Skip to main content
country_code
Ctrl+K
NVIDIA NIM Metadata API - Home NVIDIA NIM Metadata API - Home

NVIDIA NIM Metadata API

NVIDIA NIM Metadata API - Home NVIDIA NIM Metadata API - Home

NVIDIA NIM Metadata API

Table of Contents

About

  • Overview
  • Architecture Overview
  • Ecosystem Overview
  • Core Concepts Overview for NIM Metadata API
  • Release Notes for NIM Metadata API

Get Started

  • About Getting Started with NIM Metadata API
  • NIM Metadata API Prerequisites
  • Quickstart for NIM Metadata API

Guides

  • Choose a Profile for Your Hardware
  • Notes for Automated Agents

Reference

  • Reference Overview for NIM Metadata API
  • Endpoints for NIM Metadata API
  • Response Fields for NIM Metadata API
  • Related Endpoints and Registry Search
  • Current Limitations of the NIM Metadata API

Troubleshooting

  • Troubleshooting NIM Metadata API

Resources

  • Resources Overview for NIM Metadata API
  • Glossary for NIM Metadata API
  • Related Endpoints and Registry Search
Is this page helpful?

Related Endpoints and Registry Search#

The NIM Metadata API returns tag-level data only. The endpoints on this page return complementary information, such as details that describe a NIM microservice as a whole and the human-readable name behind a GPU PCI identifier. For the metadata endpoints themselves, refer to Endpoints.

Container-Level Information#

Some fields describe the NIM microservice as a whole rather than one specific tag, including id, name, nim_type, vertical, nspect_id, supported_models, model_params, and sdk_example_url. These fields live on the repository object rather than in the tag-level metadata document, and the repository response returns them in a nim object.

The tag-level endpoints described in this documentation deliberately return only tag-level data. If you need both, make both calls.

Public NIM Microservices#

For public NIM microservices, use the following forms. These forms are the recommended choice, because they require no credentials:

GET /v2/repos/{orgName}/{repository}
GET /v2/repos/{orgName}/{teamName}/{repository}

The following example requests the repository object for a team-owned NIM microservice:

curl --request GET \
  --url 'https://api.ngc.nvidia.com/v2/repos/nim/meta/llama-3.1-8b-instruct' \
  --header 'accept: application/json'

Private and Entitlement-Gated NIM Microservices#

For private or entitlement-gated NIM microservices, use the organization-scoped forms:

GET /v2/org/{orgName}/repos/{repository}
GET /v2/org/{orgName}/team/{teamName}/repos/{repository}

GPU Identifier Lookup#

To turn a gpu_device PCI identifier into a human-readable GPU name, use the NGC GPU metadata endpoint:

GET /v2/gpus/pci-device-id/{pciDeviceId}

This endpoint accepts the device portion of the identifier only. A gpu_device value has the form pciDeviceId:pciVendorId, so split the value on the colon and send the first segment. For the gpu_device value 2330:10de, request 2330:

curl --request GET \
  --url 'https://api.ngc.nvidia.com/v2/gpus/pci-device-id/2330' \
  --header 'accept: application/json'

The response is an array, because one device identifier can cover several recorded subsystem variants. Read displayName for the human-readable name, such as H100 80GB HBM3. Sending the full gpu_device value returns a 404 status code.

Catalog Search by GPU#

The NGC Catalog supports filtering containers by GPU type in its user interface, which is useful for discovering candidate NIM microservices before you query their metadata individually.

Related Topics#

  • Endpoints

  • Response Fields

  • Choose a Profile for Your Hardware

previous

Response Fields for NIM Metadata API

next

Current Limitations of the NIM Metadata API

On this page
  • Container-Level Information
    • Public NIM Microservices
    • Private and Entitlement-Gated NIM Microservices
  • GPU Identifier Lookup
  • Catalog Search by GPU
  • Related Topics
NVIDIA NVIDIA
Privacy Policy | Your Privacy Choices | Terms of Service | Accessibility | Corporate Policies | Product Security | Contact

Copyright © 2024-2026, NVIDIA Corporation.

Last updated on Oct 05, 2026.