bachuntr-14b-LoRA

QLoRA adapter for Qwen3-14B, fine-tuned for automated BAC/IDOR vulnerability testing.

Model Details

Property	Value
Base Model	Qwen/Qwen3-14B
LoRA Rank	64
LoRA Alpha	16
Method	RSLoRA, all projection layers
Training Examples	413 train / 67 val
Epochs	4
Batch Size	2 × grad_accum 8
Learning Rate	2e-4
Max Seq Length	16384
Final Loss	0.155 (from 2.069)
Adapter Size	~500 MB

Usage

With Transformers + PEFT

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-14B", device_map="auto")
model = PeftModel.from_pretrained(base, "your-username/bachuntr-14b-lora")
tokenizer = AutoTokenizer.from_pretrained("your-username/bachuntr-14b-lora")

With vLLM (recommended for production)

vllm serve Qwen/Qwen3-14B \
  --enable-lora \
  --lora-modules bachuntr-14b=./bachuntr-14b-lora \
  --served-model-name bachuntr-14b \
  --max-model-len 32768

Merge into full model

from peft import PeftModel
from transformers import AutoModelForCausalLM

base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-14B", torch_dtype="auto")
model = PeftModel.from_pretrained(base, "./bachuntr-14b-lora")
merged = model.merge_and_unload()
merged.save_pretrained("./bachuntr-14b-merged")

Training Data

Fine-tuned on distilled multi-agent security testing conversations covering:

• Account provisioning & authentication
• Endpoint discovery & API reconnaissance
• Attack planning
• BAC/IDOR exploitation (horizontal, vertical, privilege escalation)
• Multi-persona vulnerability verification
• Security report generation

7 target applications, 480 unique conversations, 22K+ message turns with full tool-calling schemas.

Intended Use

Automated security testing of web applications for Broken Access Control and IDOR vulnerabilities via the bachuntr framework.

Framework Versions

• PEFT 0.18.1
• TRL: 0.24.0
• Transformers: 5.5.0
• Pytorch: 2.10.0+cu126
• Datasets: 4.3.0
• Tokenizers: 0.22.2
• Unsloth: 2026.4.1