What a Modelfile Is: Your AI's DNA

Until now, we have been using "Standard" models exactly as they came from the registry. But what if you want a model that always speaks in a specific tone, uses a specific format, or has a larger context window by default?

To do this, you use a Modelfile.

1. The "Dockerfile" for AI

If you are familiar with Docker, you know that a Dockerfile contains a list of instructions to build a container. A Modelfile is exactly the same concept for LLMs.

It is a simple plain-text file that tells Ollama:

1. Which base model to use (e.g., Llama 3).
2. What the system instructions are (e.g., "You are a helpful assistant").
3. What the technical parameters are (e.g., Temperature, Context window).
4. What the "Stop sequences" are (e.g., Stop when you see the word "End").

---

2. Why use a Modelfile?

Reason A: Consistency

If you are building an app, you don't want to send a massive "System Prompt" every time the user sends a message. By using a Modelfile, you "bake" that personality into the model itself. You just run ollama run my-custom-model.

Reason B: Technical Optimization

You can create a version of Llama 3 that always defaults to a 32k context window or a 0.2 temperature, so you don't have to set those parameters manually every time you open the terminal.

Reason C: Security and Privacy

You can bake internal company rules or "Guardrails" into the Modelfile so that the model refuses to answer questions outside of its intended scope.

---

3. The Lifecycle of a Custom Model

The workflow follows three steps:

1. Write: Create a file named Modelfile (no extension) with your instructions.
2. Create: Run ollama create [new-name] -f Modelfile. This "compiles" your instructions into a new model entry.
3. Run: Use ollama run [new-name] to start chatting with your creation.

---

4. A Simple Example Preview

Here is what a basic Modelfile looks like:

FROM llama3
SYSTEM "You are a polite British butler who answers every question with dignity."
PARAMETER temperature 0.2

In the next lesson, we will break down this syntax line-by-line so you can start building your own.

---

Key Takeaways

• A Modelfile is a configuration script for creating custom models.
• It allows you to bake in system prompts and technical parameters.
• It enables consistency and reproducibility in AI development.
• The ollama create command is used to turn a Modelfile into a runnable model.