This looks like an awesome project. I'd like to use it to control a small GPU cluster I have on my LAN.
Can you provide advice about how I might put back the GPU support? I don't mind maintaining my own fork.
@ibuildthecloud I share the same thinking as well. Why this feature is removed? It is such a need for ML applications!
Based on my research, it seems like perhaps we can just not delete that directory. I haven't tried this myself yet. @ibuildthecloud does that sound right?
@ajbouh please let me know if you have any success. I am interested in this as well.
@ajbouh @liyimeng On the master branch the commits are very surgical, removing features one by one. You could try reverting the commit removing GPU support (I really don't know which one it is because I believe it might fall under "DeviceManager," not sure).
When k8s 1.13 is released and I do the rebase for the 1.13 series I can evaluate keeping GPU support.
What's on master now should support GPU to my best knowledge. Let me know if it doesn't. Thanks!
Thanks for the great work!
Most helpful comment
@ajbouh @liyimeng On the master branch the commits are very surgical, removing features one by one. You could try reverting the commit removing GPU support (I really don't know which one it is because I believe it might fall under "DeviceManager," not sure).
When k8s 1.13 is released and I do the rebase for the 1.13 series I can evaluate keeping GPU support.