You can deploy this model and start earning money today!

jartechsai/
cybersec-mistral-7b

Name: cybersec-mistral-7b
Availability: InStock

safetensors

Views

No views yet

CyberSec Mistral 7B (v5)

Fine-tuned Mistral 7B for cybersecurity question answering using QLoRA/Unsloth.

What's New in v5

Clean single-pass training — eliminated domain contamination from v4's isolated correction runs
Balanced dataset — all domains represented proportionally in one shuffled training set
Expanded coverage — added penetration testing, incident response, malware analysis, cloud security, and cryptography domains
Validated before push — 23-question test suite checking both accuracy and domain isolation

Training Data (v5)

MITRE ATT&CK Enterprise: techniques, groups, software
IANA port registry + networking concepts
CVE/NVD HIGH severity vulnerabilities
CWE weakness database
OWASP Top 10 coverage
HuggingFace cybersecurity datasets (Fenrir v2.0, Cybersecurity-Dataset-v1, CybersecurityQAA, NIST)
Hand-crafted examples: penetration testing, incident response, malware analysis, cloud security, cryptography, password cracking, SIEM, IT support

Domains Covered

Network Security & Threats
IT Support & Troubleshooting
Compliance & Frameworks (NIST, MITRE ATT&CK, CIS)
Penetration Testing & CTF
Incident Response & Digital Forensics
Malware Analysis
Cloud Security (AWS, Azure, GCP)
Cryptography

Usage

python

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "jartechsai/cybersec-mistral-7b",
    max_seq_length = 2048,
    dtype = None,
    load_in_4bit = True,
)

FastLanguageModel.for_inference(model)

alpaca_prompt = """Below is a cybersecurity question or task. Provide a thorough, accurate response.

### Instruction:
{}

### Input:
{}

### Response:
{}"""

inputs = tokenizer([alpaca_prompt.format("What is a KARMA attack?", "", "")], return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.3, do_sample=True, repetition_penalty=1.3)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Model Details

Base: mistral-7b-instruct-v0.3
Method: QLoRA (r=16, alpha=16) via Unsloth
Training: 1 epoch, LR 5e-5, cosine scheduler
GPU: NVIDIA L4/A100 (Colab Pro)
Demo: Hugging Face Space