Model Version Support

We provide the capability to download and register various commits of the single model from HuggingFace. Follow the steps below for the same :

• Generate MAR files for the required HuggingFace commits by passing it's commit ID in the "--repo_version" argument
• Deploy TorchServe with any one of the versions passed through the "--repo_version" argument
• Register the rest of the required versions through the register additional models request.

Set Default Model Version

If multiple versions of the same model are registered, we can set a particular version as the default for inferencing by running the following command:

curl -v -X PUT "http://{inference_server_endpoint}:{management_port}/{model_name}/{repo_version}/set-default"