API Reference#

The TAO API provides comprehensive OpenAPI specifications for all versions. The latest v2 API introduces a unified, job-centric architecture that simplifies interactions with TAO Toolkit.

TAO API v2 (Current)#

Documentation is available for the following TAO API v2 versions:

The TAO API v2 is the latest version, featuring:

  • Unified Jobs API - Single endpoint for experiment and dataset operations

  • Environment Variable Authentication - JWT token-based authentication with environment variable support

  • Resource-Specific Metadata - Dedicated endpoints for workspaces, datasets, and jobs

  • Enhanced Job Control - Comprehensive pause, resume, cancel, and delete operations

  • Inference Microservices - Deploy trained models as scalable inference endpoints

Access v2 API Documentation:

After deploying Fine-Tuning Micro-Services, access the API documentation at:

  • Swagger UI: /api/v2/swagger

  • ReDoc: /api/v2/redoc

  • OpenAPI Specs: /api/v2/openapi.json or /api/v2/openapi.yaml

  • Example Notebooks: /api/v2/tao_api_notebooks.zip

v2 API Key Endpoints:

Authentication#

  • POST /api/v2/login - Authenticate with NGC credentials and obtain JWT token

Workspaces#

  • GET /api/v2/orgs/{org_name}/workspaces - List workspaces

  • POST /api/v2/orgs/{org_name}/workspaces - Create workspace

  • GET /api/v2/orgs/{org_name}/workspaces/{workspace_id} - Get workspace metadata

  • DELETE /api/v2/orgs/{org_name}/workspaces/{workspace_id} - Delete workspace

  • POST /api/v2/orgs/{org_name}/workspaces/{workspace_id}:backup - Backup workspace

  • POST /api/v2/orgs/{org_name}/workspaces/{workspace_id}:restore - Restore workspace

Datasets#

  • GET /api/v2/orgs/{org_name}/datasets - List datasets

  • POST /api/v2/orgs/{org_name}/datasets - Create dataset

  • GET /api/v2/orgs/{org_name}/datasets/{dataset_id} - Get dataset metadata

  • DELETE /api/v2/orgs/{org_name}/datasets/{dataset_id} - Delete dataset

  • GET /api/v2/orgs/{org_name}/datasets:get_formats - Get supported dataset formats

Jobs (Unified Endpoint)#

  • GET /api/v2/orgs/{org_name}/jobs - List all jobs

  • POST /api/v2/orgs/{org_name}/jobs - Create job (experiment or dataset)

  • GET /api/v2/orgs/{org_name}/jobs/{job_id} - Get job metadata

  • DELETE /api/v2/orgs/{org_name}/jobs/{job_id} - Delete job

  • POST /api/v2/orgs/{org_name}/jobs/{job_id}:pause - Pause job

  • POST /api/v2/orgs/{org_name}/jobs/{job_id}:resume - Resume job

  • POST /api/v2/orgs/{org_name}/jobs/{job_id}:cancel - Cancel job

  • GET /api/v2/orgs/{org_name}/jobs/{job_id}:logs - Get job logs

  • POST /api/v2/orgs/{org_name}/jobs/{job_id}:list_files - List job files

  • POST /api/v2/orgs/{org_name}/jobs/{job_id}:download_selective_files - Download specific files

  • GET /api/v2/orgs/{org_name}/jobs/{job_id}:download - Download entire job

  • POST /api/v2/orgs/{org_name}/jobs/{job_id}:publish_model - Publish model to NGC

  • POST /api/v2/orgs/{org_name}/jobs/{job_id}:remove_published_model - Remove published model

  • GET /api/v2/orgs/{org_name}/jobs:list_base_experiments - List base experiments

  • GET /api/v2/orgs/{org_name}/jobs:schema - Get job schema for action

  • GET /api/v2/orgs/{org_name}/jobs:gpu_types - Get available GPU types

Inference Microservices#

  • POST /api/v2/orgs/{org_name}/inference_microservices:start - Start inference microservice

  • GET /api/v2/orgs/{org_name}/inference_microservices/{job_id}:status - Get microservice status

  • POST /api/v2/orgs/{org_name}/inference_microservices/{job_id}:inference - Make inference request

  • POST /api/v2/orgs/{org_name}/inference_microservices/{job_id}:stop - Stop microservice

AutoML#

  • GET /api/v2/orgs/{org_name}/automl:get_param_details - Get AutoML parameter details

Python SDK and CLI#

For programmatic and command-line access to the v2 API, use the TAO Python SDK and CLI:

pip install nvidia-tao

See Remote Client for comprehensive SDK and CLI documentation.

TAO API v1 (Legacy)#

Documentation is available for the following TAO API v1 versions:

Note

TAO API v1 is in maintenance mode. We recommend migrating to v2 for new projects to take advantage of the unified job-centric architecture and enhanced features.

Migration from v1 to v2#

If you’re using TAO API v1, consider migrating to v2 for these benefits:

Key Improvements

  • Simplified Architecture: Single unified jobs endpoint instead of separate experiment and dataset endpoints

  • Better Authentication: Environment variable support for easier CI/CD integration

  • Enhanced Job Control: Comprehensive pause, resume, cancel, and delete operations

  • Resource Management: Full CRUD operations on all resources

  • Inference Microservices: Deploy models as scalable inference endpoints

Breaking Changes

  • Endpoint structure changed from /api/v1/ to /api/v2/

  • Unified /jobs endpoint replaces separate /experiments and dataset action endpoints

  • Job creation now single-step instead of two-step (create + run action)

  • Authentication uses JWT tokens with environment variables instead of file-based config

  • Metadata access uses resource-specific endpoints

Migration Resources