Request a model addition in the Hugging Face collection on Azure

At the moment the Hugging Face collection on Azure ML / AI contains +10,000 open models from the Hugging Face Hub, leveraging open-source inference solutions such as Text Generation Inference (TGI), Text Embeddings Inference (TEI), or the Hugging Face Inference Toolkit, among others to come.

To request a model addition into the Hugging Face collection on the Azure ML / AI catalog (it’s shared among those services), you can either navigate to the model card on the Hugging Face Hub via https://hf.co/models and then click the “Deploy” button and look for the “Deploy to Azure ML” option, that will automatically check whether a given model is already in the collection or not, and it will prompt a “Request to add”.

Alternatively, you can also open an issue with the model/s you’d like to see on the Hugging Face collection on Azure.

Before requesting a model addition, you need to make sure that the model or models that you’d like to see on the Hugging Face collection on Azure match the following criteria:

• Have any of the Transformers, Diffusers or Sentence-Transformers tags on the Hugging Face Hub, meaning that the model architecture is compatible with any of those. If the model you’d like to see in the collection doesn’t match this criteria, you can maybe check the Contributing a new model to Transformers guide on how to add new modular-based model architectures into Transformers.

• Make sure that the task-tag is any of the tasks supported at the moment, as per the listing at Azure AI - Supported Tasks. If it’s not there neither among the listing of upcoming tasks in that same page, then feel free to open an issue requesting the addition of that given task. Again as per the point above, it needs to be a Transformers, Diffusers, or Sentence-Transformers compatible task.

• If you want to benefit from the production-like inference solutions and the OpenAI-compatible interfaces for text generation and embeddings, you should also make sure that the given model has either the text-generation-inference (shortened as tgi) or text-embeddings-inference (shortened as tei) tags, respectively.

• There might be some cases where the “Deploy to Azure ML” button is not enabled (shouldn’t be the case, but can happen), in that case feel free to open an issue sharing the model card and we’ll identify if that model can land on our catalog in Azure.

• The Hugging Face Hub models MUST be public and ideally without any kind of gating restrictions, private or gated models won’t be considered at the moment.

• The Hugging Face Hub models with trust_remote_code are not allowed for security reasons, unless manually verified or coming from an already verified organization.

• And finally, the model weights need to be in Safetensors format and have passed the JFrog, ClamAV, and the rest of the security checks performed in the Hugging Face Hub.

< > Update on GitHub