How NeMo Studio Works#

NVIDIA NeMo Studio operates as a web application that communicates with the underlying NeMo microservices as the backend. It provides a graphical interface layer on top of the NeMo microservices platform. Interact with the system without writing code or using command-line tools.

Architecture Overview#

The following diagram shows the high-level architecture of NeMo Studio.

flowchart TD User(User) Studio("NVIDIA NeMo Studio") NeMo("NVIDIA NeMo Microservices") User -->|Interacts through<br/>Web Browser| Studio Studio -->|REST API Requests| NeMo NeMo -->|Responses| Studio Studio -->|Displays Results| User style User fill:#2196F3,stroke:#000000,color:#fff style Studio fill:#76b900,stroke:#000000,color:#fff style NeMo fill:#76B900,stroke:#000000,color:#fff

NeMo Studio Architecture#

  • NVIDIA NeMo Studio: The frontend application that you can interact with through your web browser. When you interact or submit job requests, Studio sends API requests to the NeMo microservices running on your cluster and renders the responses in the interface.

  • NVIDIA NeMo Microservices: The NVIDIA NeMo microservices platform that runs under the hood to process requests and return responses.

API Communication#

NeMo Studio communicates with the NeMo microservices running on your cluster through the REST APIs exposed by the corresponding microservices.

  • Entity Management: Studio communicates with NeMo Entity Store to create, retrieve, update, and delete entities such as projects, datasets, and models.

  • Data Operations: File uploads and downloads are handled through NeMo Data Store.

  • Customization Jobs: Studio submits fine-tuning jobs to NeMo Customizer and monitors their progress.

  • Evaluation Jobs: Evaluation requests are sent to NeMo Evaluator for processing.

  • Inference: Model testing in the Models page communicates with NeMo Entity Store, NeMo Deployment Management, and NIM Proxy.