The base TensorFlow docker images do not have Cuda configured correctly

[praveen-palanisamy]

The Issue:

import tensorflow fails with ImportError: libcuda.so.1: cannot open shared object file: No such file or directory using openai/retro-agent:tensorflow as the base docker image.

The issue exists in openai/retro-agent:tensorflow-1.4 also. Haven't tested other tags.

[endrift]

You need to use nvidia-docker (--runtime nvidia), at which point it will work properly.

[praveen-palanisamy]

The issue happens when used with retro-contest run <...> locally.

[endrift]

Were you passing -N?

[tristansokol]

I'm having the same issue, trying to get the baseline rainbow agent running

retro-contest run --agent $DOCKER_REGISTRY/rainbow-agent:v1 \
    --results-dir results --no-nv  --use-host-data\
    SonicTheHedgehog-Genesis GreenHillZone.Act1

It seems like you need to build a different image to test locally if you don't have a gpu (like my macbook) but I am not sure how since I am already using tensorflow not tensorflow-gpu

[HanSeoulOh]

Hi @tristansokol , were you able to build different images? New to docker and facing the same issue (macbook doesn't have gpu)

[tristansokol]

@HanSeoulOh, not yet (since I believe that I am still building CPU utilizing images & have tried using some other images that start install tensorflow). My current hypothesis is that tensorflow on my machine just can't find my installed Cuda libraries. I'm kind of lost though.

[endrift]

If you have an Nvidia GPU with nvidia-docker2 installed you should not pass --no-nv.