Skip to main content
Last updated on

AI Runtime

AI Runtime is a fully managed, serverless GPU compute offering on Databricks for training, fine-tuning, and other deep learning workloads. It is for ML engineers and data scientists who want GPUs without provisioning clusters. For classic compute with Databricks Runtime ML, see Databricks Runtime for Machine Learning.

  • Serverless GPUs: On-demand A10s and H100s, accessible through the Databricks CLI, notebooks, jobs, and Ray.
  • Managed environments: pre-built environments and dataloaders ready for training and fine-tuning, or bring your own Docker container.
  • Built-in observability: experiment tracking via MLflow, logging, and debugging agents.
  • Orchestration: schedule your GPU workloads and use them in CI/CD pipelines.

Get started​

Learn about AI Runtime and run your first workload.

    • 🚀 Quickstart
    • Run your first workload on AI Runtime from your terminal.
    • 🧭 Overview
    • Learn the key things about AI Runtime: what GPUs are available, how it works, and limitations.
    • 💡 Examples
    • Browse notebook tutorials and runnable Databricks CLI recipes organized by task.

Connect to GPUs​

Access GPUs through the Databricks CLI, notebooks, Jobs, or via Ray.

    • Databricks CLI
    • Submit and manage distributed GPU training jobs from your laptop using YAML job configs.
    • Notebooks
    • Attach a notebook to AI Runtime and develop interactively, with no cluster setup.
    • Jobs
    • Schedule a notebook on AI Runtime as a recurring job.
    • Ray
    • Run Ray Core, Ray Data, Ray Train, and Ray Tune on AI Runtime.

Manage environments​

Run workloads in a managed environment or in your own Docker image.

    • Pre-built environments
    • Start from a minimal Standard environment or a Databricks AI environment pre-loaded with ML frameworks.
    • Custom Docker images
    • Bring your own Docker image when your workload requires custom system libraries or dependencies.

Develop workloads​

Load data efficiently and develop workloads with coding agents.

    • Load data into GPUs
    • Stream data from Unity Catalog volumes with fault-tolerant loaders built for high GPU utilization.
    • Coding agents
    • Use Genie Code and the databricks-ai-runtime agent skill to create, submit, and manage GPU workloads.

Monitor, debug, and deploy​

Track experiments, inspect logs, diagnose failures, and schedule production workloads.

    • View logs
    • View training output, monitor GPU resource usage while your code runs, and debug with Genie Code.
    • Schedule workloads
    • Deploy training code with Declarative Automation Bundles, schedule runs, and build multi-task GPU and CPU workflows.

Guides​

Explore migration guides, tutorials and technical resources for AI Runtime.

On this page