Connect NeMo Microservices#

Model fine-tuning uses a separately operated NeMo Microservices deployment. These endpoints are optional for the base POC Factory installation and AI agent generation.

If your deployment has no endpoints, start with the NVIDIA NeMo Microservices collection and its installation and administration documentation. The collection states that NeMo Microservices was sunset on October 1, 2026, with no new releases. Confirm a compatible deployment with your platform owner before enabling this integration.

Prerequisites#

Ask the platform owner for these inputs before configuring POC Factory.

  1. Platform API, NIM Proxy, and Data Store base URLs reachable from the POC Factory backend, with any required platform authorization or routing configuration.

  2. A base model available through NIM and a matching Customizer configuration, plus an available Evaluator service.

  3. A valid POC Factory inference profile and persistent, writable fine-tuning artifact and dataset storage.

Connect the Endpoints#

Choose deployment-managed endpoints for a shared service, or permit users to supply them for their workflows.

Environment variable

Helm value

NEMO_PLATFORM_URL

config.nemo.platformUrl

NEMO_NIM_URL

config.nemo.nimUrl

NEMO_DATASTORE_URL

config.nemo.datastoreUrl

ENABLE_USER_FINETUNE_ENDPOINTS_OVERRIDES

config.overrides.enableUserFinetuneEndpointsOverrides

  1. Set the three base URLs through your existing Compose or Helm configuration. Use service base URLs rather than an operation URL such as /v1/models.

  2. Set the endpoint override policy to false for deployment-managed endpoints, or true when users may supply alternatives. This is separate from inference-credential overrides.

  3. Apply the configuration through the deployment’s normal rollout or GitOps process.

  4. Open Create → Model Finetuning → Dataset & Training. At the training-target step, validate the configured endpoints or save user-provided values when permitted. Confirm that a fine-tuneable base model is discovered before starting training.

Next Steps#

Follow Fine-tune a model for the POC Factory workflow, or Troubleshooting if endpoint validation or model discovery fails.