CML with Tensorboard use case
This repository contains a sample project using CML with Tensorboard.dev to track model training in real-time. When a pull request is made, the following steps occur:
- GitHub will deploy a runner machine with a specified CML Docker environment
- A Tensorboard.dev page will be created
- CML will report a link to the Tensorboard as a comment in the pull request
- The runner will execute a workflow to train a ML model (
The key file enabling these actions is
Secrets and environmental variables
In this example,
.github/workflows/cml.yaml contains two environmental variables that are stored as repository secrets.
|GITHUB_TOKEN||This is set by default in every GitHub repository. It does not need to be manually added.|
To access your Tensorboard credentials:
- On your local machine, run
tensorboard dev upload
- Accept the TOS and follow the authentication procedure.
- When you have authenticated, copy your credentials out of
~/.config/tensorboard/credentials/uploader-creds.json(this is the typical path for OSX/Linux systems). Paste these credentials as the secret TB_CREDENTIALS.
Cloning this project
Note that if you clone this project, you will have to configure your own TB credentials for the example.