Collection of Dockerfiles useful for NLP and Deep Learning. To download the docker images visit: floydhub's Docker Hub.
- Dockerfiles are generated using two inputs:
matrix.ymland jinja template file inside./dl/FRAMEWORKdirectory.matrix.ymlprovides variable values for jinja template files.
- The
$renderlist inmatrix.ymlcontrols what version of the framework to render. - For each version config in
matrix.yml, any key starts with_are global keys, which will get automatically injected into each variant config for that version.
-
Most of the cases, you only need to update
./dl/FRAMEWORK/matrix.ymlto generate a set of dockerfiles for a new version of a framework. If not, you will need to update the jinja file to account for build step changes. -
Install floydker:
cd floydker && pipenv shell && pipenv install. -
Render dockerfiles:
cd .. && floydker render .. -
Commit new docker images to git and push:
git commit -a.
Dockerfiles are organized into the following directory structure:
CATEGORY/PROJECT_NAME/VERSION/Dockerfile-ENV
CATEGORY/PROJECT_NAME/VERSION/Dockerfile-ENV.gpu
Automated build scripts will generate the following tags for images based on the above dockerfile paths:
floydhub/PROJECT_NAME:VERSION-ENV
floydhub/PROJECT_NAME:VERSION-ENV-gpu
Contains docker images for popular deep learning frameworks including: Tensorflow, PyTorch and Torch.