Tensorflow Serving grouped convolutions

Question

I am doing handwriting detection using keras and tensorflow. I have the model ready and working with hdf5 file. When I tried to serve it using tensorflow serving, I get the following error :

grpc.framework.interfaces.face.face.LocalError: LocalError(code=StatusCode.UNIMPLEMENTED, details="Generic conv implementation does not support grouped convolutions for now.
 [[{{node conv2d_1/convolution}} = Conv2D[T=DT_FLOAT, _output_shapes=[[?,40,40,20]], data_format="NHWC", dilations=[1, 1, 1, 1], padding="SAME", strides=[1, 1, 1, 1], use_cudnn_on_gpu=true, _device="/job:localhost/replica:0/task:0/device:CPU:0"](_arg_conv2d_1_input_0_0, conv2d_1/kernel/read)]]")

I used the two files mentioned in this article to generate the model from my hdf5 file(which was working). Please provide any inputs that may help me solve this?

Answer 1

I had a similar error but with "Fused conv implementation does not support..." instead of "Generic conv implementation does not support..." as the error message. It turned out to be a TensorFlow version issue for me too -- version 1.13.2 was what I needed. To fix it, I did this BEFORE doing "import tensorflow as tf":

 !pip install tensorflow==1.13.2

This solved it for me.