CUDA 8.0 build is not in Docker registry about docker-torch-rnn HOT 5 OPEN

Thanks for the report, as soon as I have some time I will try to get everything up and running. As of right now the best thing to do is trying to build the images from source (but builds may fail since the Dockerfiles are an year old).

If someone has some time to fix the build scripts I would also accept such pull requests and then I will build and upload the images over to dockerhub.

from docker-torch-rnn.

Thanks for the quick response! I've cloned and done a local build for now. FYI the CUDA8.0 version did successfully build locally on my machine, no errors :) Now to run it and see if it's all happy...

from docker-torch-rnn.

So the procedure would be to clone this repo, cd to CUDA/8.0, and run docker build . -t torch-rnn:cuda8.0, is that correct? I tried that and got the following output:

Sending build context to Docker daemon  3.584kB
Step 1/27 : FROM nvidia/cuda:8.0
8.0: Pulling from nvidia/cuda
bc391f6cea37: Pull complete 
Digest: sha256:9aabe6ec8965db35d38b7250892f8421c6a7bff9bbd305d2fbf44dc491b730f0
Status: Downloaded newer image for nvidia/cuda:8.0
 ---> a60a3c83aeed
Step 2/27 : MAINTAINER Cristian Baldi "[email protected]"
 ---> Running in d739abcd138c
Removing intermediate container d739abcd138c
 ---> d9565476bca6
Step 3/27 : ENV DEBIAN_FRONTEND noninteractive
 ---> Running in 42cfe88b50d1
Removing intermediate container 42cfe88b50d1
 ---> b277fe876e6d
Step 4/27 : ENV DEBCONF_NONINTERACTIVE_SEEN true
 ---> Running in 59891504d98a
Removing intermediate container 59891504d98a
 ---> 56565074c282
Step 5/27 : RUN apt-get update
 ---> Running in 65ac726097c3
OCI runtime create failed: container_linux.go:296: starting container process caused "exec: \"/bin/sh\": stat /bin/
sh: no such file or directory": unknown

I have zero prior experience with Docker, so it's likely that I'm being dumb about something. I'm on Ubuntu 16.04. I have CUDA 8.0 and nvidia-docker installed correctly, as far as I can tell (docker run --runtime=nvidia --rm nvidia/cuda nvidia-smi seems to work).

from docker-torch-rnn.

One of the problems I ran into when building the CUDA 8 version locally was that the Dockerfile in the CUDA/8.0/ directory references an image that doesn't exist on DockerHub.

The CUDA 8 files are either nvidia/cuda:8.0-runtime or nvidia/cuda:8.0-devel. They are listed here .

@adelespinasse maybe this can get you in the right direction if you're still looking.

from docker-torch-rnn.

Also it appears as though the ubuntu images that the nvidia ones derive from have removed 'sudo' so there's some finagling to get around that as well.

from docker-torch-rnn.