OpenWeights

Use an existing deployment

Someone runs OpenWeights and gave you access — upload and download with the hf CLI.

You are not running the stack yourself. An operator gave you a deployment URL and an API key. All you need is the standard Hugging Face client — no fork, no patched wheel.

Steps

1. Install the client.

pip install "huggingface_hub[hf_xet]"

2. Point hf at the deployment.

Set two variables: the deployment URL your operator gave you (as HF_ENDPOINT) and your API key (as HF_TOKEN).

export HF_ENDPOINT=https://cas.example-deployment.com
export HF_TOKEN=<your-key>

3. Upload.

hf upload me/my-model ./model-dir

4. Download.

hf download me/my-model --local-dir ./out

The bytes come back byte-for-byte identical. If a download 404s right after an upload, the first pin is still forming on Sia — wait and retry.

A read + write key (the console default) does both upload and download. If your operator gave you a single-scope key, a write key uploads and a read key downloads.

More detail

On this page