Local AI with Ollama

[Local AI with Ollama] Customizing Models with Modelfile

Ollama allows you to create customized versions of AI models using a simple configuration file called a Modelfile. This guide walks you through the process of customizing a model, adjusting its behavior, and running your personalized version—all from the command line.

Step 1 : Understand the Modelfile

A Modelfile is a text file (usually named just Modelfile with no extension) that specifies how you want to customize your model. You can set parameters like:

  • The base model to start from
  • Model behavior (system prompt)
  • Creativity (temperature)
  • And more

Step 2 : Create a Modelfile

Create a new file named Modelfile in your working directory. Here's an example of a simple Modelfile that customizes the Llama 3.2 model:

FROM llama3.2

PARAMETER temperature 0.3

SYSTEM """
You are BriefAI. A very smart assistant who answers questions succinctly and informatively.
"""

  • FROM llama3.2 tells Ollama to use the Llama 3.2 model as the base.
  • PARAMETER temperature 0.3 makes the model less creative and more direct.
  • The SYSTEM block defines the assistant's personality, now explicitly introducing it as "BriefAI".

Step 3 : Build Your Custom Model

Use the ollama create command to build your model from the Modelfile. Specify a new name for your custom model (e.g., BriefAI):

ollama create BriefAI -f Modelfile

Ollama will process the Modelfile and create a new model called BriefAI based on your specifications.

Step 4 : Verify Your Custom Model

List your available models to confirm the new one was created:

ollama list

You should see BriefAI listed alongside other models.

Step 5 : Run and Test Your Custom Model

Start a chat session with your custom model:

ollama run BriefAI

Try asking questions like:

  • What is your name?
  • Tell me about the solar system.

The model should respond according to the system prompt and parameters you set in the Modelfile (e.g., introducing itself as BriefAI, being succinct and informative).

Step 6 : Edit and Iterate

You can further customize your Modelfile with more parameters, different prompts, or other settings supported by Ollama. Re-run the ollama create command to update your custom model.

Step 7 : Remove a Custom Model

If you want to delete your custom model to save space:

ollama rm BriefAI

With Modelfile, Ollama gives you the power to build AI that fits your exact needs.