Swift TensorFlow with CUDA Docker environment

Image details

• Ubuntu 18.04 Bionic
• CUDA 10.0
• Swift 5.0
• TensorFlow for Swift v0.4

Required

• Follow nvidia-docker instructions

Option 1. Build Docker Image

• From within the directory of the Dockerfile execute the following command to build the image

docker build -t swift_tf_cuda_docker .

• To run with bash and a shared volume

docker run --runtime=nvidia -v <host_path>:<container_path> -ti swift_tf_cuda_docker:latest bash

Option 2. Pull image from Docker Hub

• Go to the repository page bryankp/swift_tf_cuda
• Pull the latest image

docker pull bryankp/swift_tf_cuda:latest

• To run with bash and a shared volume

docker run --runtime=nvidia -v <host_path>:<container_path> -ti bryankp/swift_tf_cuda_docker:latest bash

Example Programs

• Official Tensorflow swift-models

Further Details

• TensorFlow for Swift
• The initial run of a TF program will take a decent amount time to start this is normal
• For troubleshooting issues see the TF installation guide and nvidia-docker instructions