Ushitha/ushitha-coder-network-corrector

Network Security Config LoRA

Fine-tuned LoRA adapter on Qwen/Qwen2.5-7B-Instruct.

Trained on 246 Cisco router/switch configuration pairs across 10 categories: Basic Router, Basic Switch, VLAN, ACL, Trunking, NAT, OSPF, EIGRP, DHCP, SSH/Telnet.

What it does

Give it an insecure or AI-generated Cisco config — it will:

1. Identify every security vulnerability (Critical / Important / Best-Practice)
2. Explain why each issue matters
3. Output a fully corrected, production-hardened configuration
4. Show the security score improvement

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
import torch

base      = "Qwen/Qwen2.5-7B-Instruct"
lora_repo = "Ushitha/ushitha-coder-network-corrector"

tokenizer = AutoTokenizer.from_pretrained(base)
model = AutoModelForCausalLM.from_pretrained(
    base, torch_dtype=torch.bfloat16, device_map="auto")
model = PeftModel.from_pretrained(model, lora_repo)

insecure_config = """
hostname Router
interface GigabitEthernet0/0
 ip address 192.168.1.1 255.255.255.0
 no shutdown
"""

messages = [
    {"role": "system",  "content": "You are a network security expert..."},
    {"role": "user",    "content": f"Review this config:\n{insecure_config}"},
]
text   = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
output = model.generate(**inputs, max_new_tokens=2048, temperature=0.1, do_sample=True)
print(tokenizer.decode(output[0], skip_special_tokens=True))

Training Details

Parameter	Value
Base model	Qwen/Qwen2.5-7B-Instruct
Technique	QLoRA 4-bit NF4
LoRA rank / alpha	16 / 32
Training examples	246
Epochs	3
Effective batch size	8
Learning rate	0.0002
Hardware	NVIDIA A40 48 GB