Creating Custom Models with Modelfiles

Just like Docker uses a Dockerfile to build custom containers, Ollama uses a Modelfile to define custom system instructions, adjust model parameters, and configure default context lengths.

1. Writing your First Modelfile

Create a text file named Modelfile (no extension) in your workspace:

# 1. Specify the base model
FROM qwen2.5:7b

# 2. Adjust parameters (e.g. context window size to 8k tokens)
PARAMETER num_ctx 8192
PARAMETER temperature 0.3

# 3. Define the system prompt instruction
SYSTEM """
You are a private company support bot. You must follow these rules:
1. Always reply in English.
2. Only write answers if they relate to developer setup instructions.
3. Be professional and concise.
"""

2. Compiling the Model via CLI

Compile your Modelfile into a custom model name using the ollama create command:

# Compile and name your model "company-helper-bot"
ollama create company-helper-bot -f ./Modelfile

3. Running and Verifying

Verify that your custom model is registered:

ollama list

You will see company-helper-bot listed. Launch and test the model in your terminal:

ollama run company-helper-bot