|
|
--- |
|
|
license: cc0-1.0 |
|
|
base_model: mlx-community/Qwen2.5-Coder-7B-Instruct-4bit |
|
|
tags: |
|
|
- mlx |
|
|
- cybersecurity |
|
|
- nist |
|
|
- security-controls |
|
|
- compliance |
|
|
- fine-tuned |
|
|
language: |
|
|
- en |
|
|
--- |
|
|
|
|
|
# HackIDLE-NIST-Coder v1.1 (MLX 4-bit) |
|
|
|
|
|
**The most comprehensive NIST cybersecurity model** - Fine-tuned on 530,912 examples from 596 NIST publications. |
|
|
|
|
|
## Model Overview |
|
|
|
|
|
This is an MLX-optimized 4-bit quantized model fine-tuned specifically for NIST cybersecurity expertise. Version 1.1 includes significant improvements over v1.0: |
|
|
|
|
|
- **+7,206 training examples** (530,912 total) |
|
|
- **+28 new documents** (596 NIST publications) |
|
|
- **CSWP series added**: CSF 2.0, Zero Trust Architecture, Post-Quantum Cryptography |
|
|
- **Improved quality**: Fixed 6,150 malformed DOI links |
|
|
|
|
|
## Training Results |
|
|
|
|
|
- **Training iterations**: 1,000 (+ 200 checkpoint recovery) |
|
|
- **Best validation loss**: 1.512 (12.5% improvement) |
|
|
- **Training loss**: 1.420 (final) |
|
|
- **Trainable parameters**: 11.5M (0.151% of 7.6B total) |
|
|
- **Training time**: ~5 hours on M4 Max |
|
|
|
|
|
## Installation |
|
|
|
|
|
```bash |
|
|
pip install mlx-lm |
|
|
``` |
|
|
|
|
|
## Usage |
|
|
|
|
|
```python |
|
|
from mlx_lm import load, generate |
|
|
|
|
|
model, tokenizer = load("ethanolivertroy/HackIDLE-NIST-Coder-v1.1-MLX-4bit") |
|
|
|
|
|
prompt = "What is Zero Trust Architecture according to NIST SP 800-207?" |
|
|
response = generate(model, tokenizer, prompt=prompt, max_tokens=500) |
|
|
print(response) |
|
|
``` |
|
|
|
|
|
## Other Formats |
|
|
|
|
|
- **Ollama**: [etgohome/hackidle-nist-coder:v1.1](https://ollama.com/etgohome/hackidle-nist-coder) |
|
|
- **Dataset**: [ethanolivertroy/nist-cybersecurity-training](https://huggingface.co/datasets/ethanolivertroy/nist-cybersecurity-training) |
|
|
|
|
|
## License |
|
|
|
|
|
CC0 1.0 Universal (Public Domain) - All NIST publications are in the public domain. |
|
|
|
|
|
--- |
|
|
|
|
|
**Version**: 1.1 |
|
|
**Release Date**: October 2025 |
|
|
|
