> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tuneplane.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Hugging Face

> Link a Hugging Face account so jobs can read gated models and push exports.

## Hugging Face

Linking an account does two things a job cannot otherwise do: read a gated or private model, and
push an export back to the Hub.

**1. Link**

Authorise through Hugging Face OAuth. The token is encrypted at rest with the deployment's
`TUNEPLANE_SECRET_ENC_KEY`.

**2. Submit a job that needs it**

Preflight checks that the model your config names is reachable and that you are authorised for it
— before the job is queued, rather than after it has waited and started downloading.

**3. Push an export**

```bash theme={"theme":{"light":"github-light","dark":"github-dark-dimmed"}}
tp export my-grpo --checkpoint checkpoints/step_500 \
  --checkpoint-format nemo-dcp --push-repo alice/qwen-support-v3
```

> [!WARNING]
>
> Write scopes are off by default (`TUNEPLANE_HF_OAUTH_WRITE_ENABLED`). The safe default is that nothing
> is ever uploaded; an administrator opts the deployment in before `--push-repo` can work.

## On an air-gapped network

There is no Hugging Face to link to, and there does not need to be. An internal mirror is configured
deployment-wide with `TUNEPLANE_HF_ENDPOINT` and every job uses it without anyone linking anything. See
[air-gapped deployments](https://docs.tuneplane.ai/en/ops/airgapped.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.