[tensorflow serving]Executor failed to create kernel. Invalid argument: NodeDef mentions attr 'index_type' not in Op

1
votes

When I run grpc to request serving the serving post this fail~I make sure it all runing in TF 1.6.0

2018-03-29 17:09:35.204261: E external/org_tensorflow/tensorflow/core/common_runtime/executor.cc:643] Executor failed to create kernel. Invalid argument: NodeDef mentions attr 'index_type' not in Op output:T; attr=T:type>; NodeDef: rnn/MultiRNNCellZeroState/BasicLSTMCellZeroState/zeros = Fill[T=DT_FLOAT, _output_shapes=[[?,8]], index_type=DT_INT32, _device="/job:localhost/replica:0/task:0/device:CPU:0"](rnn/MultiRNNCellZeroState/BasicLSTMCellZeroState/concat, rnn/MultiRNNCellZeroState/BasicLSTMCellZeroState/zeros/Const). (Check whether your GraphDef-interpreting binary is up to date with your GraphDef-generating binary.). [[Node: rnn/MultiRNNCellZeroState/BasicLSTMCellZeroState/zeros = Fill[T=DT_FLOAT, _output_shapes=[[?,8]], index_type=DT_INT32, _device="/job:localhost/replica:0/task:0/device:CPU:0"](rnn/MultiRNNCellZeroState/BasicLSTMCellZeroState/concat, rnn/MultiRNNCellZeroState/BasicLSTMCellZeroState/zeros/Const)]]

1
tensorflow-serving
because of the tf version. For example, I train the net in tf 1.6, but my c++ is tf 1.4. So, for me. if don't want to change c++/tf version. I export graph use python/tf 1.4. but train use 1.5 or later - Fisherman

1 Answers

1
votes

This long dump might obfuscate the important piece of information that is:

"Check whether your GraphDef-interpreting binary is up to date with your GraphDef-generating binary"

The TensorFlow version in the tensorflow_model_server might be older than that used for building the model and hence does not know how to instantiate that operator.

You can either try rebuild your model on an older TensofFlow release or try creating the Tensorflow serving with the latest source.