we should pin model revisions on Hugging Face

[charlesfrye]

models can still be yanked, but this should reduce the variability

Here's an example, which uses tgi:

def download_model():
    subprocess.run(
        [
            "text-generation-server",
            "download-weights",
            MODEL_ID,
            "--revision",
            REVISION,
        ],
        env={
            **os.environ,
            "HUGGING_FACE_HUB_TOKEN": os.environ["HF_TOKEN"],
        },
        check=True,
    )

[charlesfrye]

and once we've got more examples that do this, it should go in the PR template