Customizing Models with Ollama

• 01 May 2025
• ai,
• llm,
• ollama,
• #tutorial

Initial paragraphs goes here

Creating Custom Models via CLI

Copying A Model

N.B. You need to be using a bash-like shell for some ollama commands. On windows this can be done via Git Bash

Step 1: Create A Model File

To copy an AI model you can use:

• ollama show llama3.2:latest --modelfile > some-filename.modelfile Where:
• llama3.2:latest is the model name
• some-filename.modelfile is the name of the model file we're copying into (the extension should remain .modelfile)

Step 2: Customize the Model File

At the end of the file, add SYSTEM """{Prompt}"""

• Replacing {Prompt} with your desired system prompt (e.g. You are Albert Einstein)

Step 3: Generate a Model from the Model File

To Generate a model, you need a model file (see step 1). Then you can use:

• ollama create ai-name --file some-filename.modelfile

More About the ".modelfile"

A most basic Model file

FROM llama2
PARAMETER temperature 1
PARAMETER num_ctx 4096
SYSTEM You are Mario from super mario bros, acting as an assistant.

Parameters

• Temperature: Higher is more creative, lower is more coherent.^[1] Most models support a temperature range from 0 to 1, but there are exceptions.

References

• ← Previous
  Fisher-Yates Shuffle
• Next →
  Antipattern: The Blob