What is a Modelfile in Ollama?

A Modelfile is a plain-text blueprint that tells Ollama how to build a custom model. It starts with a FROM line naming a base model, then adds optional SYSTEM, PARAMETER, TEMPLATE and LICENSE instructions. You run `ollama create -f Modelfile` to bake it into a reusable model — like a Dockerfile, but for a local LLM.

How do I set a system prompt for an Ollama model?

Add a SYSTEM instruction to your Modelfile — for example `SYSTEM """You are a SQL tutor."""`. This generator wraps your prompt in the triple-quote block form so multi-line prompts work, then you build the model once with `ollama create`. After that, the system prompt is baked in and you never retype it.

How do I change the temperature or context length in Ollama?

Add PARAMETER lines: `PARAMETER temperature 0.3` and `PARAMETER num_ctx 8192`. Temperature (default 0.8) controls randomness; num_ctx (default 2048) sets the context window in tokens. Set them in the parameters grid above and the tool writes the exact lines. A larger num_ctx uses more memory.

How do I create a model with ollama create?

Save the generated text to a file named Modelfile, open a terminal in the same folder, and run `ollama create my-model -f Modelfile`. Ollama reads the FROM line, applies your settings and builds the model locally. Then `ollama run my-model` starts a chat. No file is uploaded anywhere.

What are the default parameter values in Ollama?

Per the Ollama Modelfile spec: temperature 0.8, top_k 40, top_p 0.9, min_p 0.0, repeat_penalty 1.1, repeat_last_n 64, num_ctx 2048, num_predict -1 (unbounded), and seed 0 (random). The reference table on this page lists every parameter with its default, valid range and effect.

Why does my Modelfile only show a few PARAMETER lines?

The "Only emit changed params" toggle is on by default, so the tool writes a parameter only when its value differs from the Ollama default. That keeps the file short and readable. Turn the toggle off to write every parameter explicitly — useful when you want a self-documenting file or plan to tweak values later.

Does this tool upload my system prompt anywhere?

No. Everything runs in your browser — the Modelfile, the commands and the validation are all generated locally from the Ollama spec pinned in this page's code. There is no server call, no account and no logging. You can even use it offline once the page has loaded.

Can I build a model from a local GGUF file or LoRA adapter?

Not in this version. It targets `FROM ` — a model name from the Ollama library. Building from a local GGUF path or attaching a LoRA adapter with ADAPTER is a niche flow; follow the Ollama docs for that. This tool focuses on the common case: base model plus system prompt plus parameters.

Developer · AI

Ollama Modelfile Generator

Turn a base model, system prompt and sampling parameters into a valid Ollama Modelfile — plus the exact ollama create and ollama run commands. Built entirely in your browser from the official spec. No signup, no upload.

By Induwara Ashinsana— Executive Director, Ryzera TechnologiesUpdated Jul 9, 2026

Build your Modelfile

Ollama spec · client-side

Presets

Model name

Used in the ollama create command. Lowercase, digits and . _ - only.

Base model (FROM)

Any Ollama library tag, or a namespace/model path.

System prompt (SYSTEM)

The persistent instruction baked into the model. Leave blank to omit the SYSTEM block.

Sampling parameters

Only emit changed params

temperature

top_k

top_p

min_p

repeat_penalty

repeat_last_n

num_ctx

num_predict

seed

Stop sequences

Each becomes one PARAMETER stop line. Press Enter or Add.

Generated Modelfile

FROM llama3.1

SYSTEM """You are a helpful assistant."""

1 · Build the model

ollama create my-model -f Modelfile

Save the Modelfile next to your terminal, then run this.

2 · Run it

ollama run my-model

Starts an interactive chat with your custom model.

Everything is generated in your browser from the official Ollama Modelfile spec. No account, no upload, no network calls.

How it works

A Modelfile is Ollama's build recipe for a custom model — the local-LLM equivalent of a Dockerfile. This generator maps each form field to an instruction from the official Ollama Modelfile grammar and assembles them in the order Ollama expects. Nothing is invented: every default and valid range is pinned from the Ollama documentation and the reference table below renders from those same constants, so the tool and the docs cannot drift.

FROM — always the first line: FROM <base model>. The spec requires it, so the tool blocks generation until you supply a base model such as llama3.1.
SYSTEM — if you enter a system prompt, it is emitted as a triple-quote block: SYSTEM """…""". The block form keeps multi-line prompts intact, and any literal """ inside your text is escaped so it cannot close the block early.
PARAMETER — one line per setting: PARAMETER <name> <value>. With "Only emit changed params" on (the default), a parameter is written only when it differs from the Ollama default, keeping the file minimal. Turn it off to write every parameter explicitly. Each stop sequence becomes its own PARAMETER stop "…" line, which the spec allows you to repeat.
TEMPLATE and LICENSE — emitted verbatim in """ blocks only when the advanced fields are filled, so most files stay clean.
Commands — the tool assembles ollama create <name> -f Modelfile and ollama run <name>. Your name is lowercased and stripped to [a-z0-9._-]to match Ollama's naming rules; if it changes, a notice tells you what it became.

Validation runs on every keystroke. Out-of-range values (for example a temperature above 2.0) raise an inline warning before the output, so you never copy a file Ollama will reject. Because the whole thing is a pure, deterministic function of the form, the same inputs always produce the same Modelfile — and an internal cross-check confirms the number of PARAMETER lines in the output matches the number of non-default values you set.

Parameter reference

Every supported PARAMETER with its Ollama default, valid range and effect. These are the exact values the generator writes and diffs against.

Parameter	Default	Range	What it does
temperature	0.8	0.0 – 2	Creativity vs. determinism. Higher is more random; lower is more focused. Ollama default 0.8.
top_k	40	0 – 200	Sample only from the K most likely next tokens. Higher gives more diversity; lower is more conservative.
top_p	0.9	0.0 – 1	Nucleus sampling: keep the smallest set of tokens whose probability sums to top_p. Works with top_k.
min_p	0.0	0.0 – 1	Drop tokens below this fraction of the most likely token's probability. An alternative to top_p.
repeat_penalty	1.1	0.0 – 2	Penalise repeated tokens. Higher penalises repetition more strongly; 1.0 disables the penalty.
repeat_last_n	64	-1 – 2,048	How far back the repeat penalty looks. 0 disables it; -1 uses the full context window (num_ctx).
num_ctx	2048	256 – 131,072	Context window size in tokens. Larger fits more history but uses more memory (VRAM/RAM).
num_predict	-1	-2 – 8,192	Max tokens to generate per reply. -1 = generate until a stop; -2 = fill the remaining context.
seed	0	0 – 4,294,967,295	Random seed. A fixed non-zero value makes generations reproducible for the same prompt; 0 is random.

Worked examples

Low-temperature SQL tutor (minimal file)

Only changed params · on

Name sql-tutor · base llama3.1 · temperature 0.3 · num_ctx 8192
Only 0.3 ≠ 0.8 and 8192 ≠ 2048, so just two PARAMETER lines appear.
FROM llama3.1
SYSTEM """You are a SQL tutor. Always answer with ANSI SQL and explain briefly."""
PARAMETER temperature 0.3
PARAMETER num_ctx 8192
→ ollama create sql-tutor -f Modelfile · ollama run sql-tutor

Creative writer with stop tokens (explicit file)

Only changed params · off

Name story-bot · base mistral · temperature 1.2 · top_p 0.95 · repeat_penalty 1.15 · num_predict 512
Stop sequences <|end|> and User: each get their own quoted line.
FROM mistral
SYSTEM """You are a vivid fiction writer."""
PARAMETER temperature 1.2 … (all 9 params written because the toggle is off)
PARAMETER stop "<|end|>"
PARAMETER stop "User:"
→ ollama create story-bot -f Modelfile · ollama run story-bot

Edge case: messy model name

Sanitisation

You type the name "My SQL Tutor!" with spaces and a bang.
Ollama names allow only lowercase, digits and . _ - characters.
The tool sanitises it to my-sql-tutor and shows a notice.
→ ollama create my-sql-tutor -f Modelfile

Frequently asked questions

Sources & references

Parameter defaults and ranges on this page were last cross-checked against the Ollama documentation on 2026-07-09. They are reviewed whenever Ollama publishes a change to the Modelfile spec.

Related tools

LiveAI

System Prompt Generator

Turn a short form into a structured system prompt — role, task, tone, output format, guardrails and few-shot examples — for ChatGPT Custom Instructions, a Custom GPT, or the Claude, OpenAI or Gemini system field. Generic Markdown or Claude XML style, with a live preview and copy/download. Runs entirely in your browser; nothing is uploaded, no AI rewriting.

Open tool

LiveAI

AI Model Card Generator

Generate a complete, Hub-valid Hugging Face model card (README.md) from a form — correctly-keyed YAML frontmatter (license, pipeline_tag, datasets, model-index) plus the standard documentation sections. Runs in your browser, nothing you type is uploaded.

Open tool

LiveAI

AI Chat Template Generator

Wrap a system prompt and a multi-turn conversation in the exact special tokens an open model expects — ChatML, Llama 3.x, Llama 2, Mistral, Gemma, Phi-3, Qwen, DeepSeek, Alpaca or Vicuna — and copy the ready-to-paste prompt for raw completion endpoints (Ollama, llama.cpp, vLLM). Fully in-browser.

Open tool

Rate this tool

Be the first to rate

Comments & feedback

Spotted a bug or want an improvement? Tell us — our team reviews every comment, and good ideas get built. Comments are public and anonymous.

Found a bug, edge case, or want to suggest an improvement?

Email me at [email protected] — most fixes ship within 24 hours.