Relay.frontend.from_tensorflow fail to load LSTM


#1

I am working on loading a tf model, and I got these errors

(missing_operators))
NotImplementedError: The following operators are not implemented: {‘TensorArrayReadV3’, ‘TensorArrayWriteV3’, ‘TensorArrayGatherV3’, ‘TensorArrayV3’, ‘TensorArraySizeV3’, ‘TensorArrayScatterV3’}

I located the error, and I am sure the error was caused by the bidirectional lstm layer.

sequence = tf.squeeze(conv7, axis=1, name=‘squeeze’) # (1,25,512)

fw_cell_list = [tf.nn.rnn_cell.LSTMCell(nh, forget_bias=1.0) for
nh in [hidden_nums] * hidden_layers]

Backward direction cells

bw_cell_list = [tf.nn.rnn_cell.LSTMCell(nh, forget_bias=1.0) for
nh in [hidden_nums] * hidden_layers]

stack_lstm_layer, _, _ = rnn.stack_bidirectional_dynamic_rnn(
fw_cell_list, bw_cell_list, sequence,
dtype=tf.float32,
scope=‘lstm’
)

But I notice that the tvm do support lstm, maybe I just need to write my code in another way? I am new to tvm , could anyone give me some suggestions ? Thank you !