Re-package tensorflow-serving-api==1.9.0 (discussion about future of this package)

[illagrenan]

Release: https://pypi.org/project/tensorflow-serving-api/1.9.0rc2/

[ghego]

1.9.0 has been out for a while. Is this going to be repackaged?

[illagrenan]

I will try to release a new version later today (Thu) or tomorrow (Fri).

[ghego]

actually, I was just checking the official package and it seems to be supporting py3 too. have a look: https://pypi.org/project/tensorflow-serving-api/#files

[illagrenan]

You're right! That's new. So I think my project is no longer needed. 😄 It looks like tensorflow-serving-api supports Python 3 from version >=1.8.0.

[ghego]

as far as I understand they did release it with tf 1.9 but somehow the tensorflow_model_server command is not available. I'm confused..

[ghego]

My bad, I needed to update the linux package too. https://www.tensorflow.org/serving/setup#aptget

[spate141]

@illagrenan Thank you for all your work! 🎉

[ghego]

The official package seems to require tensorflow and does not work with tensorflow-gpu. @illagrenan thanks for the good work! maybe worth keeping this alive for one more release while they fix the issue?

[illagrenan]

@ghego I understand the problem. I have never understood the splitting of the GPU and the non-GPU version of Tensorflow.

You can try this:

pip install 'tensorflow-serving-api~=1.9.0' --no-dependencies
pip install 'tensorflow-gpu~=1.9.0'

[ghego]

Thanks that solved the issue. Do you know if it can be done in a requirements file or conda environment.yml ? I don't seem to have found a way to do that.

[illagrenan]

Unfortunately, it looks like it is not possible in requirements.txt. And I'm not familiar with Conda. 😞 Ref: https://stackoverflow.com/questions/31798280/pass-no-deps-in-pip-requirements-txt

[illagrenan]

But it shouldn't be so hard to handle this situation with your build script / Dockerfile. For example, I would do this in Dockerfile:

RUN pip install -r requirements.txt
RUN pip uninstall tensorflow-gpu
RUN pip install tensorflow