Uploading Custom Models

In this tutorial, you'll learn how to share your custom local LLMs by uploading them to the Ollama model registry. We’ll start with a quick recap of model customization and then walk through the steps to publish and distribute models seamlessly.

Recap

We’ve already covered:

• Customizing existing models with a Modelfile definition
• Building a new model using the ollama create command

Why Use a Model Registry?

Imagine Jane, an AI developer, has fine-tuned a model for her application. Distributing the raw Modelfile means each teammate must pull and rebuild the model locally every time there’s an update—error prone and time-consuming:

A registry-based approach mirrors how container images work with Docker Hub:
Just push updates once, and everyone can pull the latest version:

Publishing Your Model

Follow these steps to upload and share your custom model on Ollama:

1. Create an Ollama Account
   Sign up at ollama.com and verify your email. You’ll use this account to push and manage models.

2. Configure Your SSH Public Key
   Ollama uses SSH keys to authenticate model uploads. Locate or generate your public key and add it to your account settings.

   Platform	Default Public Key Path
   macOS	~/.ollama/id_ed25519.pub
   Linux	~/.ollama/id_ed25519.pub
   Windows	C:\Users\<username>\.ollama\id_ed25519.pub

   Note

   If you don’t have an SSH key pair yet, generate one with:

   ssh-keygen -t ed25519 -f ~/.ollama/id_ed25519 -C "[email protected]"
   

   

3. Tag Your Model for Your Namespace
   Prefix your local model name with your Ollama username:

   $ ollama cp harris your_username/harris
   copied 'harris' to 'your_username/harris'
   

   Warning

   Make sure to replace your_username with your actual Ollama username to avoid naming conflicts.

4. Push the Model to the Registry
   Upload your tagged model:

   $ ollama push your_username/harris
   retrieving manifest
   pushing dde5aa3fc5ff... 100%  2.0 GB
   pushing 966de95ca8a6... 100%  1.4 KB
   ...
   success
   You can find your model at: https://ollama.com/your_username/harris
   

5. View and Run Your Published Model
   Open your model’s page to see details like architecture, parameters, and license—similar to Llama 3.2 and other public models. You’ll also find the ollama run command to pull and launch the model locally:

   

Once configured, any update to your Modelfile can be published with:

$ ollama push your_username/harris

Teammates can instantly pull the latest build:

$ ollama run your_username/harris

Proceed to the demonstration below to see this process in action!

Links and References

• Ollama Documentation
• Docker Hub
• Llama 3.2 Model by Meta AI