Where integrations are used

You create an integration once, then pick it in each form that needs it. A form lists the current project’s integrations of the type it needs, and lets you create one of that type right there in the form.

Where	Field you pick it in	Integration type	Required
Model Endpoints, create	HF Token Integration	Hugging Face	Yes, for the models in Model Endpoints and Hugging Face
Model Endpoints, create	GitHub Integration under Link with Github	GitHub	Yes, once you choose Link with Github
Fine-Tuning, create	Select Hugging Face Integration	Hugging Face	Yes
Fine-Tuning, create	Select Weight & Biases Integration	Weights & Biases	Only when you enter a Run Name
Model Evaluation, create	Huggingface Integration (Optional)	Hugging Face	Only when the Hugging Face dataset is private
Models Library, a model’s deploy form	HF token integration	Hugging Face	Marked “Required for this model.” or “Optional for this model.” for each model

Model Endpoints and Hugging Face

The create form asks for HF Token Integration in three cases:

• The model comes from Hugging Face, as it does for most frameworks.
• It is a fine-tuned model.
• It is a vLLM model served as PEFT from E2E storage.

Some frameworks, such as YOLOv8 and the BAAI frameworks, skip this field.

The form checks that the token can reach the model. The check runs as soon as you pick the integration, or when you select Check access on forms that have that button.

For a gated model, the model’s owner must also have granted access to the Hugging Face account that owns the token. Request access on the model’s page on Hugging Face, then check again.

Model Endpoints and GitHub

For the Custom Container, Custom Recipe, and TIR Optimised frameworks, Download Type offers Link with Github. Enter GitHub Repository as https://github.com/<owner>/<repo>/tree/<branch>, pick a GitHub Integration, and the form checks that the token can reach the repository.

Fine-Tuning and Weights & Biases

Every fine-tuning job needs a Hugging Face integration, picked in Select Hugging Face Integration. To log the job’s metrics to Weights & Biases, also pick a Weights & Biases integration in Select Weight & Biases Integration. A Run Name, if you give one, needs that integration too.

Related pages

• Create a Hugging Face integration
• Create a GitHub integration
• Create a Weights & Biases integration