AgentDoG-Qwen2.5-7B Security Agent

A specialized 7B parameter AI agent model optimized for security, safety, and agentic workflows. This model is based on AI45Research/AgentDoG-Qwen2.5-7B and provided in GGUF format with weighted/imatrix quantization for efficient local inference.

Overview

AgentDoG-Qwen2.5-7B is designed for conversational AI applications with a focus on security and safety-critical tasks. The model is available in multiple quantization levels (from 2GB to 6.4GB) allowing you to balance quality, speed, and resource requirements based on your hardware.

Instructions

When using this model, follow these steps:

1. **Select appropriate quantization level** based on your hardware and quality requirements:

- **Recommended for most users**: Q4_K_M (4.8GB) - fast with good quality

- **Optimal balance**: Q4_K_S (4.6GB) - best size/speed/quality ratio

- **Higher quality**: Q5_K_M (5.5GB) or Q6_K (6.4GB) - for better results with more resources

- **Lower resource**: IQ3_M (3.7GB) or IQ4_XS (4.3GB) - for constrained environments

- **Minimal size**: IQ2_S (2.7GB) or higher IQ2 variants - only if extremely resource-limited

2. **Download the model** from HuggingFace:

```bash

# Example for Q4_K_M (recommended)

wget https://huggingface.co/mradermacher/AgentDoG-Qwen2.5-7B-i1-GGUF/resolve/main/AgentDoG-Qwen2.5-7B.i1-Q4_K_M.gguf

```

3. **Load the model** in your preferred GGUF-compatible runtime:

**llama.cpp:**

```bash

./main -m AgentDoG-Qwen2.5-7B.i1-Q4_K_M.gguf -p "Your prompt here" -n 512

```

**Ollama:**

```bash

# Create Modelfile

echo "FROM ./AgentDoG-Qwen2.5-7B.i1-Q4_K_M.gguf" > Modelfile

ollama create agentdog -f Modelfile

ollama run agentdog

```

**LM Studio:** Import the .gguf file through the GUI

4. **Configure for security tasks** by providing context-appropriate system prompts emphasizing:

- Security analysis and threat detection

- Safety-critical decision making

- Agentic behavior with proper guardrails

- Conversational interaction with security focus

5. **Optimize performance** based on your use case:

- Adjust context window size (`-c` flag in llama.cpp)

- Configure temperature for deterministic (0.1-0.3) or creative (0.7-0.9) outputs

- Set appropriate token limits for responses

- Enable GPU acceleration if available (`-ngl` flag)

Usage Examples

**Security Analysis:**

```

Analyze the following code for potential security vulnerabilities:

[paste code here]

```

**Threat Assessment:**

```

Given this network log, identify any suspicious patterns or potential security threats:

[paste logs]

```

**Safety Evaluation:**

```

Review this system configuration for safety compliance and identify risks:

[paste configuration]

```

Quantization Reference

Constraints

Additional Resources