Skip to content

PySDK

Overview

The DeGirum Python Software Development Kit (PySDK) is a Python package designed to provide APIs for performing machine learning (ML) model inferences. PySDK is built around three core concepts:

  1. AI Inference Engines: These are responsible for executing inferences on AI models.
  2. AI Model Zoos: Repositories where AI models are stored and accessed.
  3. AI Models: The specific models used for inference.

The interaction between these components enables seamless integration of AI-powered functionalities into your applications.

AI inference engines in PySDK power different types of AI inference, which are categorized based on the location of the inference engines, providing flexibility in deployment scenarios to meet various needs.

AI Inference Types

PySDK supports three types of AI inference, catering to different deployment scenarios:

  1. Local Inference:

    • The client application communicates directly with an AI hardware accelerator installed on the same machine.

  2. AI Server Inference:

    • The AI hardware accelerator is managed by the DeGirum AI Server software stack.
    • The client application communicates with the AI Server, which may reside on the same machine or on a different one connected via a local network.

  3. Remote Inference:

    • The client application connects to the DeGirum AI Hub over the Internet to perform inferences on devices hosted in the DeGirum Device Farm.

AI Model Zoo Types

PySDK provides access to two types of AI model zoos:

  1. Local Model Zoo:

    • A collection of AI models stored in a local directory on the machine with the AI hardware accelerator.
    • For local inference, the model zoo resides on the same machine as the application.
    • For AI server inference, the model zoo is located on the machine running the AI Server.

  2. AI Hub Model Zoo:

    • A collection of AI models hosted on the DeGirum AI Hub.
    • Users can manage model zoos via the DeGirum AI Hub web interface.
    • Two categories of model zoos:
      • Public Model Zoo: Accessible to all registered AI Hub users.
      • Private Model Zoo: Restricted to members of a specific organization.

This modular structure of inference engines and model zoos ensures flexibility and scalability for various AI deployment needs.