Ollama

Modelfile Commands

Create custom models with Modelfiles. Define base models, system prompts, parameters, and templates to build specialized AI assistants.

7 commands

Pro Tips

Use PARAMETER temperature 0.7 to control creativity (0=deterministic, 1=creative).

SYSTEM instruction sets the persistent persona - use for role-specific assistants.

Inherit from quantized base models (e.g., FROM llama2:7b-q4_0) to save memory.

Common Mistakes

Custom models inherit the license of the base model - check usage restrictions.

Large num_ctx values (context length) dramatically increase memory requirements.

Commands

Create custom model

$ ollama create my-assistant -f Modelfile

Build a custom model from a Modelfile.

Basic Modelfile

$ FROM llama2 SYSTEM You are a helpful assistant. PARAMETER temperature 0.7

Simple Modelfile with system prompt and parameters.

Set context length

$ FROM llama2 PARAMETER num_ctx 4096

Configure model context window size.

Custom template

$ FROM llama2 TEMPLATE "{{ .System }}\n\nUser: {{ .Prompt }}\nAssistant:"

Define custom prompt template format.

Push to registry

$ ollama push username/my-model

Upload custom model to Ollama registry.

From quantized base

$ FROM llama2:7b-q4_0 SYSTEM You are a coding assistant.

Create model from quantized variant for efficiency.

Set stop tokens

$ FROM llama2 PARAMETER stop "User:" PARAMETER stop "Assistant:"

Define tokens that stop generation.