The outcome

Keep GPU inference observable, cost-bounded, recoverable, and isolated across workspaces and environments.

DGX Cloud Lepton is a NVIDIA AI application and inference platform for Python Photons, containers, endpoints, jobs, and GPU resources. Turning Python model code into a Photon service, deploying a Photon or container as a GPU endpoint, and operating bounded AI jobs and inference workloads in a managed workspace. This guide narrows that broad capability into one repeatable outcome, with checkpoints that keep the source material and your judgment in the loop.

Before you begin

Set the boundary before the tool starts.

Choose one real task, identify who will use the result, and decide what evidence or test will make the result acceptable. Gather only the source material needed for that task. If the work contains confidential, personal, regulated, or client-owned information, confirm that the platform and account are approved before sharing it.

Troiana principle

AI should make the work easier to inspect. If the workflow removes the source, the owner, or the review step, redesign the workflow.

Step by step

A workflow you can repeat.

  1. 01

    Separate workspaces and credentials by environment, assign least privilege, and define access, data handling, retention, SLOs, replica and resource ceilings, budgets, and incident ownership.

  2. 02

    Create an endpoint from a reviewed Photon ID or pinned container, attach scoped secrets, keep access private, expose one required port, and configure startup allowance and replica bounds.

  3. 03

    Monitor status, events, logs, replicas, GPU utilization, queue, latency, errors, output parity, storage, and spend using redacted metadata and actionable alerts.

  4. 04

    Test bad artifacts, slow startup, memory exhaustion, malformed requests, scale limits, provider failure, secret rotation, access revocation, rollback, stop, and removal.

  5. 05

    Canary an endpoint update instead of rerunning production in place, retain the prior revision, audit access and cost, remove retired resources, and verify billing cessation.

Working standard

What good use looks like.

  • Use immutable Photon or image versions.
  • Update production with a canary.
  • Verify endpoint removal stops billing.

Older material may call endpoints deployments and the product Lepton AI; current documentation uses NVIDIA DGX Cloud Lepton. A Photon simplifies packaging but does not remove responsibility for unsafe handlers, arbitrary code, dependency and model licenses, public exposure, secrets, logs, autoscaling, or GPU cost. Pin versions, use private access, isolate secrets, cap resources, and update with rollback.

Official references

Check the current product documentation.

Features, plan limits, availability, and data controls change. These official pages are the starting points used for this collection.