# Access gated models (Llama, FLUX) and fix 401 errors

> Accept the model licence, create a read or fine-grained token, log in with hf auth login, set HF_TOKEN for CI, and fix 401/403 errors.

- Canonical: https://guides-ai.com/guides/hugging-face-gated-models-access/
- Plate 16.04 · Topic: Hugging Face (https://guides-ai.com/topics/huggingface/)
- Published: 12 Oct 2026 · 3 min read
- Source site: guides-ai — https://guides-ai.com/

Models like Llama and FLUX are gated: you must accept their licence and authenticate before any download works. Miss either step and every request returns a 401 or 403.

## 1. Accept the licence on the model page

Log in at huggingface.co, open the model, and click **Agree and send request to access repo** (some repos label it **Acknowledge license**). If approval is automatic you get access immediately; if the author reviews manually, you wait. Access is granted per user, never to a whole organization. The flow is described in the official [gated-models guide](https://huggingface.co/docs/hub/models-gated).

## 2. Create a read or fine-grained token

Go to Settings, then **Access Tokens**, then **New token**. A **read** token downloads any repo you can read, including gated ones you have been granted. For production prefer a **fine-grained** token scoped to only the models you need. Token roles are explained in the [security-tokens docs](https://huggingface.co/docs/hub/security-tokens). Never print a token to logs or commit it.

## 3. Log in with hf auth login

```bash
hf auth login
```

Paste the token when prompted; it is stored under `HF_HOME`. Confirm which account it belongs to:

```bash
hf auth whoami
```

Prints your username. This must be the account that holds the licence approval.

## 4. Set HF_TOKEN for scripts and CI

macOS / Linux:

```bash
export HF_TOKEN=hf_xxx
```

Windows (PowerShell):

```powershell
$env:HF_TOKEN = "hf_xxx"
```

`HF_TOKEN` overrides the stored login for that process, which is how you authenticate in CI without running `hf auth login`. In GitHub Actions put it in a repository secret, not the workflow file.

## 5. Use the token in Python

```python
from huggingface_hub import login
login()  # reads HF_TOKEN, or prompts if unset
```

Logs the whole process in. Or authenticate one call only:

```python
from huggingface_hub import hf_hub_download
hf_hub_download("meta-llama/Llama-3.1-8B-Instruct", "config.json", token="hf_xxx")
```

The `token=` argument works the same way on `from_pretrained` in transformers and diffusers.

## 6. Fix 401 and 403 errors

- **401 Unauthorized / "Repository Not Found"** — no token, or a bad one. The Hub hides gated repos from anonymous callers, so a missing token looks like a missing repo. Run `hf auth login` or set `HF_TOKEN`, then `hf auth whoami` to confirm.
- **403 with `GatedRepoError`** — you are authenticated, but "Access to model ... is restricted and you are not in the authorized list." You have not accepted the licence, or manual approval is still pending. Open the model page, agree, and wait if approval is manual.
- **403 with header `x-error-code: GatedRepo` on a valid read token** — the token's account is not the approved one. Log in with the same account that accepted the licence.

## Verify it worked

```bash
hf download meta-llama/Llama-3.1-8B-Instruct config.json
```

Prints a local file path with no 401 or 403. Auth and licence are both in place.

---

Next: [download models with the hf CLI](/guides/hugging-face-cli-download-models/) · [generate images with FLUX locally](/guides/generate-images-with-flux-locally/).