Illegal Instruction when running ./predict

[tuancorn97]

Hi there,
I have just downloaded your models and installed all the requirement, but it did not work when I tried predicting the images with
./predict \ --docker-image nima-cpu \ --base-model-name MobileNet \ --weights-file $(pwd)/models/MobileNet/weights_mobilenet_technical_0.11.hdf5 \ --image-source $(pwd)/src/tests/test_images/42039.jpg

It then showed an error of Illegal Instruction in the file entrypoint.predict.cpu.sh:
entrypoints/entrypoint.predict.cpu.sh: line 12: 6 Illegal instruction (core dumped) python -m evaluater.predict --base-model-name $BASE_MODEL_NAME --weights-file $WEIGHTS_FILE --image-source $IMAGE_SOURCE

[Christopher Lennan]

Hm, not sure what is happening there. We upgraded to TF2.0 today, can you try again with the newest version and see if it works?

[RonSv]

Hello. Same problem. With newest version also.

[Christopher Lennan]

Which operating system and Docker version are you running?

[RonSv]

Probably the reason is cpu. (Intel Xeon X5675) tensorflow/serving#1342

[RonSv]

I solve it building tensorflow binary without avx instruction. There are compiled WHL for different processors - https://github.com/yaroslavvb/tensorflow-community-wheels/issues

Then add in dockerfile

COPY tensorflow-2.0.0-cp36-cp36m-linux_x86_64.whl /tensorflow-2.0.0-cp36-cp36m-linux_x86_64.whl

RUN pip uninstall tensorflow -y \
    && pip install tensorflow-2.0.0-cp36-cp36m-linux_x86_64.whl \
    && rm -rf /tensorflow-2.0.0-cp36-cp36m-linux_x86_64.whl

[Sps113]

get from here https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-2.0.0-cp36-cp36m-manylinux2010_x86_64.whl