Pytorch Serving

Overview

PyTorch Serving is a robust and convenient starter pack designed specifically for deploying lightweight responsive web applications that utilize fast.ai PyTorch models. Built on the Starlette framework with the Uvicorn ASGI server, this toolkit simplifies the process of model serving, ensuring you can swiftly transition your machine learning models into production. A great example of its capabilities can be seen in the demo application, PlantDoc, which serves as a digital plant doctor, effectively diagnosing plant diseases using image classification.

The essence of this starter pack lies in its seamless integration with various tools and services, making deployment straightforward and efficient. With the ability to set up your environment quickly, this solution is ideal for developers looking to leverage their trained models in a practical, user-friendly application.

Features

• Lightweight Framework: Built on Starlette, it provides an efficient collection of tools specifically for creating ASGI services.

• Model Serving: Integrates with Uvicorn ASGI server for responsive and high-performance model serving, ensuring quick inference times.

• Docker Integration: Easily deploy your trained models using Zeit Now, which handles Docker containerization, simplifying the deployment process.

• Starter App: The included PlantDoc app demonstrates how to deploy a plant disease classification model, giving users a hands-on reference.

• One-Time Setup: The setup process is streamlined, requiring only the installation of Node.js and the Now CLI for initial configuration.

• Customization Options: Users can easily customize the app for their specific model by modifying the modelDefinition.json file to include their model's parameters and classes.

• Unique Deployment URLs: Each deployment generates a unique URL, allowing for quick access and testing of your application.

• Local Testing Capability: Supports local testing of the app server, ensuring that users can validate their deployments before going live.