Documentation — ollama-toolkit

ollama-toolkit is a Python CLI utility for exporting and importing Ollama models (.tar.gz) to streamline backups, sharing, or migration across machines.

Designed for developers, researchers, and ML teams looking to package and deploy local Ollama models with ease.

---

What’s It For?

• Backup a custom model locally with all its layers
• Transfer models to another environment without re-downloading
• Archive models for offline or long-term use
• Recreate a complete Ollama model from a .tar.gz archive

---

Key Features

• Export a model with all necessary files
• Generate a self-contained .tar.gz archive
• Automatically recreate the model using ollama create
• Built-in progress bar via tqdm
• Compatible with any locally installed Ollama model

---

Installation

Requirements

• Python ≥ 3.7
• Ollama installed and accessible from your terminal
• tqdm (installed automatically)

Install from PyPI or Source

Install via PyPI:

pip install ollama-panongbene

Or install from source:

git clone https://github.com/openDataSenegal/ollama_toolkit.git
cd ollama-toolkit
pip install .

Once installed, two global CLI commands are available:

ollama-export <model_name>
ollama-import <archive.tar.gz> <model_name>

---

CLI Commands

ollama-export

Exports an Ollama model to a .tar.gz archive including the Modelfile and all binary layers.

ollama-export <model_name> [-o OUTPUT] [-v]

Arguments

Argument	Description
model_name	Name of the model to export (e.g., phi:latest)
-o, --output	Optional output filename for the archive (.tar.gz)
-v, --verbose	Enable detailed logs (e.g., Modelfile content, layers)

Example

ollama-export llama3:instruct -o llama3_instruct.tar.gz -v

---

ollama-import

Imports a model from a .tar.gz archive and recreates it via ollama create.

ollama-import <archive.tar.gz> <model_name>

Arguments

Argument	Description
archive.tar.gz	Path to the archive containing model files
model_name	Name to assign to the model (used during model creation)

Example

ollama-import phi_latest.tar.gz phi:latest

---

Archive Structure

Each export archive includes:

phi_latest.tar.gz
├── Modelfile
├── layer0.bin
├── layer1.bin
├── ...

This structure allows full offline restoration.

---

Use Cases

• Seamless model transfers across machines or teams
• Use Ollama in offline or air-gapped environments
• Archive models for reproducibility and long-term access
• Avoid redundant large downloads
• Share custom Modelfile setups and binary layers

---

Can I Store Archives on External Drives or Git?

Yes — archives are fully portable. You can version them using Git LFS or store them on external media for future reuse.

---

Are Custom or Private Models Supported?

Absolutely. As long as the model is installed locally in Ollama, you can export and re-import it — including custom layers and any Modelfile configuration.

---

Author

• Developed and maintained by Panongbene Sawadogo
• Contact: amet1900@gmail.com

---

License

MIT License — Free for commercial and educational use.

---