This repository fulfills a simple purpose: to provide models for ollama to machines that are not connected to the internet.
- Download the model you want to use from the download links section.
- Extract the downloaded file
.tar.gzfile. Then extract the.tarfile located inside the extracted folder. - Move the extracted folder
modelsto the root of your.ollamafolder (e.g./usr/share/ollama/.ollama/models). See where-are-models-stored, from the officialollamaFAQ for more information. - You should now be able to use the model you downloaded. Executing
ollama run MODEL_NAMEshould work as expected, without it trying to download the model.
Important
Due to Contabo limiting my bandwidth to only 100 Mbit/s because my STORAGE VPS is exceeding their set average bandwidth, I’ve added torrents and purchased a Seedbox. Feel free to download and seed them!
With the current setup, downloading the files via the Torrents is the fastest way to get the models.
| Model | Parameters | Last Modified | Size | Download Links |
|---|---|---|---|---|
| codellama | 7B | 2024-08-18 16:48 | 3.4G | Storage VPS / Torrent |
| deepseek-coder-v2 | 16B | 2025-01-02 23:18 | 5.2G | Storage VPS / Torrent |
| gemma2 | 2B | 2024-08-18 15:45 | 1.5G | Storage VPS / Torrent |
| gemma2 | 9B | 2024-08-18 18:24 | 4.8G | Storage VPS / Torrent |
| gemma2 | 27B | 2024-08-19 17:23 | 14G | Storage VPS / Torrent |
| llama2-uncensored | 7B | 2024-08-18 16:32 | 3.4G | Storage VPS / Torrent |
| llama3.1 | 8B | 2024-08-19 17:31 | 4.1G | Storage VPS / Torrent |
| llama3.1 | 70B | 2024-08-18 19:51 | 35G | Storage VPS / Torrent |
| llama3.2-vision | 11B | 2024-11-13 19:53 | 6.7G | Storage VPS / Torrent |
| llama3.2 | 1B | 2024-10-07 00:38 | 1.2G | Storage VPS / Torrent |
| llama3.2 | 3B | 2024-10-07 00:36 | 1.8G | Storage VPS / Torrent |
| llama3 | 8B | 2024-08-18 20:13 | 4.1G | Storage VPS / Torrent |
| mistral-nemo | 12B | 2024-08-19 15:27 | 6.3G | Storage VPS / Torrent |
| mistral | 7B | 2024-08-18 20:08 | 3.6G | Storage VPS / Torrent |
| moondream | 1.8B | 2024-08-18 16:29 | 1.4G | Storage VPS / Torrent |
| nomic-embed-text | --- | 2025-01-08 21:31 | 241M | Storage VPS / Torrent |
| phi3 | 3.8B | 2024-08-18 16:20 | 1.9G | Storage VPS / Torrent |
| qwen2.5-coder | 7B | 2024-11-13 20:27 | 4.3G | Storage VPS / Torrent |
| qwen2.5-coder | 14B | 2024-11-13 20:25 | 8.2G | Storage VPS / Torrent |
| qwen | 0.5B | 2024-08-19 14:48 | 356M | Storage VPS / Torrent |
| qwen | 32B | 2024-08-19 15:01 | 16G | Storage VPS / Torrent |
To export the model yourself, you can use the ollama-exporter.sh script, that I created. The script will create a .tar.gz file of the model you want to export.
Caution
This script will delete everything in the models folder to make sure that only the model you want to export is present. Make sure to back up any models you want to keep! I did this on a separate VM.
Usage: ollama-exporter.sh [OPTIONS]
Options:
-m, --model-name Name of the model to pull and backup (e.g. "moondream", "gemma2:2b", "llama3.1:70b").
-d, --dest-folder Path to the destination folder where the tar.gz file will be moved.
-f, --model-folder Path to the ollama model folder (default: /usr/share/ollama/.ollama/models).
--no-delete Do not delete the original model folder contents after moving.
-h, --help Display this help message and exit.
If no options are provided, the script will display this help message.
Examples:
ollama-exporter.sh -m moondream -d /path/to/backup -f /custom/path/to/models
ollama-exporter.sh -m llama3.1:70b -d /home/pyenb/Daten/Backups/Ollama/modelsAs the model name, everything out of Ollamas registry should work. Everything else should be self-explanatory.
The finished .tar.gz file structure will look like this:
MODEL_NAME.tar.gz
└── MODEL_NAME.tar
└── models
├── blobs
│ └── sha256-...
└── manifests
└── registry.ollama.ai
└── library
└── MODEL_NAME
└── latest
Example output
pyenb@ollama:~$ sudo ./ollama-exporter.sh -m llama3.1:70b -d /home/pyenb/Daten/Backups/Ollama/models
Pulling the Ollama model: llama3.1:70b...
pulling manifest
pulling a677b4a4b70c... 100% 39 GB
pulling 11ce4ee3e170... 100% 1.7 KB
pulling 0ba8f0e314b4... 100% 12 KB
pulling 56bb8bd477a5... 100% 96 B
pulling 654440dac7f3... 100% 486 B
verifying sha256 digest
writing manifest
removing any unused layers
success
Compressing the model folder...
37.2GiB 0:02:41 [ 235MiB/s] [==========================================================================] 100%
Moving the tar.gz file to the destination folder using rclone...
2024/08/18 19:51:15 NOTICE: Config file "/root/.config/rclone/rclone.conf" not found - using defaults
Transferred: 35.464 GiB / 35.464 GiB, 100%, 1.243 MiB/s, ETA 0s
Checks: 2 / 2, 100%
Deleted: 1 (files), 0 (dirs)
Renamed: 1
Transferred: 1 / 1, 100%
Elapsed time: 7m36.3s
Press Enter to continue with deletion or Ctrl+C to cancel...
Deleting the original model folder contents...
Model moved and cleaned up successfully.With GitHub LFS, a "data pack" costs approximately €5 per month for 50GB of storage. The Storage-VPS I'm using also costs around 5€ per month, but with 800GB of SSD storage. So I decided to host the files myself. The only downside is mostly the download speed, so consider supporting me, so that I can upgrade the server.
You can either create the model yourself by following the manual export instructions, or you can open an issue, and I will try to get it done as soon as possible.
Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.
This repository is for research purposes only, the use of this code is your responsibility.
I take NO responsibility and/or liability for how you choose to use any of the source code available here. By using any of the files available in this repository, you understand that you are AGREEING TO USE AT YOUR OWN RISK. Once again, ALL files available here are for EDUCATION and/or RESEARCH purposes ONLY.
