Problem when deploying the model with Tensorflow serving about thai-word-segmentation HOT 4 CLOSED

I didn't give specific names for the input variables in the prediction graph so the names are default what TF gives and are different from what the save_model function specifies. Please look at predict_example.py how to obtain the input variables.

from thai-word-segmentation.

@statguy Thank you so much for replying.

Did you mean these input variables?

g_inputs = 'IteratorGetNext:1' g_lengths = 'IteratorGetNext:0' g_training = 'Placeholder_1:0' g_outputs = 'boolean_mask_1/Gather:0'

Unfortunately, I have tried them but also got an error related to tensor alias.
grpc.framework.interfaces.face.face.AbortionError: AbortionError(code=StatusCode.INVALID_ARGUMENT, details="input tensor alias not found in signature: IteratorGetNext:0. Inputs expected to be in the set {lengths, inputs, training}.")

client.py

text = "ทดสอบ"
inputs = [ThaiWordSegmentLabeller.get_input_labels(text)]
lengths = [len(text)]
request.model_spec.name = 'word'
request.model_spec.signature_name = 'word_segmentation'

g_inputs = 'IteratorGetNext:1'
g_lengths = 'IteratorGetNext:0'
g_training = 'Placeholder_1:0'
g_outputs = 'boolean_mask_1/Gather:0'

request.inputs[g_inputs].CopyFrom(
tf.contrib.util.make_tensor_proto(inputs,shape=[inputs.shape[0]],dtype=tf.int64))
request.inputs[g_lengths].CopyFrom(
tf.contrib.util.make_tensor_proto(lengths,dtype=tf.int64))
request.inputs[g_training].CopyFrom(
tf.contrib.util.make_tensor_proto(False,dtype=tf.bool))

from thai-word-segmentation.

I updated predict_example.py to use the signature names. Does it help?

from thai-word-segmentation.

@statguy Thank you so much for fixing the issue and I would like to thank again for contributing such an amazing application.

This model works perfectly with my training data. I am going to write an article about how I apply it in my domain and I will refer to your article and Github.

from thai-word-segmentation.