Ucas-HaoranWei/GOT-OCR2.0 ? reverse-engineered prompt

Build me a working OCR app from this repo that can take an image and extract the text from it using the GOT OCR 2.0 model.

I want it to handle a few simple modes, plain text OCR, formatted OCR that keeps the layout a bit better, and a focused mode where I can give a box or color to read just part of the image. It should also support the multi crop mode for harder images, and let me render the extracted result into a simple HTML page so I can review it in the browser.

Make the setup easy to run on a CUDA machine, include clear install steps, and use the official model weights from Hugging Face if needed. If anything needs current docs or updated install advice, look it up online. Please make the demo scripts work cleanly end to end with a sample image and sensible defaults.