Update Provider Deployment
Use /app/form/updateProviderDevelopment to edit an existing provider
Deployment. Open it from Administration > GenAI Admin > LLM Models >
Deployments by choosing the row’s edit action.
The route name is updateProviderDevelopment, but the form updates a provider
Deployment through updateLlmProviderDeployment.
Read-Only Fields
| Field | Example | Description |
|---|---|---|
| Host Id | selected-host-id | Host that owns the Deployment. |
| Provider Deployment Id | 7ee18d9d-... | Stable Deployment identifier referenced by Credentials, Pricing, and Alias Routes. |
| Aggregate Version | 4 | Current optimistic-concurrency version. A stale version is rejected. |
Do not remove or alter these values. The active field is backend-managed and
is not included in the form.
Editable Fields
| Field | Example | Description |
|---|---|---|
| LLM Registration | dev — groq / llama-3.3-70b-versatile | Host-scoped approval; its label combines environment, provider, and physical model. |
| Provider Account | OpenAI Production | Host-scoped billing and quota owner. |
| Deployment Name | openai-gpt4o-ca-prod | Unique operator-friendly name within the host. |
| Provider Type | groq | Provider identity. Changing it reloads Physical Model Id options. |
| Physical Model Id | gpt-4o | Exact upstream model served by the endpoint. |
| Provider Endpoint | nvidia-free-embeddings | Reusable transport/authentication profile that owns Protocol and Base URL. |
| Deployment Revision Id | nvidia-free-embedding-demo/r1 | Operator revision of this exact callable configuration. |
| Physical Runtime Id | nvidia/integrate-api/free-embeddings | Stable external service/process identity. |
| Capacity Domain Id | nvidia-free-embedding-demo | Runtime capacity domain; protected lanes must not share one. |
| Runtime Capacity | {"maxParallelRequests":32,...} | Required positive parallel, queue, cold-start, stream-setup, and request timeout bounds. |
| Readiness Policy | IMMEDIATE | IMMEDIATE or WARM_BEFORE_ELIGIBLE. |
| Expected Sidecar Identity | Empty | Profile/digest only for a managed sidecar; leave empty for a native hosted Endpoint. |
| Region | ca-central-1 | Optional placement or residency region. Leave it empty for a global endpoint. |
| Transport Bounds | {"requestTimeoutMs":60000} | Optional non-secret transport metadata object. |
Registration and Account selectors list non-deleted labels under the selected host.
Provider Type comes from model_provider; Physical Model Id comes from the
provider-to-model reference relation; Region comes from the host’s region
reference data.
The form does not edit quotaGroupId. The selected Provider Account owns that
value, and Portal derives it through the existing Account relationship. Changing
the Account changes the value used by the next publication, but it does not
rewrite an already published gateway snapshot.
Structured Fields
Runtime Capacity, Expected Sidecar Identity, and Transport Bounds support JSON and YAML. Choose Apply after editing. For example:
{
"connectTimeoutMs": 5000,
"requestTimeoutMs": 60000
}
Transport-bound properties remain control-plane annotations unless the publication contract explicitly maps them to supported gateway settings.
Base URL and Provider Protocol are edited only on the Provider Endpoints tab.
Deployment updates reference that record through providerEndpointId.
Identity Changes
The provider type, provider protocol, physical model, and endpoint form the callable identity. If the provider endpoint or physical model changes, publish the updated configuration and test connectivity through the tenant gateway. The selected Account’s provider type must match the Deployment.
An update does not bypass Credential, Pricing, Alias Route, or publication requirements. Publish performs the final cross-record review.
For the hosted NVIDIA Deployment, preserve Provider Type nvidia, Physical
Model Id nvidia/nemotron-3-embed-1b, and Endpoint nvidia-free-embeddings.
On the Provider Endpoints tab, that endpoint owns Protocol openai_embeddings
and Base URL https://integrate.api.nvidia.com/v1. Rotate env:NVIDIA_API_KEY
through Credentials.
Save The Update
Choose Update Provider Deployment. The form sends
lightapi.net/genai/updateLlmProviderDeployment/0.1.0 with
providerDeploymentId and aggregateVersion, then returns to
Administration > GenAI Admin > LLM Models after success.
An update changes the Portal control-plane record; it does not rewrite a previously published gateway snapshot. Publish the intended new configuration and test it through the tenant gateway before expecting supported runtime behavior.
Common Problems
- Stale aggregate version: reopen the form from the Deployments tab and apply the change to the latest record.
- Registration or Account is unavailable: confirm it is not deleted and belongs to the selected host.
- Structured edit is blocked: correct the Transport Bounds JSON/YAML draft and choose Apply, or choose Reset to restore the last valid value.
- Provider mismatch: select an Account whose provider type matches the Deployment.
- 403 on Update: confirm access to
lightapi.net/genai/updateLlmProviderDeployment/0.1.0and the required write permission.
For route eligibility and gateway consumption, see the Deployments tab guide.