Installation and setup

For monitoring experiments using TrueFoundry you need to setup mlfoundry:

mlfoundry python library

Step 1: Signup or Login on TrueFoundry

Step 2: Setup MLFoundry

mlfoundry is the python library from TrueFoundry used for experiment tracking.

Install ServiceFoundry client library

On your machine, run (preferably within a virtual environment or conda environment):

pip install mlfoundry

Login from CLI

From the CLI, run the following and follow the link to authenticate:

mlfoundry login --host https://app.example.yourdomain.com

You will be prompted to complete the authentication by clicking on the link.

Output:

Opening:- https://app.truefoundry.com/authorize/device?userCode=XXXX
Please click on the above link if it is not automatically opened in a browser window.
Successfully logged in to 'https://app.truefoundry.com' as 'abc-tfy'
([email protected])

Troubleshooting

How can I log in to mlfoundry on a device in a non-interactive mode?

If you are running automated training jobs, you can use the TFY_HOST and TFY_API_KEY environment variable to set the API Key. You do not need to use the mlfoundry login command if you are using environment variables.

The API key passed via the environment variable takes precedence over the API key passed via the login command.

export TFY_HOST=https://app.example.yourdomain.com  # e.g. https://app.truefoundry.com
export TFY_API_KEY="your-api-key"

Troubleshooting for Mac OS users with M1 processors

Users can face issues installing mlfoundry packages on Mac OS computers with M1 chip. Here are a few links that will help you troubleshoot.

Error in installing numpy. Follow this link to fix the issue.

While installing MLFoundry user might face the following error.

SetuptoolsDeprecationWarning: setuptools.installer is deprecated. Requirements should be satisfied by a PEP 517 installer.

Follow the link to solve the issue.