Deploying ClearML as an Azure ML Alternative

Description and use

ClearML replaces Azure ML with a self‑hosted MLOps platform that runs on any infrastructure, supports GPU workloads, and provides experiment tracking, pipelines, hyper‑parameter tuning, and Triton model serving.

Problem

Azure Machine Learning ties experiment tracking, pipelines, and model serving to Azure-specific APIs and managed‑compute pricing, limiting flexibility and cost control.

Outcome

ClearML provides a fully self‑hosted alternative to Azure ML, enabling experiment tracking, pipeline orchestration, hyper‑parameter tuning, and model serving with Triton on any Linux infrastructure, while eliminating Azure‑specific costs and dependencies.

Workflow

• Set up Linux server with Docker, Docker Compose, and DNS records for subdomains
• Configure system limits and create directories for ClearML data
• Download and edit docker‑compose.yml to use Traefik routing and bridge networks
• Create .env with ClearML host URLs
• Start ClearML services with docker compose
• Set up Traefik with TLS via Let’s Encrypt and dynamic routing
• Create admin account and generate credentials in ClearML UI
• Deploy an agent on the same or a separate machine, installing clearml‑agent and initializing with credentials
• Start the agent daemon, optionally specifying GPU queues
• Install ClearML SDK in a Python virtual environment
• Use the SDK to track experiments, build pipelines, run hyper‑parameter optimization, and serve models with Triton

Evidence

ClearML Server with Docker Compose and Traefik, configures agents for remote execution, tracks an experiment, builds a pipeline, runs hyperparameter optimization, and serves a model with Triton.

Source