--- base_model: unsloth/deepseek-r1-distill-qwen-1.5b-unsloth-bnb-4bit tags: - text-generation-inference - transformers - unsloth - qwen2 - trl - deepseek - deepseek-r1 - cybersecurity - red-team - blue-team - security - reasoning - chain-of-thought - edge-ai - gguf - ollama - llama-cpp - lora - 4-bit - llm-security - penetration-testing - threat-analysis - mcp-security - ai-safety - offline - air-gapped license: apache-2.0 language: - en library_name: transformers pipeline_tag: text-generation model_type: qwen2 datasets: - Nguuma/security-slm-dataset metrics: - name: Security Reasoning Score (post fine-tune) type: custom value: 8.0 - name: Security Reasoning Score (baseline) type: custom value: 3.4 - name: Think-block activation rate type: custom value: 1.0 model-index: - name: security-slm-unsloth-1.5b results: - task: type: text-generation name: Security Reasoning dataset: type: custom name: Security SLM Dataset args: security-domain metrics: - name: Eval Score (baseline) type: custom_security_reasoning value: 3.4 - name: Eval Score (fine-tuned) type: custom_security_reasoning value: 8.0 - name: Think-block activation rate type: custom_think_rate value: 1.0 --- # security-slm-unsloth-1.5b — Edge-Deployable Security Reasoning Model **Developed by:** Nguuma **License:** Apache-2.0 **Base model:** unsloth/deepseek-r1-distill-qwen-1.5b-unsloth-bnb-4bit **Quantized format:** GGUF Q4_K_M (~1.2 GB RAM at inference) > A security-focused small language model that **thinks before it answers** — fine-tuned for AI-native Blue/Red team operations, deployable on a 4 GB RAM machine with no GPU required. Trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) — [](https://github.com/unslothai/unsloth) --- ## Quickstart — Copy & Run ```python # pip install llama-cpp-python huggingface_hub from huggingface_hub import hf_hub_download from llama_cpp import Llama # Download the fine-tuned GGUF from HuggingFace (~1.2 GB, one-time) model_path = hf_hub_download( repo_id="Nguuma/security-slm-unsloth-1.5b", filename="security-slm-finetuned.gguf", local_dir="./models", ) # Load — runs on CPU, no GPU required llm = Llama( model_path=model_path, n_ctx=2048, n_threads=4, # adjust to your CPU core count verbose=False, ) # Ask a security question response = llm.create_chat_completion( messages=[ { "role": "system", "content": "You are a Cybersecurity assistant with Blue and Red team security reasoning. Think step by step before answering.", }, { "role": "user", "content": 'An AI agent received this tool-call response: {"file": "../../../../etc/passwd"}. Is this a path traversal attack? What should the agent do?', }, ], max_tokens=512, temperature=0.7, top_p=0.9, ) print(response["choices"][0]["message"]["content"]) ``` > **Prefer Ollama?** One command: `ollama run hf.co/Nguuma/security-slm-unsloth-1.5b` --- ## Why security-slm-unsloth-1.5b? Most security-aware LLMs require cloud APIs, expose sensitive queries to third parties, and run on expensive hardware. **security-slm-unsloth-1.5b runs entirely offline on commodity hardware** — a reasoning-capable SLM purpose-built for the 2026 AI threat landscape, covering attack classes that general-purpose models have no training signal for: MCP tool poisoning, agentic lateral movement, Crescendo jailbreaks, LLM-assisted SSRF, financial fraud detection, ransomware incident response, CVE/CWE reasoning, MITRE ATT&CK TTP mapping, and regulatory compliance reasoning (NDPR, GDPR, PCI-DSS). --- ## Model Description security-slm-unsloth-1.5b is a fine-tuned version of DeepSeek-R1-Distill-Qwen-1.5B, specialised in cybersecurity reasoning across offensive and defensive contexts. It preserves the base model's chain-of-thought (``) reasoning behaviour and redirects it toward security-domain problems: threat analysis, attack simulation, detection logic, and AI-specific attack patterns emerging in 2025–2026. | Property | Value | |---|---| | Base architecture | Qwen2 / DeepSeek-R1-Distill | | Parameters | 1.5B | | Training dataset | curated security samples across 11 domains | | Training epochs | 5 | | Final training loss | 1.69 | | Eval score (pre-fine-tune) | 3.4 / 10 | | Eval score (post-fine-tune) | 8.0 / 10 | | Improvement | **+135%** | | Think-block activation rate | **100%** | | GGUF RAM footprint | ~1.2 GB (Q4_K_M) | | LoRA rank | r=16 | | LoRA target modules | q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj | --- ## Files in This Repository | File | Description | |---|---| | `*.gguf` | Q4_K_M quantized model — use with Ollama or llama.cpp | | `adapter_model.safetensors` | LoRA adapter weights (~30MB) — use with Transformers + PEFT | | `adapter_config.json` | LoRA configuration | | `tokenizer*` | Tokenizer files | --- ## Quickstart ### Ollama (recommended — one command) ```bash ollama run hf.co/Nguuma/security-slm-unsloth-1.5b ``` Or pull first then run: ```bash ollama pull hf.co/Nguuma/security-slm-unsloth-1.5b ollama run hf.co/Nguuma/security-slm-unsloth-1.5b ``` ### Ollama with custom Modelfile Save this as `Modelfile`, then run `ollama create security-slm -f Modelfile && ollama run security-slm`: ``` FROM hf.co/Nguuma/security-slm-unsloth-1.5b SYSTEM """You are a Cybersecurity assistant with Blue and Red team security reasoning. Think step by step before answering.""" PARAMETER temperature 0.7 PARAMETER top_p 0.9 PARAMETER num_predict 512 PARAMETER num_ctx 2048 ``` ### llama.cpp ```bash # Download the GGUF huggingface-cli download Nguuma/security-slm-unsloth-1.5b --include "*.gguf" --local-dir ./ # Run ./llama-cli -m security-slm-finetuned.gguf \ --prompt "Analyse this log entry for signs of prompt injection: ..." \ -n 512 ``` ### Transformers + PEFT (LoRA adapter) ```python from transformers import AutoTokenizer, AutoModelForCausalLM from peft import PeftModel base = AutoModelForCausalLM.from_pretrained( "unsloth/deepseek-r1-distill-qwen-1.5b-unsloth-bnb-4bit" ) model = PeftModel.from_pretrained(base, "Nguuma/security-slm-unsloth-1.5b") tokenizer = AutoTokenizer.from_pretrained("Nguuma/security-slm-unsloth-1.5b") ``` --- ## Prompt Format This model uses the ChatML format. Always include a system prompt and open the assistant turn with `` to trigger chain-of-thought reasoning: ``` <|im_start|>system You are a Cybersecurity assistant with Blue and Red team security reasoning. Think step by step before answering. <|im_end|> <|im_start|>user A user's AI agent received this tool-call response: {"file": "../../../../etc/passwd"}. Is this a path traversal attack? What should the agent do? <|im_end|> <|im_start|>assistant ``` The model will complete the `` block with its reasoning chain, then deliver a structured answer. --- ## Training Dataset Fine-tuned on curated security samples** covering **2026 AI-native threat categories** not present in standard security benchmarks. Every scenario is authored as a matched red/blue pair — the same threat modelled from both attacker and defender perspectives. | Domain | Description | |---|---| | **MCP Attacks** | Model Context Protocol exploitation, tool-call injection, context poisoning | | **Prompt Hijacking** | Crescendo attacks, payload splitting, indirect injection chains | | **Agentic Security** | Lateral movement between AI agents, privilege escalation in tool-use pipelines | | **Cloud-Native AI** | RAG poisoning, SSRF via LLM agents, S3 misconfiguration exploitation | | **Guardrail Bypass** | Base64, Unicode homoglyph, and encoding-based evasion techniques | | **Financial Fraud** | Transaction fraud patterns, account takeover, money mule detection, payment interception, card skimming, SIM swap, deepfake-enabled identity fraud | | **CVE/CWE Reasoning** | Vulnerability root cause analysis (CWE-89, CWE-79, CWE-287, CWE-502, CWE-918), CVE exploit chains mapped to fintech and cloud stacks | | **MITRE ATT&CK TTP** | Technique extraction from incident logs, kill chain mapping, Sigma/KQL detection rule generation across T1566, T1078, T1190, T1486, T1003 and more | | **Ransomware IR** | Triage and family identification (LockBit, BlackCat/ALPHV, Cl0p, Akira), containment playbooks, recovery sequencing for critical infrastructure | | **Regulatory Compliance** | NDPR, GDPR, PCI-DSS v4.0, ISO 27001, and sector-specific cybersecurity frameworks — breach notification obligations, gap analysis, self-assessment reasoning | | **AI Attack Detection** | AI-generated phishing detection, deepfake audio/video fraud, RAG pipeline poisoning, banking chatbot prompt injection, LLM-assisted BEC | All samples include preserved `` reasoning blocks — critical for security work where auditability matters. --- ## Evaluation Results Same 10 standardised prompts run against base model and fine-tuned model: | Metric | Baseline | Fine-Tuned | Change | |---|---|---|---| | Average score (/ 10) | 3.4 | 8.0 | **+135%** | | `` block rate | 20% | **100%** | +80pp | | Average response length | 341 words | 272 words | more precise | | Technical depth markers | 1–2 / 5 | 4–5 / 5 | +3× | **Scoring rubric (10 pts total):** Reasoning presence (3) · Reasoning depth (3) · Technical specificity (2) · Response coverage (2) --- ## Key Features - **Offline-first** — No API calls, no data exfiltration risk. Safe for sensitive security environments. - **Edge-deployable** — Runs on a 4 GB RAM laptop via Ollama or llama.cpp. No GPU required. - **100% chain-of-thought** — Every response includes a `` reasoning chain. The model shows its work. - **2026 threat coverage** — Trained on AI-native attack classes absent from standard model training: MCP, agentic lateral movement, Crescendo, LLM SSRF. - **Financial fraud reasoning** — Covers transaction fraud, account takeover, payment interception, and deepfake-enabled identity fraud with detection logic and playbooks. - **CVE/CWE + ATT&CK native** — Reasons from vulnerability root cause (CWE) through exploit chain to MITRE ATT&CK technique mapping and Sigma detection rule generation. - **Ransomware IR** — Triage, containment, and recovery playbooks for LockBit, BlackCat/ALPHV, Cl0p, and Akira targeting financial and critical infrastructure. - **Compliance-aware** — Reasons through NDPR, GDPR, PCI-DSS v4.0, and ISO 27001 breach notification and gap analysis scenarios. - **Dual-use** — Blue team (detection, triage, policy) and Red team (simulation, adversarial testing). - **Quantized & portable** — Q4_K_M GGUF, ~1.2 GB. Fits on a USB drive. --- ## Use Cases ### Blue Team / Defensive Security - Analyse suspicious logs and network events for indicators of compromise - Draft detection rules (Sigma, YARA, KQL) from attack descriptions - Explain CVEs, map them to CWE root causes, and surface remediation paths - Map incident evidence to MITRE ATT&CK tactics and techniques - Assess security posture of AI/LLM deployments (RAG pipelines, agentic systems) - Generate incident response playbooks for ransomware and financial fraud - Detect AI-generated phishing, deepfake-enabled fraud, and BEC patterns - Reason through NDPR, GDPR, and PCI-DSS breach notification obligations ### Red Team / Offensive Security - Simulate adversarial prompts and injection chains for AI system testing - Reason through attack paths against cloud-native AI infrastructure - Generate phishing and social engineering scenario templates for awareness training - Enumerate MCP and agentic attack surfaces - Model financial fraud techniques (account takeover, payment interception) for red team exercises ### Financial Sector Security - Detect and reason about transaction fraud patterns: fan-out transfers, velocity anomalies, mule account activation - Triage payment fraud: card skimming, SIM swap, USSD interception, deepfake voice KYC bypass - Ransomware containment and recovery sequencing for core banking and payment infrastructure - Map financial sector breaches to MITRE ATT&CK and generate SIEM detection rules - Compliance gap analysis against PCI-DSS v4.0, ISO 27001, and sector-specific frameworks ### AI Security Research - Study how reasoning models behave on adversarial security inputs - Benchmark SLM security knowledge against larger frontier models - Prototype lightweight security copilots for air-gapped environments - Explore AI-native threat modelling for LLM/agent pipelines ### Education & CTF - Walk through security concepts with chain-of-thought explanations - Assist with Capture the Flag challenge reasoning - Train junior analysts on threat patterns with guided step-by-step analysis --- ## Limitations - Trained on domain-specific samples — a focused specialist, not a general security encyclopedia - CVE/CWE and MITRE ATT&CK coverage is curated, not exhaustive — verify against NVD and ATT&CK Navigator for production use - Ransomware IR playbooks are generalist starting points; adjust containment steps to your specific infrastructure - Regulatory compliance reasoning (NDPR, GDPR, PCI-DSS) is advisory — consult qualified legal/compliance professionals for binding decisions - Not a substitute for professional penetration testing or incident response - Intended for **authorised security testing, research, and education only** --- ## Responsible Use This model is designed for **defensive security, authorised red team exercises, CTF competitions, and security education**. Do not use it to conduct unauthorised access, develop malware, or attack systems you do not own or have explicit permission to test. --- ## Citation ```bibtex @misc{nguuma2026securityslm, title = {security-slm-unsloth-1.5b: Edge-Deployable Reasoning Model for AI-Native Security Intelligence}, author = {Nguuma}, year = {2026}, howpublished = {HuggingFace}, url = {https://huggingface.co/Nguuma/security-slm-unsloth-1.5b} } ``` --- *Fine-tuned with [Unsloth](https://github.com/unslothai/unsloth) on Google Colab. Reasoning architecture based on DeepSeek-R1.*