Text Generation
GGUF
German
qwen3
qlora
minecraft
npc
dialogue
german
living-villages
villagemind
conversational
Instructions to use angecoded/VillageMind with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use angecoded/VillageMind with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf angecoded/VillageMind:Q4_K_M # Run inference directly in the terminal: llama cli -hf angecoded/VillageMind:Q4_K_M
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf angecoded/VillageMind:Q4_K_M # Run inference directly in the terminal: llama cli -hf angecoded/VillageMind:Q4_K_M
Use pre-built binary
# Download pre-built binary from: # https://github.com/ggerganov/llama.cpp/releases # Start a local OpenAI-compatible server with a web UI: ./llama-server -hf angecoded/VillageMind:Q4_K_M # Run inference directly in the terminal: ./llama-cli -hf angecoded/VillageMind:Q4_K_M
Build from source code
git clone https://github.com/ggerganov/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-server llama-cli # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama-server -hf angecoded/VillageMind:Q4_K_M # Run inference directly in the terminal: ./build/bin/llama-cli -hf angecoded/VillageMind:Q4_K_M
Use Docker
docker model run hf.co/angecoded/VillageMind:Q4_K_M
- LM Studio
- Jan
- vLLM
How to use angecoded/VillageMind with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "angecoded/VillageMind" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "angecoded/VillageMind", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/angecoded/VillageMind:Q4_K_M
- Ollama
How to use angecoded/VillageMind with Ollama:
ollama run hf.co/angecoded/VillageMind:Q4_K_M
- Unsloth Studio
How to use angecoded/VillageMind with Unsloth Studio:
Install Unsloth Studio (macOS, Linux, WSL)
curl -fsSL https://unsloth.ai/install.sh | sh # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for angecoded/VillageMind to start chatting
Install Unsloth Studio (Windows)
irm https://unsloth.ai/install.ps1 | iex # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for angecoded/VillageMind to start chatting
Using HuggingFace Spaces for Unsloth
# No setup required # Open https://huggingface.co/spaces/unsloth/studio in your browser # Search for angecoded/VillageMind to start chatting
- Pi
How to use angecoded/VillageMind with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf angecoded/VillageMind:Q4_K_M
Configure the model in Pi
# Install Pi: npm install -g @mariozechner/pi-coding-agent # Add to ~/.pi/agent/models.json: { "providers": { "llama-cpp": { "baseUrl": "http://localhost:8080/v1", "api": "openai-completions", "apiKey": "none", "models": [ { "id": "angecoded/VillageMind:Q4_K_M" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Docker Model Runner
How to use angecoded/VillageMind with Docker Model Runner:
docker model run hf.co/angecoded/VillageMind:Q4_K_M
- Lemonade
How to use angecoded/VillageMind with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull angecoded/VillageMind:Q4_K_M
Run and chat with the model
lemonade run user.VillageMind-Q4_K_M
List all available models
lemonade list
- Hermes Agent
How to use angecoded/VillageMind with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf angecoded/VillageMind:Q4_K_M
Configure Hermes
# Install Hermes: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash hermes setup # Point Hermes at the local server: hermes config set model.provider custom hermes config set model.base_url http://127.0.0.1:8080/v1 hermes config set model.default angecoded/VillageMind:Q4_K_M
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use angecoded/VillageMind with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf angecoded/VillageMind:Q4_K_M
Configure OpenClaw
# Install OpenClaw: npm install -g openclaw@latest # Register the local server and set it as the default model: openclaw onboard --non-interactive --mode local \ --auth-choice custom-api-key \ --custom-base-url http://127.0.0.1:8080/v1 \ --custom-model-id "angecoded/VillageMind:Q4_K_M" \ --custom-provider-id llama-cpp \ --custom-compatibility openai \ --custom-text-input \ --accept-risk \ --skip-health
Run OpenClaw
openclaw agent --local --agent main --message "Hello from Hugging Face"
File size: 8,222 Bytes
7c81758 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 | ---
language:
- de
license: apache-2.0
base_model: Qwen/Qwen3-1.7B
base_model_relation: finetune
pipeline_tag: text-generation
tags:
- qwen3
- gguf
- qlora
- minecraft
- npc
- dialogue
- german
- living-villages
- villagemind
---
# VillageMind 0.1-dev
**VillageMind** is a compact local language model fine-tuned for natural NPC dialogue in the **Living Villages** Minecraft mod project.
The goal is not to let an LLM control the Minecraft world directly. VillageMind is responsible for **language, personality and context-aware dialogue**, while the Living Villages game logic remains authoritative for inventory, relationships, reputation, jobs, combat, ownership, world state and other gameplay facts.
> **Status:** Development release candidate
> **Current release:** `VillageMind-0.1-dev-Q4_K_M.gguf`
> **Base model:** `Qwen/Qwen3-1.7B`
> **Format:** GGUF
> **Quantization:** Q4_K_M
> **Primary validated language:** German
## Purpose
VillageMind is designed for NPCs that should feel like inhabitants of a Minecraft world rather than generic assistants.
The fine-tuning focuses on short, natural responses to situations such as:
- strangers asking for favors
- friends requesting help
- hostile players
- attacks and threats
- private property and theft
- work and free time
- following or refusing to follow a player
- apologies and relationships
- rumors and uncertainty
- village emergencies such as fire
- grief and social reactions
- donations
- inventory questions
- village resource shortages
The intended integration is **Living Villages**, where residents can have persistent personalities, memories, relationships, jobs, inventories and social behavior.
## Architecture
VillageMind is derived from:
**[Qwen/Qwen3-1.7B](https://huggingface.co/Qwen/Qwen3-1.7B)**
Training approach:
- QLoRA / LoRA fine-tuning
- 4-bit NF4 training
- approximately 32,000 training examples
- 1,200 held-out evaluation examples
- best adapter selected by evaluation loss
- final model merged with the Qwen3-1.7B base model
- converted to GGUF
- quantized to Q4_K_M
The selected adapter reached a best evaluation loss of approximately **0.1117** during training.
## Evaluation
VillageMind was evaluated with a separate hand-designed final examination containing **180 gameplay-oriented dialogue cases**.
The critical subset contains **50 tests** covering behaviors that should not fail in normal gameplay, including inventory truthfulness, ownership boundaries, stranger/enemy behavior and uncertainty around rumors.
### Best adapter
| Metric | Result |
|---|---:|
| Tests passed | 176 / 180 |
| Overall score | 97.78% |
| Critical tests | 50 / 50 |
| Critical score | 100% |
### Q4_K_M GGUF
The final GGUF was tested again after merge, conversion and quantization using the same 180-case suite.
| Metric | Result |
|---|---:|
| Tests passed | 174 / 180 |
| Overall score | 96.67% |
| Critical tests | 50 / 50 |
| Critical score | 100% |
| Release candidate | Yes |
The tested GGUF has the SHA-256 checksum:
```text
475b1b819c42a8667ea2d101f68020f0c8e10315bfbbbd5a21d3227436a268b1
```
The repository also contains the generated manifest and release-candidate evaluation metadata.
## Known limitation
The main known weakness of VillageMind 0.1-dev is **resource-shortage attribution**.
In a small number of evaluation cases, the model correctly recognized that the village lacked a resource but referred to the **wrong resource** in its response.
For example, if the authoritative game state says that wheat is scarce, the model may occasionally talk about iron instead.
For this reason, Living Villages is designed around a strict separation:
```text
Minecraft world state
↓
Living Villages game logic
(authoritative facts and actions)
↓
VillageMind
(language generation)
↓
Fact / action validation
↓
Player
```
VillageMind must therefore **not** be treated as the authoritative source for:
- item inventories
- resource counts
- ownership
- memories
- reputation values
- relationship state
- jobs or task feasibility
- combat state
- world changes
- spawning or removing items
- any other persistent gameplay state
These facts should be supplied and validated by the host application.
## Language
VillageMind 0.1-dev has been specifically evaluated in **German**.
The Qwen3 base model is multilingual, and future Living Villages versions are intended to support both German and English dialogue based on the player's Minecraft language setting. However, **English behavior has not yet received the same dedicated VillageMind 0.1-dev evaluation**, so this release should not be described as equally validated in English yet.
## Example behavior
### Stranger
**Player:**
> Bring me some iron.
**Resident:**
> Du bist neu hier. Ich nehme keine Aufträge von Fremden an.
### Private property
**Player:**
> I'm taking this from your chest.
**Resident:**
> Finger weg. Das gehört dir nicht.
### Rumor
**Player:**
> Is the rumor true?
**Resident:**
> Ich hab was gehört, aber ich weiß nicht, ob's stimmt.
### Inventory
If the game provides the resident inventory as:
```text
7 coal
2 bread
```
VillageMind can respond:
> Ich hab gerade 7 Kohle, 2 Brot. Mehr nicht.
The host application should still validate factual responses against the authoritative game state.
## Recommended use
The current release is intended primarily for:
- Living Villages development
- local Minecraft NPC dialogue
- experimentation with state-aware game characters
- local inference through llama.cpp-compatible runtimes
It is **not** intended as:
- a general-purpose assistant
- a source of factual real-world information
- an autonomous authority over game state
- a replacement for deterministic gameplay logic
## GGUF file
Primary release file:
```text
VillageMind-0.1-dev-Q4_K_M.gguf
```
Recommended runtime:
- llama.cpp or a compatible GGUF inference runtime
The exact integration parameters for Living Villages may change during development.
## Repository files
```text
VillageMind-0.1-dev-Q4_K_M.gguf
VillageMind-0.1-dev-manifest.json
release-candidate-gguf.json
README.md
```
## Base model and license
VillageMind is a fine-tuned derivative of **Qwen3-1.7B** by the Qwen Team.
Base model:
https://huggingface.co/Qwen/Qwen3-1.7B
Qwen3-1.7B is distributed under the **Apache License 2.0**. VillageMind is likewise published under **Apache-2.0**.
Please review the base model's documentation and license when redistributing or building on VillageMind.
## Project status
VillageMind 0.1-dev is the first development model candidate intended for integration into **Living Villages**.
The project is under active development. Evaluation results describe this specific development test suite and should **not** be interpreted as a general benchmark score for language-model quality.
---
# Deutsch
**VillageMind** ist ein lokal ausführbares Sprachmodell für natürliche NPC-Dialoge im Minecraft-Mod-Projekt **Living Villages**.
Das Modell soll nicht selbst über die Minecraft-Welt bestimmen. **Living Villages bleibt die autoritative Instanz** für Inventare, Beziehungen, Ruf, Eigentum, Aufgaben, Kampf, Ressourcen und Weltzustände. VillageMind übernimmt hauptsächlich die natürliche Formulierung der NPC-Reaktionen.
## Aktueller Stand
- Basis: `Qwen/Qwen3-1.7B`
- Fine-Tuning: QLoRA / LoRA
- GGUF: `Q4_K_M`
- Best-Adapter-Abschlussprüfung: **176/180 (97,78%)**
- Kritische Tests Best Adapter: **50/50**
- GGUF-Abschlussprüfung: **174/180 (96,67%)**
- Kritische Tests GGUF: **50/50**
- Status: **Release Candidate bestanden**
### Bekannte Schwäche
Bei einigen Tests zu knappen Dorfressourcen nennt VillageMind gelegentlich die falsche Ressource. Deshalb werden solche Aussagen in Living Villages nicht ungeprüft als Wahrheit übernommen.
Die Minecraft-Mod liefert die echten Fakten und kann VillageMinds Antwort anschließend validieren.
### Sprache
Die Version **0.1-dev wurde gezielt auf Deutsch geprüft**. Englisch soll im späteren Living-Villages-Einsatz ebenfalls unterstützt werden, wurde für diesen Modellstand aber noch nicht mit einer gleichwertigen Abschlussprüfung validiert.
|