Tencent-Hunyuan/Hy3 — reverse-engineered prompt

Build me a practical setup for running Tencent Hy3 as a self hosted chat and reasoning model. I want a simple way to download either the main model or the FP8 version, start a server on a machine with lots of GPU memory, and expose an OpenAI compatible API so I can plug existing apps into it.

Please use the serving approach the README recommends, and make sure I can switch between quick answers and deeper reasoning for coding, math, and long context tasks. Add a tiny example client that sends a chat request, plus clear commands for starting the service and testing that it works.

If this repo already includes finetuning support, keep that usable and explain the basic flow in plain English, but the main goal is reliable deployment and easy local or server testing. I would also like friendly docs that mention hardware expectations, recommended model settings, and when to choose the FP8 model. Look up current docs online if you need to.