Ollama Modelfile Sample
Ollama Modelfile Cheat Sheet
Basic Structure
A Modelfile is a recipe for building a model.
FROM <base_model> # Optional settings PARAMETER <key> <value> SYSTEM <text> TEMPLATE <text> ADAPTER <path> LICENSE <text> MESSAGE <role> <text>
FROM (Required)
Base model you build on.
FROM llama3 FROM mistral FROM ./local-model.gguf
PARAMETER
Controls runtime behavior.
Common Parameters
PARAMETER temperature 0.7 PARAMETER top_p 0.9 PARAMETER top_k 40 PARAMETER num_ctx 4096 PARAMETER repeat_penalty 1.1 PARAMETER seed 42 PARAMETER num_predict 128
Description
- temperature – randomness (0 = deterministic, 1 = highly random)
- top_p – nucleus sampling probability cutoff
- top_k – limits token selection pool
- num_ctx – context window size
- repeat_penalty – reduces repetition
- num_predict – max tokens to generate
SYSTEM
Defines assistant behavior.
SYSTEM You are a helpful assistant that explains things simply.
TEMPLATE
Controls prompt formatting.
TEMPLATE """
{{ if .System }}<|system|>
{{ .System }}
{{ end }}
<|user|>
{{ .Prompt }}
<|assistant|>
"""
MESSAGE
Few-shot examples.
MESSAGE user What is 2+2? MESSAGE assistant 4
ADAPTER
Attach LoRA adapters.
ADAPTER ./my-lora
LICENSE
Attach a license.
LICENSE MIT
Full Example
FROM llama3
PARAMETER temperature 0.7
PARAMETER top_p 0.9
PARAMETER num_ctx 4096
SYSTEM You are a precise and concise coding assistant.
TEMPLATE """
{{ if .System }}<|system|>
{{ .System }}
{{ end }}
<|user|>
{{ .Prompt }}
<|assistant|>
"""
MESSAGE user Write a Python function to add two numbers.
MESSAGE assistant def add(a, b): return a + b
Build and Run
ollama create mymodel -f Modelfile ollama run mymodel
Common Pitfalls
- Missing FROM → build fails
- Too large num_ctx → high memory usage
- Incorrect TEMPLATE → broken responses
- High temperature → hallucinations
- Low temperature → overly rigid output
Mental Model
- FROM = base model
- SYSTEM = behavior
- TEMPLATE = prompt structure
- PARAMETER = tuning
- MESSAGE = examples
- ADAPTER = extensions