Added tensorflow-cpu docker image #15
Conversation
| @@ -0,0 +1,89 @@ | |||
| FROM tensorflow/tensorflow:latest-py3 as mlbench-worker-base-cpu | |||
There was a problem hiding this comment.
Thanks a lot for the PR. looks great. We should use a fixed-version base image so updates to the TF image don't accidentally break our images
|
What is the status on this pull request? It seems to address issue #13, if we fix the Travis failure and use a fixed-version base image. |
@giorgiosav This has been fixed in #26 . I guess both #13 and this PR can be closed. Better check with @Panaetius |
|
I'm not sure on the state on this. The main goal was having a tensorflow cpu only image (without cuda dependency so it runs in a gcloud cluster that doesn't have GPUs, as tensorflow requires cuda by default even if you don't use a gpu in your code). As far as I know, that part works. The other part is having tf compiled with openmpi support as a base image, so we can directly compare tf and pytorch openmpi. I'm not sure what the state of that is, and we have since given tf a lot lower priority, so it's not that important, which is why this PR has been around for ages. |
|
@martinjaggi should we close this? |
No description provided.