Closed nelson-liu closed 7 years ago
looks like it's mostly OK...just complaining that there's no CUDA. i'll force CPU in the travis config?
nevermind, looks like it needs cuda even if it runs in CPU-only mode :/ yet another step to add to the already lengthy process of running the python server.
tensorflow-gpu seems to have the ability to run CPU code...so I think it is a reasonable default.
The main reason i made this change is because setting up the solver to run on docker installs from requirements.txt, and we probably want the GPU version on there.