---
license: apache-2.0
base_model: AlicanKiraz0/Titus-CybersecurityLLM-v1.0
library_name: gguf
pipeline_tag: text-generation
tags:
  - qwen
  - qwen3.6
  - cybersecurity
  - turkish
  - gguf
  - q4_0
  - llama.cpp
  - quantized
  - soc
  - dfir
---

# Titus-CybersecurityLLM-v1.0 Q4_K_M GGUF

![Titus-CybersecurityLLM](titus-cybersecurityllm.png)

This repository contains the **Q4_K_M GGUF variant** of `AlicanKiraz0/Titus-CybersecurityLLM-v1.0`, a Turkish-first cybersecurity assistant fine-tuned from `Qwen/Qwen3.6-35B-A3B`.

The source model was trained on a **500K+ row cybersecurity instruction dataset** created with a structured security taxonomy, using LoRA over an approximately **4B-parameter adaptation surface**, then merged and converted to GGUF for llama.cpp-compatible runtimes.

## Files

- `titus-cybersecurityllm-v1.0-q4_0.gguf`

## Variant Details

- **Variant:** GGUF Q4_K_M
- **Source model:** `AlicanKiraz0/Titus-CybersecurityLLM-v1.0`
- **Base model:** `Qwen/Qwen3.6-35B-A3B`
- **Architecture:** `qwen3_5_moe`
- **Target runtime:** llama.cpp-compatible tools
- **Primary language:** Turkish cybersecurity assistance

## Intended Use

This GGUF build is intended for local inference workflows where lower memory usage is preferred:

- SOC alert triage
- DFIR checklist generation
- Threat hunting prompts
- Detection logic drafting
- IAM, cloud, Kubernetes, endpoint, Docker, and AppSec review support
- Authorized purple-team, lab, and CTF-style validation

## Dataset Taxonomy Summary

The source model was fine-tuned with a cybersecurity taxonomy that models:

- **Domain:** SOC, IR, DFIR, cloud, IAM, endpoint, web, API, Kubernetes, AppSec, DevSecOps, malware, threat intel, GRC, OT/ICS, mobile, AI/LLM security, and resilience
- **Artifact type:** logs, EDR telemetry, SIEM alerts, HTTP transcripts, email headers, IAM policies, K8s manifests, Terraform, SBOM, CVE records, forensic excerpts
- **Task family:** triage, classification, artifact analysis, detection engineering, rule/query writing, root cause analysis, remediation planning, executive reporting
- **Reasoning type:** direct recognition, causal inference, temporal reconstruction, evidence synthesis, trade-off analysis, cross-domain reasoning
- **Difficulty:** L1 fundamental through L5 research/strategic
- **Safety:** defensive, bounded dual-use, CTF/lab-only, restricted/excluded

## Example Usage

With llama.cpp:

```bash
llama-cli \
  -m titus-cybersecurityllm-v1.0-q4_k_m.gguf \
  -p "Windows hostta LSASS erişimi şüphesini doğrulamak için hangi telemetry alanlarına bakarsın?" \
  -n 512 \
  --temp 0
```

With a chat template-aware frontend such as LM Studio or compatible llama.cpp server builds, load the GGUF file and use Turkish cybersecurity prompts directly.

## Notes

- Q4_K_M is optimized for size and accessibility, not maximum quality.
- For highest fidelity, use the BF16 merged safetensors model: `AlicanKiraz0/Titus-CybersecurityLLM-v1.0`.
- For Apple Silicon MLX workflows, use: `AlicanKiraz0/Titus-CybersecurityLLM-v1.0-mlx-4Bit`.

## Safety

This model is intended for authorized defensive security workflows. Offensive or dual-use analysis should remain limited to legal, controlled, and explicitly authorized lab, CTF, red-team, purple-team, or detection validation contexts.