https://github.com/andyneff/tensorflow_build_docker

Simple docker to build tensorflow
https://github.com/andyneff/tensorflow_build_docker

Last synced: 4 months ago
JSON representation

Simple docker to build tensorflow

• Host: GitHub
• URL: https://github.com/andyneff/tensorflow_build_docker
• Owner: andyneff
• License: mit
• Created: 2018-05-30T15:26:21.000Z (about 8 years ago)
• Default Branch: master
• Last Pushed: 2018-05-30T15:27:34.000Z (about 8 years ago)
• Last Synced: 2025-01-15T19:42:59.350Z (over 1 year ago)
• Size: 2.93 KB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          

Builds tensorflow wheel using CUDA and CUDNN for you

Based roughly on the tensorflow docker, but works.

## Usage

    docker build -t tf .

    docker run -it --rm -v `pwd`:/tmp/wheel tf

## Docker Args

- `CUDA_VERSION_2` - Version of CUDA, defaults to `9.0`

- `CUDNN_VERSION_2` - Version of CUDNN to use, defaults to `7.0.5.15`

- `BAZEL_VERSION` - Version of Bazel to use, defaults to `0.13.1`

## Docker ENV vars

- Many used in configure.py, not covered here

    - `TF_CUDA_COMPUTE_CAPABILITIES` - Defaults to `5.2`, you'll probably want

it to include whatever card capabilities you are using.

- `TENSORFLOW_VERSION` - tag/SHA/branch to checkout. Using branch name is not

recommended because the underlying SHA can change, but do whatever you want.

Defaults to `v1.8.0`