 2 numbers originally have trailing -1 (actual data can never go below 0).I am trying to design a python 3 Tensor flow LS