Instructions to use Mungert/DMind-3-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Mungert/DMind-3-GGUF with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Mungert/DMind-3-GGUF")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("Mungert/DMind-3-GGUF", dtype="auto")

llama-cpp-python

How to use Mungert/DMind-3-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="Mungert/DMind-3-GGUF",
	filename="DMind-3-bf16.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use Mungert/DMind-3-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf Mungert/DMind-3-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf Mungert/DMind-3-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf Mungert/DMind-3-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf Mungert/DMind-3-GGUF: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 Mungert/DMind-3-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf Mungert/DMind-3-GGUF: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 Mungert/DMind-3-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Mungert/DMind-3-GGUF:Q4_K_M

Use Docker

docker model run hf.co/Mungert/DMind-3-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use Mungert/DMind-3-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Mungert/DMind-3-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Mungert/DMind-3-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Mungert/DMind-3-GGUF:Q4_K_M

SGLang

How to use Mungert/DMind-3-GGUF with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Mungert/DMind-3-GGUF" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Mungert/DMind-3-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Mungert/DMind-3-GGUF" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Mungert/DMind-3-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use Mungert/DMind-3-GGUF with Ollama:
```
ollama run hf.co/Mungert/DMind-3-GGUF:Q4_K_M
```

Unsloth Studio new

How to use Mungert/DMind-3-GGUF 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 Mungert/DMind-3-GGUF 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 Mungert/DMind-3-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for Mungert/DMind-3-GGUF to start chatting

Pi new

How to use Mungert/DMind-3-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf Mungert/DMind-3-GGUF: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": "Mungert/DMind-3-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use Mungert/DMind-3-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf Mungert/DMind-3-GGUF: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 Mungert/DMind-3-GGUF:Q4_K_M

Run Hermes

hermes

Docker Model Runner
How to use Mungert/DMind-3-GGUF with Docker Model Runner:
```
docker model run hf.co/Mungert/DMind-3-GGUF:Q4_K_M
```

Lemonade

How to use Mungert/DMind-3-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Mungert/DMind-3-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.DMind-3-GGUF-Q4_K_M

List all available models

lemonade list

DMind-3 GGUF Models

Model Generation Details

This model was generated using llama.cpp at commit 05fa625ea.

Quantization Beyond the IMatrix

I've been experimenting with a new quantization approach that selectively elevates the precision of key layers beyond what the default IMatrix configuration provides.

In my testing, standard IMatrix quantization underperforms at lower bit depths, especially with Mixture of Experts (MoE) models. To address this, I'm using the --tensor-type option in llama.cpp to manually "bump" important layers to higher precision. You can see the implementation here:
👉 Layer bumping with llama.cpp

While this does increase model file size, it significantly improves precision for a given quantization level.

I'd love your feedback—have you tried this? How does it perform for you?

Click here to get info on choosing the right GGUF model format

🔮 DMind-3: The Age of Foresight

From local logic to global foresight. In a world of isolated systems, the one who sees the whole board wins.

We have armed the individual with a shield (DMind-3-nano) and a brain (DMind-3-mini). We have enabled sovereign intelligence to defend and to reason. Yet, in the interconnected chaos of global markets, even the sharpest mind can be blindsided by a tsunami forming on the other side of the world. Local optimization is not enough. True sovereignty requires not just reaction, but pre-emption.

Web3 is a single, planet-scale financial machine. Capital flows like weather patterns, and risks cascade across protocols like lightning. To navigate this reality, one cannot merely analyze a single smart contract or a single chain. One must perceive the entire system—its flows, its pressures, its emergent properties. This requires a perspective that transcends the local, a form of intelligence that can synthesize global, cross-domain information into actionable foresight.

DMind-3 is our answer. It is not an incremental upgrade; it is a categorical leap. While nano provides intuition and mini provides logic, max delivers foresight. It is the Oracle in the cloud, the strategic command center that sees the entire battlefield. It was built not to answer questions, but to question the answers, to model the unseen, and to chart a course through the complexity of a new financial era.

🛡️ DMind-3-nano is your Shield. ⚔️ DMind-3-mini is your Spear. 🔮 DMind-3 is your Oracle.

Welcome to the Age of Foresight.

🏛️ DMind-3: The Macro-Strategic Financial Engine

1. Evolution & Legacy

The DMind-3 series was conceived as a complete, multi-layered cognitive architecture for the sovereign individual. Nano secures the present transaction. Mini formulates the immediate strategy. Max defines the long-term campaign.

This final piece of the trilogy moves beyond the tactical and into the strategic. It was born from the recognition that the most significant opportunities and the most devastating risks in Web3 are systemic. They are not found in code, but in the interplay between code, capital, and human psychology at a global scale. DMind-3 is engineered to be a Macro-Strategic Financial Engine, providing institutional-grade foresight as a utility for developers, funds, and the agent ecosystems built upon the DMind stack.

2. ⚙️ Model Details

Property	Value
Model Name	DMind-3
Organization	DMindAI
Base Architecture	gpt-oss-20b (Customized Transformer w/ Multi-Scale RoPE)
Parameter Count	21 Billion
Precision	BF16 / FP16 (Native)
Context Window	256k tokens
Deployment	Cloud API & Private Enterprise VPC

3. 🔬 Methodology: Hierarchical Predictive Synthesis (HPS)

DMind-3 introduces Hierarchical Predictive Synthesis (HPS). While C³-SFT (used in mini) teaches the model to correct its own reasoning, HPS teaches it to synthesize multiple, conflicting, time-variant data streams into a coherent probabilistic forecast. It operates on a nested hierarchy of abstraction, from raw on-chain events to complex macroeconomic indicators.

(Figure 1: The HPS training paradigm, showing multi-level data fusion and probabilistic future state generation)

Mathematical Formalization

The HPS objective function seeks to minimize the divergence between the model's predicted distribution of future states and the actual observed outcomes, weighted by strategic importance:

$\mathcal{L}_{\text{HPS}}(\theta) = - \mathbb{E}_{\mathcal{D}} \left[ \sum_{t=1}^{T} \sum_{i=1}^{M} \omega_i \cdot \log P_\theta(S'_{t+1} \mid S_t, A_t, M_i) \right] + \lambda \sum_{l=1}^{L} \| \Omega_l(\theta) - \Omega_l(\theta_{\text{ref}}) \|_F$

where:

Symbol	Description
$S_{t}$	The state of the global market at time $t$
$A_{t}$	The set of all actions (transactions, governance votes) at time $t$
$M_{i}$	The $i$ -th modality of data (e.g., on-chain, news, social sentiment)
$\omega_i$	The attention weight assigned to the strategic importance of modality $i$
$\Omega_l$	The parameter matrix at layer $l$ of the network, regularized to prevent catastrophic forgetting

Dual-State Inference Mechanism

Similar to DMind-3-mini, the model supports a dual-state inference mechanism triggered by a special token:

$\hat{y} = \begin{cases} \operatorname*{arg\,max}\limits_{y} P_\theta(y \mid x, \mathcal{C}_{\text{global}}) & \text{if } \tau = \emptyset \quad (\text{Standard Mode}) \\ \operatorname*{arg\,max}\limits_{y} P_\theta(y \mid x, \mathcal{C}_{\text{global}}, \mathcal{R}_{\text{risk}}, \mathcal{H}_{\text{hist}}) & \text{if } \tau = \texttt{<FORESIGHT>} \quad (\text{Strategic Mode}) \end{cases}$

This forces the model to not just predict, but to weigh the importance of different data sources when constructing its view of the future.

4. 💡 Intended Use: Institutional-Grade Web3 Intelligence

DMind-3 is designed to power the next generation of DeFi analytics, risk management platforms, and autonomous agent orchestrators.

Key Capabilities:

🔮 Macro-Strategic Foresight: Identify emerging cross-chain narratives, predict market regime shifts, and model the impact of major economic events on crypto asset correlations.
🏛️ Automated Institutional Research: Generate deep, data-driven reports on novel protocols, perform automated tokenomics valuation, and assess long-term protocol viability.
🌊 Systemic Risk Assessment: Model contagion risk across DeFi, detect liquidity black holes before they form, and run stress tests on entire ecosystems based on simulated market shocks.
🤖 Agent Fleet Orchestration: Serve as the central "strategic brain" for fleets of mini and nano agents, providing high-level directives and market context.

5. 📚 The Brain, Shield & Oracle Ecosystem

The DMind-3 series is a vertically integrated stack designed for sovereign intelligence.

(Figure 2: The full DMind-3 Cognitive Architecture, from on-device reflexes to cloud-native foresight)

The Oracle (DMind-3): Runs in the cloud. Provides macro-strategic foresight, systemic risk analysis, and orchestrates the agent fleet.
The Brain (DMind-3-mini): Runs on your local high-performance machine. Executes complex, bespoke strategies and performs deep, focused research under the Oracle's guidance.
The Shield (DMind-3-nano): Runs in your browser or wallet. Provides real-time, intuitive transaction security and intent recognition, acting as the final line of defense.

6. 📚 Training Data

DMind-3 was trained on a corpus of over 500,000 curated, high-signal documents and a multi-terabyte stream of structured on-chain data.

Data Source	Proportion	Description
Institutional Alpha Reports	35%	Comprehensive reports from premier crypto-native funds and TradFi institutions, deconstructed into causal models.
Global Macroeconomic Data	25%	Time-series data from sources like the Federal Reserve (FRED), World Bank, and IMF, correlated with on-chain metrics.
Cross-Chain Indexed Data	20%	A complete, indexed history of transactions, state changes, and logs across all major EVM chains, Solana, and Cosmos ecosystems.
Financial Post-Mortems & Audits	10%	In-depth analysis of systemic failures, economic exploits, and protocol hacks, focusing on pre-mortem indicators and contagion pathways.
Geopolitical & Regulatory Feeds	10%	Real-time feeds on global regulatory changes, policy proposals, and geopolitical events impacting digital asset markets.

7. 🏆 Performance Benchmarks

Evaluated on three key benchmarks: DMind Benchmark (Web3 Native Logic), FinanceQA (Financial Domain Knowledge), and AIME 2025 (Advanced Mathematical Reasoning).

(Figure 3: LLM Performance Evaluation - 3 Benchmarks: DMind Benchmark, FinanceQA, AIME 2025)

The evaluation compares DMind-3 (21B) against top-tier frontier models (GPT-5.1, Claude Sonnet 4.5) and other efficient models. Despite its optimized size, the Max model demonstrates exceptional efficiency, particularly in specialized domain tasks where it outperforms significantly larger generalist models.

8. ⚖️ Limitations & Disclaimer

Not a Financial Advisor (NFA): DMind-3 is a powerful analytical tool for generating insights and modeling risks. It is not a registered financial advisor. All outputs should be independently verified and are not a solicitation to trade.
Probabilistic Nature: All forecasts are probabilistic and based on the data available up to the knowledge cutoff. The model cannot predict black swan events and is subject to the inherent unpredictability of markets.
Knowledge Cutoff: The core model has a knowledge cutoff of June 2025. While it can process real-time data provided via the API, its foundational understanding is based on its training corpus.

🚀 If you find these models useful

Help me test my AI-Powered Quantum Network Monitor Assistant with quantum-ready security checks:

👉 Quantum Network Monitor

The full Open Source Code for the Quantum Network Monitor Service available at my github repos ( repos with NetworkMonitor in the name) : Source Code Quantum Network Monitor. You will also find the code I use to quantize the models if you want to do it yourself GGUFModelBuilder

💬 How to test:
Choose an AI assistant type:

TurboLLM (GPT-4.1-mini)
HugLLM (Hugginface Open-source models)
TestLLM (Experimental CPU-only)

What I’m Testing

I’m pushing the limits of small open-source models for AI network monitoring, specifically:

Function calling against live network services
How small can a model go while still handling:
- Automated Nmap security scans
- Quantum-readiness checks
- Network Monitoring tasks

🟡 TestLLM – Current experimental model (llama.cpp on 2 CPU threads on huggingface docker space):

✅ Zero-configuration setup
⏳ 30s load time (slow inference but no API costs) . No token limited as the cost is low.
🔧 Help wanted! If you’re into edge-device AI, let’s collaborate!

Other Assistants

🟢 TurboLLM – Uses gpt-4.1-mini :

**It performs very well but unfortunatly OpenAI charges per token. For this reason tokens usage is limited.
Create custom cmd processors to run .net code on Quantum Network Monitor Agents
Real-time network diagnostics and monitoring
Security Audits
Penetration testing (Nmap/Metasploit)

🔵 HugLLM – Latest Open-source models:

🌐 Runs on Hugging Face Inference API. Performs pretty well using the lastest models hosted on Novita.

💡 Example commands you could test:

"Give me info on my websites SSL certificate"
"Check if my server is using quantum safe encyption for communication"
"Run a comprehensive security audit on my server"
'"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code on. This is a very flexible and powerful feature. Use with caution!

Final Word

I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is open source. Feel free to use whatever you find helpful.

If you appreciate the work, please consider buying me a coffee ☕. Your support helps cover service costs and allows me to raise token limits for everyone.

I'm also open to job opportunities or sponsorship.

Thank you! 😊

Downloads last month: 302

GGUF

Model size

21B params

Architecture

gpt-oss

Hardware compatibility

1-bit

2-bit

3-bit

4-bit

5-bit

6-bit

8-bit

16-bit

View +1 variant

Model tree for Mungert/DMind-3-GGUF

Base model

openai/gpt-oss-20b

Quantized

(208)

this model