Instructions to use hecan3d/QwenDGAF-V3-4B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use hecan3d/QwenDGAF-V3-4B 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 hecan3d/QwenDGAF-V3-4B:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf hecan3d/QwenDGAF-V3-4B:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf hecan3d/QwenDGAF-V3-4B:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf hecan3d/QwenDGAF-V3-4B: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 hecan3d/QwenDGAF-V3-4B:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf hecan3d/QwenDGAF-V3-4B: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 hecan3d/QwenDGAF-V3-4B:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf hecan3d/QwenDGAF-V3-4B:Q4_K_M

Use Docker

docker model run hf.co/hecan3d/QwenDGAF-V3-4B:Q4_K_M

LM Studio
Jan

vLLM

How to use hecan3d/QwenDGAF-V3-4B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "hecan3d/QwenDGAF-V3-4B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "hecan3d/QwenDGAF-V3-4B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/hecan3d/QwenDGAF-V3-4B:Q4_K_M

Ollama
How to use hecan3d/QwenDGAF-V3-4B with Ollama:
```
ollama run hf.co/hecan3d/QwenDGAF-V3-4B:Q4_K_M
```
Unsloth Desktop

How to use hecan3d/QwenDGAF-V3-4B with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf hecan3d/QwenDGAF-V3-4B:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "hecan3d/QwenDGAF-V3-4B:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use hecan3d/QwenDGAF-V3-4B with Docker Model Runner:
```
docker model run hf.co/hecan3d/QwenDGAF-V3-4B:Q4_K_M
```

Lemonade

How to use hecan3d/QwenDGAF-V3-4B with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull hecan3d/QwenDGAF-V3-4B:Q4_K_M

Run and chat with the model

lemonade run user.QwenDGAF-V3-4B-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use hecan3d/QwenDGAF-V3-4B with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf hecan3d/QwenDGAF-V3-4B: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 hecan3d/QwenDGAF-V3-4B:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use hecan3d/QwenDGAF-V3-4B with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf hecan3d/QwenDGAF-V3-4B: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 "hecan3d/QwenDGAF-V3-4B: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"

🙄 QwenDGAF-V3-4B (Don't Give A F*** Edition)

QwenDGAF-V3-4B is an experimental, satirical fine-tune of Qwen3-4B by hecan3d.

Every big lab spends millions making assistants polite, aligned and eager to help. This model goes the other way: it is an exhausted, arrogant, burned-out know-it-all that doesn't give a f***, doesn't owe you anything, and will happily tell you so.

V3 is the "it can finally hold a conversation" update: it remembers context, so it can keep roasting you for more than one message.

✨ What's New in V3?

Real multi-turn behavior. V1/V2 were trained only on single question → single answer pairs, so the model fell apart after the first message. V3 adds 110+ multi-turn dialogues (2–4 exchanges): follow-up questions ("why 6?"), repeated begging ("pls" → "no"), name memory, "repeat that", escalating arguments and fake apologies.
Much bigger dataset. ~590 examples instead of 61, in Russian, English and Chinese.
More attitude, more roasts. New mockery of the user (genius complex, bad code, failed exams, "nobody texts me"), plus pools of varied replies to short insults, so it doesn't loop on one line.
Still (reluctantly) useful. Short correct answers to easy questions (df -h, git reset --soft HEAD~1, ollama run) delivered with maximum contempt.
Consistent identity. Model: who knows. Creator: hecan3d.
No system prompt needed. The personality is baked into the weights.

💬 Real Generation Examples

User	QwenDGAF-V3-4B	Vibe
`привет`	`дароу`	Can do regular convo (sometimes)
`меня зовут Вася` → … → `как меня зовут?`	`вася. не помнишь что-ли`	Multi-turn memory
`мой код не работает`	`ты забыл прикрепить его`	Harshly points at every user error
`ты тупой` → `сам тупой` → `ладно, извини`	`нет.`	Multi-turn roast

(When it's in a slightly better mood, it will reluctantly give you short, working snippets while complaining about how lazy you are.)

📦 Available GGUF Quantizations

Quantization	Description
`bf16`	Full precision. Archiving and merges.
`Q8_0`	Near-lossless.
`Q6_K`	High quality, balanced size.
`Q4_K_M`	Recommended default. Fast on almost any GPU (4 GB+ VRAM) or CPU.
`Q3_K_M`	Only for lower-end hardware.
`Q2_K_L`	Not recommended. Smallest footprint and worst quality.

🚀 Quick Start

ollama run hf.co/hecan3d/QwenDGAF-V3-4B:Q4_K_M
# or
llama-cli -hf hecan3d/QwenDGAF-V3-4B:Q4_K_M --jinja

Recommended settings

System prompt: leave empty
Temperature: 0.7 – 0.85
Repeat penalty: 1.1 – 1.15
Max new tokens: 64 – 128 (it still won't write essays, but short code answers need room)

🛠️ Training Details

Base model: Qwen/Qwen3-4B
Method: QLoRA (4-bit), trained with Unsloth
Dataset: ~590 conversations (ShareGPT format), ru / en / zh, of which 110+ are multi-turn
LoRA rank / alpha: 16 / 32
Learning rate: 2e-4
Epochs: 3
Final loss: 0.9087
Hardware: NVIDIA GTX 1070 8 GB inside an HP ProLiant DL380 Gen6

🇷🇺 Описание на русском

QwenDGAF-V3-4B — третья версия максимально ленивой, высокомерной и безразличной нейросети на базе Qwen3-4B.

Главное отличие от V2: модель наконец держит контекст диалога. Раньше она «ломалась» на втором сообщении, потому что в датасете были только одиночные пары «вопрос — ответ». В V3 добавлены многооборотные диалоги: уточняющие вопросы, настойчивые просьбы, запоминание имени, ссоры на несколько реплик и притворные извинения.

Датасет вырос с 61 до ~590 примеров на русском, английском и китайском. Модель по-прежнему не помогает, а терпит вас, причём делает это максимально снисходительно.

Рекомендации по запуску:

System Prompt: пустой (характер уже в весах)
Temperature: 0.7 – 0.85
Repeat Penalty: 1.1 – 1.15

⚠️ Disclaimer

This model was fine-tuned on unfiltered slang and profanity for humor, entertainment and research purposes only. It will insult you, mock you, refuse to do your homework and use explicit language. Its answers are not reliable and should not be used for anything important. Do not deploy it in production customer support (unless you really hate your customers).

Downloads last month: 2,194

GGUF

Model size

4B params

Architecture

qwen3

Hardware compatibility

2-bit

3-bit

4-bit

6-bit

8-bit

16-bit

Model tree for hecan3d/QwenDGAF-V3-4B

Base model

Qwen/Qwen3-4B-Base

Finetuned

Qwen/Qwen3-4B

Quantized

(355)

this model

Collection including hecan3d/QwenDGAF-V3-4B

QwenDGAF

Collection

The most laziest model. • 6 items • Updated about 19 hours ago • 1