INSTALLATION

This is the simple part. One command. Coffee break. Done.

THE PULL COMMAND

Open your terminal and run:

ollama pull hf.co/bartowski/TheDrummer_Magidonia-24B-v4.3-GGUF:Q8_0

This downloads directly from HuggingFace. Ollama knows how to fetch GGUF files from HF if you prefix the model name with hf.co/

The download is about 25GB. Depending on your internet connection, that's 10-40 minutes. Grab a coffee. Read the chat logs in the BeaverAI Discord. The time will pass.

The model name is long. Intentionally. It tells you:

hf.co/                              - HuggingFace, not Ollama's library
bartowski/                          - The quantizer (bartowski uses iMatrix)
TheDrummer_Magidonia-24B-v4.3-GGUF  - The model and format
:Q8_0                               - The quantization level (8-bit, 0-variant)

When the download completes, you get a digest. That's the hash Ollama uses to verify the model wasn't corrupted. You don't need to do anything with it.

VERIFY THE DOWNLOAD

Once the download completes, verify the model is there:

ollama list

You should see the model in the list. The full name will be shown. It takes up about 25GB of disk space.

To see detailed info about the model:

ollama show hf.co/bartowski/TheDrummer_Magidonia-24B-v4.3-GGUF:Q8_0

This shows you the model's parameters, metadata, and the embedded chat template.

QUICK TEST

Before opening OpenWebUI, do a quick test from the command line:

ollama run hf.co/bartowski/TheDrummer_Magidonia-24B-v4.3-GGUF:Q8_0 "Hello, who are you?"

The model should load, think for a moment, and respond. It'll be your first time seeing it in action. Just a quick hello. The response is usually friendly and coherent.

Type quit and press Enter to exit.

THE NAME IS LONG

Typing that model name every time is tedious. We'll create a shorter alias later using a Modelfile. For now, you don't need to worry about it. Ollama stores the full name internally. The long name is just accuracy.

ALTERNATIVE PULLS FOR DIFFERENT HARDWARE

If you're following along but your hardware is different:

For Q6_K (smaller, still excellent):

ollama pull hf.co/bartowski/TheDrummer_Magidonia-24B-v4.3-GGUF:Q6_K

For the regular Cydonia variant instead:

ollama pull hf.co/bartowski/TheDrummer_Cydonia-24B-v4.3-GGUF:Q8_0

For the heretic version (more uncensored):

ollama pull hf.co/mradermacher/Cydonia-24B-v4.3-heretic-v2-i1-GGUF:Q8_0

All of these follow the same installation process. Same testing process. Same next steps.

HOW OLLAMA HANDLES THE CHAT TEMPLATE

This is a nice thing about modern GGUF files: the chat template is embedded.

The Magidonia GGUF includes the Mistral V7-Tekken chat template. When Ollama loads the model, it automatically detects and uses that template. This means we don't have to write a custom Modelfile to define how to format the chat messages.

In the January RPMax tutorial, we had to create YAML Modelfiles to specify the chat template. Now? Ollama handles it automatically. One less step.

The template is smart: it knows how to format system messages, user messages, and assistant messages correctly. Magidonia was trained with this template, so it understands it.

OPEN WEB UI: IMMEDIATE AVAILABILITY

Once the model is pulled and you have OpenWebUI running:

Open your browser to http://localhost:8501 (or wherever OpenWebUI is running)
Look at the model selector dropdown at the top
You'll see your new model in the list

You don't need to restart OpenWebUI or do anything special. Once ollama list shows the model, OpenWebUI sees it too. They're talking to the same Ollama backend.

If you don't see it, check:

Ollama is still running (ollama list works)
OpenWebUI is still running (browser can reach it)
Try refreshing the page

Rarely, OpenWebUI caches the model list. Hard refresh (Cmd+Shift+R on Mac) solves it.

QUICK TEST IN OPEN WEB UI

Select the model from the dropdown. Send a test message. Something simple:

"Hello Magidonia, introduce yourself."

The model should load into VRAM (you'll see a pause as it loads). Then it should respond. Natural language. No weird formatting. Just a conversation starting.

If it works, you're ready for the next section. Parameters.

If it doesn't:

Check Ollama logs (might show memory errors or corruption)
Check your disk has 25+ GB free
Try the command-line test again to isolate the issue

DONE

That's installation. One command, one verification, one test. The model is ready to go.

Next: we talk about parameters. How to tune Magidonia for the exact behavior you want.