What is Ollama's Modelfile and how to create custom models?

Ollama uses Modelfile to define and customize models. Modelfile is a text file, similar to Dockerfile, used to describe how to build and configure models.

Basic Modelfile Structure:

dockerfile
FROM llama3.1

PARAMETER temperature 0.7
PARAMETER top_p 0.9
PARAMETER num_ctx 4096

SYSTEM You are a helpful assistant.

TEMPLATE """
{{- range .Messages }}
{{- if eq .Role "user" }}User: {{ .Content }}
{{- else if eq .Role "assistant" }}Assistant: {{ .Content }}
{{- end }}
{{- end }}
Assistant:
"""

Common Instructions:

1. FROM - Specify base model

dockerfile
FROM llama3.1:8b

1. PARAMETER - Set model parameters

dockerfile
PARAMETER temperature 0.7
PARAMETER top_p 0.9
PARAMETER num_ctx 4096
PARAMETER repeat_penalty 1.1

1. SYSTEM - Set system prompt

dockerfile
SYSTEM You are an expert Python developer.

1. TEMPLATE - Customize prompt template

dockerfile
TEMPLATE """
<|user|>
{{ .Prompt }}<|end|>
<|assistant|>
"""

1. LICENSE - Add license information

dockerfile
LICENSE MIT

1. MESSAGE - Add example conversations

dockerfile
MESSAGE user Hello
MESSAGE assistant Hi there!

Creating Custom Models:

bash
# Create Modelfile
cat > Modelfile << EOF
FROM llama3.1
SYSTEM You are a coding assistant specialized in Python.
PARAMETER temperature 0.5
EOF

# Build model
ollama create my-coder -f Modelfile

# Run custom model
ollama run my-coder

View Model's Modelfile:

bash
ollama show --modelfile my-coder