Michael Limberger

Quick Reference

Keep this nearby

Everything you need on one page: model pull, parameters, card essentials, Modelfile template, OpenWebUI setup, stops, framing, context habits, and a short when-it-breaks tree.

The model

Magidonia-24B-v4.3 by TheDrummer. Pull:

Show me

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

Quantizations (approximate):

Q8_0    ~25GB    Highest quality
        Q6_K    ~17GB    Strong quality / speed balance
        Q5_K_M  ~14GB    Good quality, faster
        Q4_K_M  ~10GB    Acceptable on smaller machines

Recommended parameters

For Magidonia character chat:

temperature       1.0
        top_k             40
        top_p             0.95
        min_p             0.02
        repeat_penalty    1.0     (keep at 1.0)
        max_tokens        2048
        context_window    16384   (8K-16K sweet spot)
        stop_sequences    "User:", "\nUser:"

Character card essentials

Minimum:

1. Name
        2. Personality traits (5-7, specific not generic)
        3. Speech style (how they talk)
        4. Example dialogues (3-5 showing range)
        5. First message / opening

Optional but powerful:

6. Scene setting
        7. Core beliefs / motivations
        8. How they show up (positive framing, not a forbid list)

Ollama Modelfile template

Copy, fill SYSTEM, save as mycharacter.modelfile.

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

PARAMETER temperature 1.0
PARAMETER top_k 40
PARAMETER top_p 0.95
PARAMETER min_p 0.02
PARAMETER repeat_penalty 1.0
PARAMETER num_ctx 16384
PARAMETER num_predict 2048
PARAMETER stop "User:"
PARAMETER stop "\nUser:"

SYSTEM """
[Put your character description, speech style, examples, and opening here]
"""

Create: ollama create mycharacter -f mycharacter.modelfile Run: ollama run mycharacter

Ollama commands

List models:

Show me

ollama list

Run:

Show me

ollama run mycharacter

Delete:

Show me

ollama rm mycharacter

Show details:

Show me

ollama show mycharacter

Create from Modelfile:

Show me

ollama create mycharacter -f mycharacter.modelfile

Pull from HuggingFace:

Show me

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

Copy a model for edits:

Show me

ollama cp mycharacter mycharacter-v2

OpenWebUI setup

With Ollama running, start OpenWebUI (Docker example or your local install), open the UI, create a model, paste the SYSTEM text, set Magidonia parameters, and add stop sequences User: and \nUser:.

 - Temperature: 1.0
         - Top K: 40
         - Top P: 0.95
         - Min P: 0.02
         - Repeat Penalty: 1.0

Stop sequences

Without these, the model invents both sides. Minimum: User: and \nUser:. For named characters also stop on their name:

"Mark:"
        "\nMark:"
        "Nyx:"
        "\nNyx:"

Positive framing cheat sheet

Negative	Positive
Never break character	Always stay in character
Don't mention being AI	Maintain your persona fully
Don't sound like a robot	Speak naturally and warmly
Don't give unsolicited advice	Share thoughts when asked

Context management

Keep context around 8K to 16K. After roughly thirty to fifty messages, inject a short trait refresh:

[Traits: witty, skeptical, loves stories. Speaks plainly. Not preachy. Current mood: engaged.]

Every fifty-plus messages, summarize and consider a fresh chat with that summary at the top.

Alternative models

Same Modelfile shape works with sibling pulls:

Show me

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

Show me

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

Show me

ollama pull hf.co/bartowski/TheDrummer_Rocinante-X-12B-v1-GGUF:Q8_0

When things go wrong

Sounds generic → check card, refresh, repeat_penalty 1.0, summarize.
Too repetitive → repeat_penalty 1.0, disable DRY/XTC, raise temperature, vary prompts.
Generates both sides → add stop sequences.
Drifts over time → summarize, refresh, keep context right-sized.
Refuses creative fiction → remove forbid-lists; try Heretic if needed.
Will not download → disk space, off-peak, or smaller quant.
Missing in OpenWebUI → Ollama running, list shows model, URL localhost:11434.

Useful links

TheDrummer models:   https://huggingface.co/TheDrummer
        Bartowski quants:    https://huggingface.co/bartowski
        Ollama:              https://ollama.ai
        OpenWebUI:           https://github.com/open-webui/open-webui

Files you may keep

CHARACTER.modelfile      - card + parameters for ollama create
        CHARACTER.system.txt     - SYSTEM text for OpenWebUI copy/paste
        CHARACTER.examples.txt   - example dialogues for easy edits
        CHARACTER.known_facts.md - external memory for plot and relationships

The mindset

Characters beat blank chatbots because they have stakes. Bigger context is not automatically better. Right-sized context is. Your system prompt is a character bible, not a policy document of don'ts.

Five-minute start

Show me

ollama pull hf.co/bartowski/TheDrummer_Magidonia-24B-v4.3-GGUF:Q8_0
        ollama create mychar -f mychar.modelfile
        ollama run mychar

Then open OpenWebUI if you want the browser UI. That is enough to start building.

Debugging checklist

Won't start:     ☐ Ollama running? ☐ Model downloaded? ☐ ollama create done?
        Wrong model:     ☐ Right name? ☐ ollama list matches what you run?
        Bad answers:     ☐ Stops set? ☐ SYSTEM correct? ☐ repeat_penalty = 1.0?
        Too slow:        ☐ GPU in use? ☐ Context oversized? ☐ Quant too heavy?
        Drift:           ☐ Card vague? ☐ Window too large? ☐ Time for refresh?
        OpenWebUI blank: ☐ Ollama up? ☐ localhost:11434? ☐ Restart UI?

Remember

The model download is the easy part. The craft is writing a card that captures a person, managing context so they stay coherent, and framing instructions so Magidonia can follow them. That is the whole series.