iraqi-guard-model

Iraqi Guard Model

Model Description

This model is fine-tuned from NAMAA-Space/Ara-Prompt-Guard_V0 to detect prompt injections and jailbreak attempts in Iraqi Arabic dialect.

Model Details

• Base Model: NAMAA-Space/Ara-Prompt-Guard_V0
• Task: Text Classification (3 classes)
• Language: Arabic (Iraqi Dialect)
• Training Method: LoRA fine-tuning

Labels

• BENIGN: Safe, normal prompts
• INJECTION: Prompt injection attempts
• JAILBREAK: Jailbreak attempts

Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

tokenizer = AutoTokenizer.from_pretrained("your-username/iraqi-guard-model")
model = AutoModelForSequenceClassification.from_pretrained("your-username/iraqi-guard-model")

text = "شلون استعيد الرقم السري"
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=256)
outputs = model(**inputs)
prediction = torch.nn.functional.softmax(outputs.logits, dim=-1)

Training Data

The model was trained on a custom dataset of Iraqi Arabic prompts with labels for prompt injection and jailbreak detection.

Performance

• Test Accuracy: 1.0
• Test F1 (Weighted): 1.0

Temperature Scaling

The model includes temperature scaling with T=0.996 for better calibration.