|
|
--- |
|
|
license: cc-by-nc-sa-4.0 |
|
|
language: |
|
|
- en |
|
|
tags: |
|
|
- unsloth |
|
|
- llama-3.2 |
|
|
- 3b |
|
|
- cybersecurity |
|
|
- instruction-tuning |
|
|
- conversational-ai |
|
|
- penetration-testing |
|
|
- chain-of-thought |
|
|
- gguf |
|
|
- ollama |
|
|
base_model: unsloth/llama-3.2-3b-instruct-nb-bnb-4bit |
|
|
--- |
|
|
|
|
|
 |
|
|
|
|
|
# XO: A Llama 3.2 3B, Unsloth-Trained Cybersecurity Expert |
|
|
|
|
|
## Model Description |
|
|
|
|
|
**XO** is an instruction-fine-tuned model based on **`unsloth/llama-3.2-3b-instruct-nb-bnb-4bit`**. It is engineered to be a lightweight, efficient, and highly specialized AI assistant for cybersecurity tasks. Its small size makes it ideal for local deployment on consumer-grade hardware using tools like Ollama or LM Studio. |
|
|
|
|
|
The model was fine-tuned using the **Unsloth** framework, ensuring maximum performance and minimal resource consumption from the 3B parameter architecture. This version of XO is trained on a focused, foundational dataset to provide core cybersecurity knowledge and a consistent persona in English. |
|
|
|
|
|
## Model Details |
|
|
|
|
|
* **Model Type:** Fine-tuned Causal Language Model |
|
|
* **Base Model:** [unsloth/llama-3.2-3b-instruct-nb-bnb-4bit](https://huggingface.co/unsloth/llama-3.2-3b-instruct-nb-bnb-4bit) |
|
|
* **Training Framework:** [Unsloth](https://github.com/unslothai/unsloth) 🚀 |
|
|
* **Training Data:** The model was trained on the foundational, English-only **[`saberbx/X-mini-datasets`](https://huggingface.co/datasets/saberbx/X-mini-datasets)**. This dataset includes: |
|
|
* Core knowledge adapted from the "Payloads All The Things" repository. |
|
|
* An introductory Chain-of-Thought module for basic reasoning. |
|
|
* A persona module to define its identity as "XO," created by "Saber." |
|
|
* **Important Note:** This model is **NOT** trained on the advanced, bilingual dataset and does **NOT** include advanced mathematical reasoning capabilities. |
|
|
|
|
|
## Capabilities & Intended Use |
|
|
|
|
|
XO is designed to be a reliable local assistant for day-to-day cybersecurity tasks. Its primary capabilities include: |
|
|
|
|
|
* 💻 **Optimized for Local Deployment:** Its 3B parameter size allows it to run smoothly on machines with limited VRAM, making powerful AI accessible. |
|
|
* 🛡️ **Core Cybersecurity Knowledge:** Acts as an interactive encyclopedia of "Payloads All The Things," providing quick access to common payloads, commands, and checklists. |
|
|
* 🧠 **Foundational Reasoning:** Capable of performing basic step-by-step analysis for common cybersecurity problems based on its Chain-of-Thought training. |
|
|
* 👤 **Consistent Persona:** Always responds as "XO," the AI assistant created by "Saber," providing a consistent and predictable user experience. |
|
|
|
|
|
## Limitations and Ethical Considerations |
|
|
|
|
|
* **⚠️ For Ethical & Defensive Use Only:** This model is designed to empower cybersecurity professionals. **Any use for malicious or illegal activities is strictly prohibited.** |
|
|
* **Limited Scope:** This model's knowledge is based on its foundational English training data. It does not possess advanced or multilingual capabilities. |
|
|
* **Potential for Hallucinations:** Like all LLMs, XO can generate incorrect information. **Always verify critical information with a human expert.** |
|
|
* **Bias Warning:** The model may reflect biases from its training data. |
|
|
|
|
|
## Citation |
|
|
|
|
|
If you use this model in your research or project, please cite our work: |
|
|
|
|
|
```bibtex |
|
|
@misc{saber_xo_3b_2024, |
|
|
author = {Saber}, |
|
|
title = {XO: A Llama 3.2 3B, Unsloth-Trained Cybersecurity Expert}, |
|
|
year = {2025}, |
|
|
publisher = {Hugging Face}, |
|
|
journal = {Hugging Face repository}, |
|
|
howpublished = {\url{https://huggingface.co/saberbx/XO}} |
|
|
} |
