i use TensorRT-5.1.5.0, cuda9.0
the key code is :
lstm_cell_1 = tf.contrib.rnn.BasicLSTMCell(n_hidden, forget_bias=0.0, state_is_tuple=True)
lstm_cell_2 = tf.contrib.rnn.BasicLSTMCell(n_hidden, forget_bias=0.0, state_is_tuple=True)
lstm_cell_3 = tf.contrib.rnn.BasicLSTMCell(n_hidden, forget_bias=0.0, state_is_tuple=True)
lstm_cells = tf.contrib.rnn.MultiRNNCell([lstm_cell_1, lstm_cell_2, lstm_cell_3], state_is_tuple=True)
when i convert pb to uff:
DEBUG [/mnt/Data/yunjian/py_ann/envs/py36/lib/python3.6/site-packages/uff/converters/tensorflow/converter.py:71] Found input rnn/MultiRNNCellZeroState/BasicLSTMCellZeroState/zeros_1/Const
Converting rnn/MultiRNNCellZeroState/BasicLSTMCellZeroState/zeros_1 as custom op: Fill
when run inference:
[TensorRT] ERROR: UffParser: Validator error: rnn/MultiRNNCellZeroState/BasicLSTMCellZeroState/zeros_1: Unsupported operation _Fill
[TensorRT] ERROR: Network must have at least one output
how can i solve it , thanks