MLflow MLTF Gateway

PLEASE NOTE This is a very early alpha. If you hit an error or see missing functionality, please make an issue so we can track and improve.

This project is a remote MLFlow Project job submission service. Using the included CLI or via REST, users can submit their training code to ACCRE (Vanderbilt's reearch computing facility).

Install

Until releases are pushed to PyPi, one can install the client by executing the following

# If not already in a virtual environment
python3 -m venv venv
source venv/bin/activate

# Once within the virtual environment, install the client
# This only needs to be done once
pip install mltf-gateway

After installing, configure the client with

export MLTF_GATEWAY_URI=https://gw.mltf.vu
mltf login

This will prompt you to visit a webpage, login using either CERN or Vanderbilt credentials, and copy a code from your terminal into the resulting page. You can verify things worked correctly with mltf auth-status

Note: Auth endpoints are now resolved automatically from the gateway via OIDC discovery. The only required setting is MLTF_GATEWAY_URI. Optional environment variable overrides remain available: MLTF_CLIENT_ID, MLTF_AUTH_URL, MLTF_TOKEN_URL, MLTF_SCOPES.

$ mltf auth-status
Credentials found, expired? False
Token Subject: e0574079-d98b-4cc3-9246-1aef09bc0107
Token Issuer: https://keycloak.k8s.accre.vanderbilt.edu/realms/mltf-dual-login
Access Token:
    Issued: 2025-10-06 22:23:05+00:00
   Expires: 2025-10-11 22:23:05+00:00
 Remaining: 1 day, 2:43:28.941754
Refresh Token:
    Issued: 2025-10-01 21:15:11+00:00
   Expires: 2025-10-26 20:56:06+00:00
 Remaining: 16 days, 1:16:29.941754

Note: Auth endpoints are now resolved automatically from the gateway. MLTF_GATEWAY_URI is the only required setting; MLTF_CLIENT_ID, MLTF_AUTH_URL, MLTF_TOKEN_URL, and MLTF_SCOPES are optional overrides.

Quickstart

To demonstrate the gateway, we set up an example "Hello World" project in the demo/with-gpu subdirectory of this repository. To submit a job, enter the demo/with-gpu subdirectory and execute mltf submit which should output something similar to

$ mltf submit --name "hello-world"
Find your MLFlow run at https://mlflow-test.mltf.k8s.accre.vanderbilt.edu/#/experiments/0/runs/1d0c653826144357aa90a7de2c6f6bf8
Submitted project to MLTF: 962e168e-a61c-11f0-b4b0-bc2411853964

You can list your tasks with mltf list. Tasks given a name with --name display it alongside their ID.

$ mltf list
Tasks:
  2025-10-10@16:03:35 - 962e168e-a61c-11f0-b4b0-bc2411853964 (hello-world)

And check their status with mltf show <task_id>. If wanted, logs can be viewed with --show-logs.

$ mltf show 962e168e-a61c-11f0-b4b0-bc2411853964
Status: RUNNING

Finally, any output artifacts, parameters or logs will be uploaded to the tracking server which can be accessed from the URL provided above (future improvements will add CLI access to artifacts). The tracking API is described here and will let you upload arbitrary metrics (e.g. loss) and artifacts (e.g. output files)