xai-org/grok-1 ? reverse-engineered prompt

I want to set up a project to run the open-source Grok-1 model from xAI. It's that huge 314B parameter Mixture-of-Experts model.

Can you build me a simple Python project using JAX that can load the model weights and run a basic inference test? The main goal is to have a script that I can run from the command line which loads the model and generates a sample completion from a test input, just to verify that everything is working.

I'll handle downloading the massive checkpoint file myself, so just make sure to create a checkpoints directory where I can place it. The project should include all the necessary Python files to define the model architecture and a requirements.txt file so I can easily install all the dependencies.