This repo gives a minimal implementation of testing 1 example of the examples repo.

• The dockerfile has a sort-of minimal pytorch image

Build with: docker build . -t "minimal"

docker run -it --rm --gpus all -v $PWD:/wandb minimal

inside the docker container, you should log to wandb:

wandb login

this should be automated in a resonable way preserving credentials

then, you can run the execution of the testing with the nb_helpers.run_nbs

nb_helpers.run_nbs --path  /wandb/Intro_to_Weights_\&_Biases.ipynb

after executing this command, you should find a run.csv file with basic reporting of the execution.

• Find a better way to log to wandb, and override project name to something like examples_testing
• Orchestrate different images with different notebooks types (tensorflow, pytorch, sklearn, etc...)
• Automate this testing on a machine triggered from a github action