triton-inference-server/server ? reverse-engineered prompt

Build me a local Triton Inference Server setup that I can run with Docker and test right away.

I want to be able to start the server, point it at a model repository, load a sample model, and send an inference request from a simple client so I can see the result end to end. It should support the common flow from the README, including running on GPU if available and CPU only if not, and it should be easy to swap in different models later.

Please make the quickstart experience smooth, with clear example commands, working sample models, and enough documentation so someone new can get from clone to first prediction without guessing. If anything depends on current Triton docs or container tags, look that up online and keep it up to date.