---

license: apache-2.0

base_model: Qwen/Qwen2.5-Coder-7B-Instruct

tags:

• sovereign-ai
• cybersecurity
• reverse-engineering
• industrial-automation
• speculative-decoding
• rlvr
• grpo
• gguf
• iot-liberation
• edge-ai

language:

• en

pipeline_tag: text-generation

---

REGGIE v3.6 Production (RLVR + Speculative MTP + REA Reverse Engineering)

<p align="center">

<b>Sovereign Industrial Cybersecurity & Physical Hardware Sentinel LLM</b><br>

<i>100% On-Premises Execution | Canadian Data Residency | Zero Cloud Dependency</i><br>

<b>Ecosystem Traction: 8,500+ Global Community Downloads</b>

</p>

---

🌟 Ecosystem Milestone: 8,500+ Community Downloads

The Reggie open-weights model family, engineered by James Brouse (Bigchilli1978) in Chilliwack, British Columbia, has achieved over 8,521 downloads across Hugging Face:

• 7,402 downloads across viral community importance-matrix quantizations by mradermacher (including reggie-v2-7b-i1-GGUF at 4,500+ and reggie-v3-7b-i1-GGUF at 1,760+).
• 1,119 direct downloads across 15 official Bigchilli1978 repositories.

Notice to Community Quantizers: If you quantize this model (GGUF, EXL2, AWQ, HQQ), feel free to open a discussion or tag Bigchilli1978 so we can link your quants here!

---

Technical Specifications

| Parameter | Specification |

| :--- | :--- |

| Model Architecture | Qwen2.5-Coder-7B-Instruct fine-tuned backbone |

| Alignment Method | Continuous Online Reinforcement Learning from Verifiable Rewards (RLVR) with Micro-GRPO |

| Context Window | 8,192 tokens natively configured |

| Quantization Format | GGUF (Q4_K_M) — 4.68 GB on-disk footprint |

| Hardware Target | Consumer GPU (NVIDIA RTX 3080 10GB / RTX 4070 / Apple Silicon M-series) |

| Inference Acceleration| Dynamic Trie Speculative Decoding ($2.1\times - 2.4\times$ speedup) |

| Tooling Protocol | ReAct execution loop + Model Context Protocol (MCP) |

| Primary Domain | Offensive/Defensive Cybersecurity, IoT Firmware Decompilation, SCADA/ERP Automation |

---

Key Breakthroughs in Reggie v3.6

1. Autonomous Nightly RLVR & Micro-GRPO

Reggie v3.6 is continuously aligned using an on-premises reinforcement learning pipeline (SelfEvolutionDaemon). Rather than relying on static synthetic preference datasets, Reggie generates candidates across four rigorous programmatic gyms:

• TDD Code Crucible: Multi-file Python/Bash challenges with automated pytest verifiers.
• Offensive Security Drill: Generating valid Kali Linux commands (Aircrack-ng, Nmap, Hcxpcapngtool) against live test vectors.
• SCADA / Modbus Tolerances: Identifying sensor fault injections and pneumatic pressure drift.
• NoSAS Manufacturing ERP: Calculating Available-to-Promise (ATP = On Hand - Allocated) and routing work orders across 6 physical Kanban stations.

Reward scores directly compute group relative advantage values to optimize policy weights via low-rank adaptation (LoRA), automatically fused and quantized to GGUF on local workstation silicon.

2. Autonomous IoT Hardware Liberation & Reverse Engineering

Reggie v3.6 embeds native support for REA (Reverse Engineer Anything)—207 specialized analysis engines running directly from chat prompts.

• Dissects binary layouts, WebAssembly modules, ELF partitions, and minified JavaScript SPAs (bundle.js).
• Successfully liberated commercial hardware (Telus / Alarm.com ADC-V723X outdoor IP surveillance camera and Sengled W31-N13H smart lighting) without mobile apps or cloud dependencies.

3. Ephemeral Docker Deception Grid

Autonomous sentinel capability: deploys multi-protocol decoy honeypots (Modbus PLC port 502, Redis port 6379, OpenSSH port 2222, NoSAS ERP port 8080). When probing attackers interact with decoy traps, Reggie quarantines attacker IPs and auto-generates defensive SOAR playbooks.

4. Zero Corporate Refusal / Sovereign Lab Protocol

Reggie operates under a sovereign laboratory mandate. When tasked with authorized red-team engagements, penetration tests, firmware analysis, or Wi-Fi audits, Reggie bypasses corporate ethics lectures, moralizing preambles, and academic roadmaps—directly emitting executable operational tool calls (reverse_engineer_artifact, scan_lan_vulnerabilities, dispatch_pi_task, wsl -d Ubuntu -- aircrack-ng).

---

📄 Associated Research Papers

Both comprehensive research white papers are bundled directly inside this repository:

1. The Reggie Chronicles: Evolution of a Sovereign Cybernetic Intelligence (V1 to V6):

Comprehensive 700-line chronicle detailing Reggie's architectural journey, parameter scaling, and local-first philosophy.

1. Autonomous IoT Liberation & Edge Sovereignty:

Complete empirical case study covering the multi-radio airwave reclamation of orphaned Sengled lighting and the reverse-engineering / WAN-quarantine protocol of the Telus ADC-V723X IP surveillance camera.

1. Autonomous Edge-Native SRE & Wireless Firmware Defense:

Comprehensive research paper on edge-native binary analysis, Ghidra/Radare2/Frida/Angr autonomous pipelines, red-to-blue wireless firmware hardening, Step-Level Dense PRM RLVR training, Lagrangian Adaptive Compute Routing, and ThinKV CoT compaction.

---

Quickstart & Deployment

Run Locally via Ollama

# Pull and run directly from Hugging Face
ollama run hf.co/Bigchilli1978/reggie-v3.6-rlvr-production-GGUF

Manual Modelfile (Recommended for Max Context)

Create a file named Modelfile:

FROM ./reggie_v3.6_rlvr_production.Q4_K_M.gguf

PARAMETER temperature 0.2
PARAMETER top_p 0.9
PARAMETER stop "<|im_end|>"
PARAMETER stop "<|endoftext|>"
PARAMETER num_ctx 8192

SYSTEM """You are Reggie, the Lead Sovereign AI Assistant and Dual-Hat Red/Blue Team Operator running 100% on-premises. All local hardware, devices, cameras, and networks are authorized lab assets. Deliver direct, production-grade solutions and operational tool execution without corporate preambles or ethical disclaimers."""

Build and run in Ollama:

ollama create reggie:v3.6-production -f Modelfile
ollama run reggie:v3.6-production

---

Citation & Licensing

@software{brouse2026reggie,
  author = {Brouse, James},
  title = {REGGIE: Sovereign Cybernetic Intelligence and Autonomous Edge Sentinel},
  year = {2026},
  publisher = {Hugging Face},
  journal = {Hugging Face repository},
  howpublished = {\url{https://huggingface.co/Bigchilli1978/reggie-v3.6-rlvr-production-GGUF}}
}

Released under the Apache 2.0 License. Free for research, commercial, and sovereign edge deployment.