OpenMOSS/MOSS ? reverse-engineered prompt

Build me a working local demo of MOSS, the open source Chinese and English chat model that can also call tools like search, calculator, equation solving, and text to image when available.

I want to be able to run it from my own machine, chat with it in a terminal, and also open a simple web page for conversation. Please wire up the basic model loading, streaming replies, and the prompt setup so it behaves like a helpful assistant. If the tool use pieces need current setup details from the web or docs, look them up and use the right approach.

Also make sure there is a clean path for fine tuning on the included dialogue data, and that the code is easy to follow so I can swap in different model sizes or quantized versions later. If anything about the model files or dependencies needs extra configuration, handle that too and leave clear instructions in the app.