Deployment

Taking docTR to production means making it fast, packaging it, and serving it. This section covers each step.

Performance

GPU and Apple Silicon, half precision, torch.compile and other speed-ups.

Performance

ONNX & OnnxTR

Export models to ONNX and run them without PyTorch.

ONNX & OnnxTR

Docker

Prebuilt GPU-ready images, tags, and building your own.

Docker

Web API & demo app

A FastAPI template to serve docTR, and a Streamlit demo.

Web API & demo app

Configuration

Environment variables, model cache, offline use and AWS Lambda.

Configuration