Get started with Model Repository

Create a Model Repository and upload your first model files to it.

Before you begin

You need a TIR project in Delhi or Chennai and permission to create repositories in it. A Member gets that permission through a policy that allows creating and updating for the Inference service.

Create the repository

1. In TIR, open your project. In the side menu, under Inference, select Model Repository.
2. Select Create Repository. The Create Model Repository form opens.
3. Keep the suggested Repository Name, which starts with tir-model-, or type your own. A name has up to 50 characters. It has only lowercase letters, numbers, and hyphens, starts with a letter, and ends with a letter or number.
4. Choose a Model Type. Pick PyTorch for PyTorch and Hugging Face models, Triton for a Triton Inference Server model repository, or Custom for any other format.
5. Leave Storage Type on New EOS Bucket, so that TIR creates the bucket for you.
6. Select Create.

TIR creates the repository and shows how to upload files to it under Next Steps. Close that to return to the list, where your new repository comes first and is selected.

Upload a file

1. In the list, select the row of your repository. Its tabs open below.
2. Open the Model Files tab and select Upload.
3. To put the files at the top of the bucket, leave Destination Bucket / Folder Path as /. To use a folder, type it, such as v1.
4. Select up to 10 files, or drag them in, then select Upload.

When every file has finished, select Ok. The files are now on the Model Files tab.

Check that it works

Your repository is in the list, and its Model Files tab shows the files you uploaded. If a message appears instead, see Troubleshooting Model Repository.

Use the repository from an endpoint

To serve the model, select Deploy Model on the repository’s row. Pick a framework, then on the endpoint form select Link with Model Repository and choose this repository. See Use a repository from a Model Endpoint.

What to do next

• Upload model files with the MinIO CLI, s3cmd, or the Python SDK for large files and whole folders, and for any file over 5 GB.
• Connect to a repository with the MinIO CLI or s3cmd to copy a whole model folder in one command.
• Use a repository from an instance or notebook to download the model files or save a trained model.

For help with a message, see Troubleshooting Model Repository.