Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Pip package for tensorflow serving does not support GPU #668

Closed
dakshvar22 opened this issue Nov 23, 2017 · 7 comments
Closed

Pip package for tensorflow serving does not support GPU #668

dakshvar22 opened this issue Nov 23, 2017 · 7 comments

Comments

@dakshvar22
Copy link

i installed tensorflow serving pip package which comes with tensorflow 1.4 as a dependency. But it seems like the tensorflow package inside does not support GPU. Is there an alternative pip package?
I even installed tensorflow(GPU version) separately and then installed tensorflow serving pip package, but that seems to overwrite the GPU version. Any way by which I can stop tensorflow serving pip package from reinstalling tensorflow?

Thanks

@kirilg
Copy link
Contributor

kirilg commented Nov 27, 2017

That's interesting, so 'tensorflow-gpu' overwrites the 'tensorflow' PIP package?
The TensorFlow Serving pip package (tensorflow-serving-api) is a very thin API library that just includes the protos of the PredictionService, so it theory it shouldn't depend on GPU. What you can try doing is installing 'tensorflow-serving-api' (which will require tensorflow), and then install tensorflow-gpu which will overwrite tensorflow. I believe 'tensorflow-serving-api' should continue working, it will just get your newly installed GPU tensorflow.

@jorgemf
Copy link

jorgemf commented Dec 19, 2017

@dakshvar22 as far as I know the open source of tensorflow serving does not compile with GPU support, there are issues.

@ps48
Copy link

ps48 commented Jan 8, 2018

@dakshvar22 Even I am facing the same issue. Still didn't find any to use serving pip installable with GPU. Any updates on this?

@ps48
Copy link

ps48 commented Jan 8, 2018

The error is not in Serving pip API as it is only used in client side. Rather the issue is on the Apt-get Install for tensorflow-model-server. Issue #549 comment by @kirilg mentions apt-get doesn't support GPU as of now. I think we can close this issue.

@jorgemf
Copy link

jorgemf commented Jan 8, 2018

@ps48 the tensorflow serving pip package doesn't support GPU because before it couldn't be compiled easily with GPU support before. I think recent versions can be compile with GPU support with the command you saw if the issue #549, but the package hasn't been compiled with this options yet. So either you use a package without GPU or you compile it yourself. It is not related with apt-get.

@ps48
Copy link

ps48 commented Jan 9, 2018

@jorgemf thanks for the pormpt reply. As far as i see this means serving-api as well as tensorflow-model-server have GPU support and just need to be compiled locally. Please correct me if I'm wrong.

@jorgemf
Copy link

jorgemf commented Jan 9, 2018

@ps48 The serving-api doesn't need GPU support as it is only to make requests to the server. Tensorflow-model-server does need to be compiled with GPU support as the other issue said. This might help: https://gist.github.com/jorgemf/c791841f769bff96718fd54bbdecfd4e

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants