Hugging Face Integration

Connect models from Hugging Face Hub to Safeguard for AI model SBOM generation and dependency scanning.

Prerequisites

• A Hugging Face account (optional for public models)
• Hugging Face Access Token (for private models or higher rate limits)

Public Models

Hugging Face public models can be scanned without authentication.

Step 1: Navigate to Integrations

Go to Integrations from the sidebar and click Connect on the Hugging Face card.

Step 2: Enter Model Reference

1. Select the Public tab
2. Enter a Name for this configuration
3. Optionally add a Description
4. Enter the Hugging Face model ID:
   • Format: organization/model-name or username/model-name
   • Example: meta-llama/Llama-2-7b, openai/whisper-base
5. Click Add

Step 3: Review & Connect

1. Configure Project Name and Version
2. Click Connect to complete

---

Private Models

For private models or gated models, you'll need a Hugging Face Access Token.

Step 1: Navigate to Integrations

Go to Integrations from the sidebar and click Connect on the Hugging Face card.

Step 2: Enter Access Token

1. Select the Private tab
2. Enter a Name for this configuration
3. Optionally add a Description
4. Enter your Hugging Face Access Token
5. Click Verify Credentials

Step 3: Select Models

1. Once verified, browse your models or enter model IDs
2. Select the models you want to scan
3. Choose specific revisions (optional)

Step 4: Configure & Connect

1. Set Project Name and Version for each model
2. Click Connect to complete

---

Creating a Hugging Face Access Token

1. Sign in to Hugging Face
2. Click your profile picture then Settings
3. Select Access Tokens from the sidebar
4. Click New token
5. Enter a name (e.g., "Safeguard Integration")
6. Select token type:
   • Read - For scanning public and private models you own
   • Write - Not needed for Safeguard
7. Click Generate token
8. Copy the token immediately

---

What Gets Scanned

When scanning Hugging Face models, Safeguard analyzes:

---

Supported Model Formats

---

Troubleshooting

"Model not found"

• Verify the model ID is correct (case-sensitive)
• Check if the model exists on Hugging Face Hub
• For private models, ensure your token has access

"Access denied"

• For gated models, ensure you've accepted the model's terms
• Verify your token has read access
• Check if the model requires specific permissions

"Credentials verification failed"

• Verify the token was copied correctly
• Check that the token hasn't been revoked
• Ensure the token has the required permissions

---

Best Practices

• Use read-only tokens - Safeguard only needs read access
• Scan specific revisions - Pin to specific commits for reproducibility
• Check model cards - Review the model's documentation and license
• Prefer SafeTensors - More secure than Pickle-based formats