I'm trying to debug using docker interpreter but I keep facing GPU support problems. When running nvidia-docker (from terminal) tensorflow recognizes the GPU devices as expected but it is not the case when I'm using the same image as a docker interpreter.
From what I read at the net passing --runtime=nvidia might solve this issue. I didn't see where it is possible to pass arguments calling docker as part of the configuration file.
Is there any way to solve this issue? Is there any plan to integrate nvidia-docker as part of remote debugging option?