Mungert/DMind-1-GGUF

DMind-1 GGUF Models

Model Generation Details

This model was generated using llama.cpp at commit 7f37b6cf.

Ultra-Low-Bit Quantization with IQ-DynamicGate (1-2 bit)

Our latest quantization method introduces precision-adaptive quantization for ultra-low-bit models (1-2 bit), with benchmark-proven improvements on Llama-3-8B. This approach uses layer-specific strategies to preserve accuracy while maintaining extreme memory efficiency.

Benchmark Context

All tests conducted on Llama-3-8B-Instruct using:

• Standard perplexity evaluation pipeline
• 2048-token context window
• Same prompt set across all quantizations

Method

• Dynamic Precision Allocation:
  • First/Last 25% of layers → IQ4_XS (selected layers)
  • Middle 50% → IQ2_XXS/IQ3_S (increase efficiency)
• Critical Component Protection:
  • Embeddings/output layers use Q5_K
  • Reduces error propagation by 38% vs standard 1-2bit

Quantization Performance Comparison (Llama-3-8B)

Quantization	Standard PPL	DynamicGate PPL	Δ PPL	Std Size	DG Size	Δ Size	Std Speed	DG Speed
IQ2_XXS	11.30	9.84	-12.9%	2.5G	2.6G	+0.1G	234s	246s
IQ2_XS	11.72	11.63	-0.8%	2.7G	2.8G	+0.1G	242s	246s
IQ2_S	14.31	9.02	-36.9%	2.7G	2.9G	+0.2G	238s	244s
IQ1_M	27.46	15.41	-43.9%	2.2G	2.5G	+0.3G	206s	212s
IQ1_S	53.07	32.00	-39.7%	2.1G	2.4G	+0.3G	184s	209s

Key:

• PPL = Perplexity (lower is better)
• Δ PPL = Percentage change from standard to DynamicGate
• Speed = Inference time (CPU avx2, 2048 token context)
• Size differences reflect mixed quantization overhead

Key Improvements:

• 🔥 IQ1_M shows massive 43.9% perplexity reduction (27.46 → 15.41)
• 🚀 IQ2_S cuts perplexity by 36.9% while adding only 0.2GB
• ⚡ IQ1_S maintains 39.7% better accuracy despite 1-bit quantization

Tradeoffs:

• All variants have modest size increases (0.1-0.3GB)
• Inference speeds remain comparable (<5% difference)

When to Use These Models

📌 Fitting models into GPU VRAM

✔ Memory-constrained deployments

✔ Cpu and Edge Devices where 1-2bit errors can be tolerated

✔ Research into ultra-low-bit quantization

Choosing the Right Model Format

Selecting the correct model format depends on your hardware capabilities and memory constraints.

BF16 (Brain Float 16) – Use if BF16 acceleration is available

• A 16-bit floating-point format designed for faster computation while retaining good precision.
• Provides similar dynamic range as FP32 but with lower memory usage.
• Recommended if your hardware supports BF16 acceleration (check your device's specs).
• Ideal for high-performance inference with reduced memory footprint compared to FP32.

📌 Use BF16 if:
✔ Your hardware has native BF16 support (e.g., newer GPUs, TPUs).
✔ You want higher precision while saving memory.
✔ You plan to requantize the model into another format.

📌 Avoid BF16 if:
❌ Your hardware does not support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.

---

F16 (Float 16) – More widely supported than BF16

• A 16-bit floating-point high precision but with less of range of values than BF16.
• Works on most devices with FP16 acceleration support (including many GPUs and some CPUs).
• Slightly lower numerical precision than BF16 but generally sufficient for inference.

📌 Use F16 if:
✔ Your hardware supports FP16 but not BF16.
✔ You need a balance between speed, memory usage, and accuracy.
✔ You are running on a GPU or another device optimized for FP16 computations.

📌 Avoid F16 if:
❌ Your device lacks native FP16 support (it may run slower than expected).
❌ You have memory limitations.

---

Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference

Quantization reduces model size and memory usage while maintaining as much accuracy as possible.

• Lower-bit models (Q4_K) → Best for minimal memory usage, may have lower precision.
• Higher-bit models (Q6_K, Q8_0) → Better accuracy, requires more memory.

📌 Use Quantized Models if:
✔ You are running inference on a CPU and need an optimized model.
✔ Your device has low VRAM and cannot load full-precision models.
✔ You want to reduce memory footprint while keeping reasonable accuracy.

📌 Avoid Quantized Models if:
❌ You need maximum accuracy (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).

---

Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)

These models are optimized for extreme memory efficiency, making them ideal for low-power devices or large-scale deployments where memory is a critical constraint.

• IQ3_XS: Ultra-low-bit quantization (3-bit) with extreme memory efficiency.

  • Use case: Best for ultra-low-memory devices where even Q4_K is too large.
  • Trade-off: Lower accuracy compared to higher-bit quantizations.

• IQ3_S: Small block size for maximum memory efficiency.

  • Use case: Best for low-memory devices where IQ3_XS is too aggressive.

• IQ3_M: Medium block size for better accuracy than IQ3_S.

  • Use case: Suitable for low-memory devices where IQ3_S is too limiting.

• Q4_K: 4-bit quantization with block-wise optimization for better accuracy.

  • Use case: Best for low-memory devices where Q6_K is too large.

• Q4_0: Pure 4-bit quantization, optimized for ARM devices.

  • Use case: Best for ARM-based devices or low-memory environments.

---

Summary Table: Model Format Selection

Model Format	Precision	Memory Usage	Device Requirements	Best Use Case
BF16	Highest	High	BF16-supported GPU/CPUs	High-speed inference with reduced memory
F16	High	High	FP16-supported devices	GPU inference when BF16 isn't available
Q4_K	Medium Low	Low	CPU or Low-VRAM devices	Best for memory-constrained environments
Q6_K	Medium	Moderate	CPU with more memory	Better accuracy while still being quantized
Q8_0	High	Moderate	CPU or GPU with enough VRAM	Best accuracy among quantized models
IQ3_XS	Very Low	Very Low	Ultra-low-memory devices	Extreme memory efficiency and low accuracy
Q4_0	Low	Low	ARM or low-memory devices	llama.cpp can optimize for ARM devices

---

Included Files & Details

DMind-1-bf16.gguf

• Model weights preserved in BF16.
• Use this if you want to requantize the model into a different format.
• Best if your device supports BF16 acceleration.

DMind-1-f16.gguf

• Model weights stored in F16.
• Use if your device supports FP16, especially if BF16 is not available.

DMind-1-bf16-q8_0.gguf

• Output & embeddings remain in BF16.
• All other layers quantized to Q8_0.
• Use if your device supports BF16 and you want a quantized version.

DMind-1-f16-q8_0.gguf

• Output & embeddings remain in F16.
• All other layers quantized to Q8_0.

DMind-1-q4_k.gguf

• Output & embeddings quantized to Q8_0.
• All other layers quantized to Q4_K.
• Good for CPU inference with limited memory.

DMind-1-q4_k_s.gguf

• Smallest Q4_K variant, using less memory at the cost of accuracy.
• Best for very low-memory setups.

DMind-1-q6_k.gguf

• Output & embeddings quantized to Q8_0.
• All other layers quantized to Q6_K .

DMind-1-q8_0.gguf

• Fully Q8 quantized model for better accuracy.
• Requires more memory but offers higher precision.

DMind-1-iq3_xs.gguf

• IQ3_XS quantization, optimized for extreme memory efficiency.
• Best for ultra-low-memory devices.

DMind-1-iq3_m.gguf

• IQ3_M quantization, offering a medium block size for better accuracy.
• Suitable for low-memory devices.

DMind-1-q4_0.gguf

• Pure Q4_0 quantization, optimized for ARM devices.
• Best for low-memory environments.
• Prefer IQ4_NL for better accuracy.

🚀 If you find these models useful

❤ Please click "Like" if you find this useful!
Help me test my AI-Powered Network Monitor Assistant with quantum-ready security checks:
👉 Quantum Network Monitor

💬 How to test:
Choose an AI assistant type:

• TurboLLM (GPT-4o-mini)
• HugLLM (Hugginface Open-source)
• 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 scans
  • Quantum-readiness checks
  • Network Monitoring tasks

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

• ✅ Zero-configuration setup
• ⏳ 30s load time (slow inference but no API costs)
• 🔧 Help wanted! If you’re into edge-device AI, let’s collaborate!

Other Assistants

🟢 TurboLLM – Uses gpt-4o-mini for:

• 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)
• 🔑 Get more tokens by logging in or downloading our Quantum Network Monitor Agent with integrated AI Assistant

🔵 HugLLM – Latest Open-source models:

• 🌐 Runs on Hugging Face Inference API

💡 Example commands to you could test:

1. "Give me info on my websites SSL certificate"
2. "Check if my server is using quantum safe encyption for communication"
3. "Run a comprehensive security audit on my server"
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. 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! 😊

---

Table of Contents

• Introduction
• 1. Model Overview
• 2. Evaluation Results
• 3. Use Cases
• 4. Quickstart
  • 4.1 Model Downloads
  • 4.2 OpenRouter API
  • 4.3 OpenRouter Web Chat
• License
• Contact

Introduction

The rapid growth of Web3 technologies—blockchain, DeFi, and smart contracts—demands specialized AI large language models (LLMs) with precise domain alignment and advanced reasoning capabilities. However, General-purpose LLMs often lack the domain-specific accuracy, nuanced reasoning, and instruction-following aligned with expert expectations.

To address these limitations, we introduce DMind-1, a domain-specialized LLM fine-tuned for the Web3 ecosystem via supervised instruction tuning and reinforcement learning from human feedback (RLHF). Built on a powerful base model, DMind-1 achieves strong improvements in task accuracy, content safety, and expert-aligned interaction, significantly surpassing general-purpose models. DMind-1 represents a robust foundation for intelligent agents in the Web3 ecosystem.

1. Model Overview

DMind-1

DMind-1 is a specialized Web3 expert model built on the Qwen3-32B base. Leveraging a state-of-the-art transformer architecture, it integrates deep domain knowledge through a novel two-stage fine-tuning pipeline, establishing its distinctive strengths in Web3-specific applications.

Key Points:

• Comprehensive Domain Expertise Data: In the first stage, DMind-1 underwent Supervised Fine-Tuning (SFT) on 13,276 expert-curated knowledge items distilled from 32.7GB of Web3 documentation, covering 8 key subdomains including DeFi, tokenomics, governance, and smart contracts. These data points were extracted and structured by a team of domain experts to ensure both depth and accuracy. To enable efficient and scalable training, we employed Low-Rank Adaptation (LoRA) during the SFT stage, allowing DMind-1 to internalize specialized Web3 knowledge while preserving the general-language capabilities of its base model.

• Reinforcement Learning from Human Feedback (RLHF) To further align the model with expert expectations in realistic interaction scenarios and accuracy, we implemented an RLHF phase composed of:

  • Reward Model Training: We trained a domain-specific reward model using preference-ranked outputs collected from human experts across diverse Web3-specific question-answer and interaction scenarios. This model learned to assess which responses best reflect factual accuracy and expert-level reasoning in the Web3 domain.
  • Policy Optimization with PPO: Building on the SFT model, we fine-tuned Qwen3-32B using Proximal Policy Optimization (PPO), guided by the trained reward model. The policy network was optimized based on feedback from simulated Web3 dialogue environments, while LoRA ensured resource-efficient parameter updates and significantly reduced compute and memory requirements. This dual-stage approach enabled efficient fine-tuning of a larger model on Web3-specific tasks while achieving high alignment with human intent.

• Domain-Aligned Reasoning and Interaction: DMind-1 exhibits advanced web3-aligned reasoning and interactive capabilities in the following fields:

  • Natural Dialogue Fluency: Coherent, context-aware conversations on complex Web3 topics, with strong multi-turn consistency.

  • Complex Instruction Following: Reliable execution of multi-step instructions and conditional logic, supporting agent-driven workflows.

  • Safe and Compliant Content Generation: Outputs are aligned with domain-specific safety, ethics, and regulatory standards.

2. Evaluation Results

We evaluate DMind-1 and DMind-1-mini using the DMind Benchmark, a domain-specific evaluation suite designed to assess large language models in the Web3 context. The benchmark includes 1,917 expert-reviewed questions across nine core domain categories, and it features both multiple-choice and open-ended tasks to measure factual knowledge, contextual reasoning, and other abilities.

To complement accuracy metrics, we conducted a cost-performance analysis by comparing benchmark scores against publicly available input token prices across 24 leading LLMs. In this evaluation:

• DMind-1 achieved the highest Web3 score while maintaining one of the lowest token input costs among top-tier models such as Grok 3 and Claude 3.7 Sonnet.

• DMind-1-mini ranked second, retaining over 95% of DMind-1’s performance with greater efficiency in latency and compute.

Both models are uniquely positioned in the most favorable region of the score vs. price curve, delivering state-of-the-art Web3 reasoning at significantly lower cost. This balance of quality and efficiency makes the DMind models highly competitive for both research and production use.

3. Use Cases

• Expert-Level Question & Answering: Provides accurate, context-aware answers on blockchain, DeFi, smart contracts, and related Web3 topics.
• Compliance-Aware Support: Assists in drafting or reviewing content within regulatory and legal contexts.
• Content Generation in Domain: Produces Web3-specific blog posts, documentation, and tutorials tailored to developers and users.
• DeFi Strategy Suggestions: Generates insights and recommendations for yield farming, liquidity provision, and portfolio strategies based on user-provided data.
• Risk Management: Suggests strategies aligned with user risk profiles for more informed decision-making in volatile markets.

4. Quickstart

4.1 Model Downloads

Model	Base Model	Download
DMind-1	Qwen3-32B	Hugging Face Link
DMind-1-mini	Qwen3-14B	Hugging Face Link

4.2 OpenRouter API (Coming Soon)

Documentation for API access will be available soon.

4.3 OpenRouter Web Chat (Coming Soon)

Web chat interface documentation will be available soon.

License

• The code repository and model weights for DMind-1 is released under the MIT License.
• Commercial use, modification, and derivative works (including distillation and fine-tuning) are permitted.
• Base Models:
  • DMind-1 is derived from Qwen3-32B, originally licensed under the Qwen License.
  • Please ensure compliance with the original base model licenses when using or distributing derivatives.

Contact

For questions or support, please contact [email protected]