8000 tensorflow vs tensorflow-gpu · Issue #30 · KevinMenden/scaden · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

tensorflow vs tensorflow-gpu #30

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.

Sign up for GitHub

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

Jump to bottom
Closed
zeehio opened this issue Feb 25, 2020 · 3 comments
Closed

tensorflow vs tensorflow-gpu #30

zeehio opened this issue Feb 25, 2020 · 3 comments

Comments

@zeehio
Copy link
Collaborator
zeehio commented Feb 25, 2020 •
edited
Loading

When using pip to install scaden, scaden always pulls tensorflow and never tensorflow-gpu.

The real problem is that pip is not able to deal with GPU variants, as described at tensorflow/tensorflow#7166. If pip was able, then tensorflow would not need to have those two separate packages and everything would work as expected.

There are three possible scenarios:

  • BAD: We depend on tensorflow, and it is not possible to install scaden using the tensorflow-gpu package because they conflict. (this is the current status)
  • UGLY: We publish to pypi two packages scaden and scaden-gpu, depending on tensorflow and tensorflow-gpu respectively.
  • ALSO UGLY: We optionally depend on tensorflow and tensorflow-gpu. The instructions to install scaden through pip would be: pip install scaden[cpu] or pip install scaden[gpu]. pip install scaden would not install any tensorflow variant and lead to an error (we could warn the user then, though).

I can easily implement ALSO UGLY, the community is split between the two ugly solutions. Do you have any preference, @KevinMenden ?
@KevinMenden
Copy link
Owner

I have thought about this in the past, and my preferred solution would have been to generate a scaden-gpu package - mostly because the majority of users (I believe) won't have a GPU and would just want to do pip install scaden

However now that I think of it , ALSO UGLY might be a better solution for now. I don't really like the idea of having two different packages for the same software .... so I guess we go with this for now?

@zeehio
Copy link
Collaborator Author
zeehio commented Feb 26, 2020 •
edited
Loading

Another option is leave only the CPU version on pip, as it is now, and provide docker images and conda instructions for cpu & gpu

@KevinMenden
Copy link
Owner

That might actually be the best version for now.
For GPU, you can basically just install Scaden via pip, then uninstall tensorflow and install tensorflow-gpu. And we can easiily provide a container that has this.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@zeehio @KevinMenden
0