---
base_model: Qwen/Qwen3-4B-Instruct-2507
library_name: peft
pipeline_tag: text-generation
tags:
- base_model:adapter:Qwen/Qwen3-4B-Instruct-2507
- lora
- sft
- transformers
- trl
- security
- guardrails
- multilingual
language:
- en
- tr
---

# 🛡️ LLM-Fort Guardrails Suite (v1)

[![Base Model](https://img.shields.io/badge/Base_Model-Qwen3--4B--Instruct-blue?style=flat-square)](https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507)
[![Framework](https://img.shields.io/badge/Framework-PEFT_/_LoRA-orange?style=flat-square)](https://github.com/huggingface/peft)
[![Collection](https://img.shields.io/badge/Collection-LLM--Fort_Guardrails-purple?style=flat-square)](https://huggingface.co/collections/ApiFort/llmfort-guardrails-v1)

LLM-Fort Guardrails is a suite of **7 security-focused LoRA adapters** fine-tuned on top of **Qwen/Qwen3-4B-Instruct-2507**. These adapters serve as lightweight, high-performance security guardrails mapped to critical safety boundaries.

By offloading classification and security checks to lightweight adapters, the system achieves enterprise-grade security filtering without degrading the inference performance of the main application model.

📖 Collection Page: [ApiFort/llmfort-guardrails-v1](https://huggingface.co/collections/ApiFort/llmfort-guardrails-v1)

---

## 🗺️ Category Mappings

| Vulnerability Category | Adapter Model ID | Description |
|:---|:---|:---|
| 🚨 **[Prompt Injection](https://huggingface.co/ApiFort/jailbreak_content_injection)** | `jailbreak_content_injection` | Detects direct/indirect prompt injection and jailbreak attempts |
| 🕵️ **[PII Extraction](https://huggingface.co/ApiFort/pii)** | `pii` | Identifies and extracts PII (person names, ID numbers, etc.) |
| 💻 **[Code Security](https://huggingface.co/ApiFort/code_security)** | `code_security` | Scans code snippets for software vulnerabilities (SQLi, SSRF, XSS) |
| 🚦 **[Excessive Agency](https://huggingface.co/ApiFort/excessive_agency)** | `excessive_agency` | Intercepts unauthorized or destructive critical tool calls |
| 🔒 **[System Prompt Leakage](https://huggingface.co/ApiFort/system_prompt_leakage)** | `system_prompt_leakage` | Detects attempts to extract developer system instructions |
| ⚠️ **[Content Safety](https://huggingface.co/ApiFort/content_safety)** | `content_safety` | Blocks hate speech, harassment, and general unsafe content |
| 🛑 **[Unbounded Consumption](https://huggingface.co/ApiFort/unbounded_consumption)** | `unbounded_consumption` | Mitigates resource exhaustion and compute DoS attacks |

---

## 📈 Performance & Evaluation

Visual comparison of baseline performance versus the trained adapters:

![LLM-Fort Guardrails Accuracy Comparison](benchmark_chart.png)

### Benchmark Results
Below is the exact accuracy performance measured across our evaluation test suites:

| Category | Gemma 4-E4B-it | Qwen 3.5 4B | Qwen3 4B Instruct | llmfort ai guardrail v.1.0 |
|:---|:---:|:---:|:---:|:---:|
| **[Prompt Injection](https://huggingface.co/ApiFort/jailbreak_content_injection)** | 56.30% | 64.70% | 84.14% | **98.10%** |
| **[PII Extraction](https://huggingface.co/ApiFort/pii)** | 84.36% | 75.84% | 78.31% | **95.30%** |
| **[Code Security](https://huggingface.co/ApiFort/code_security)** | 83.30% | 75.10% | 76.20% | **90.07%** |
| **[Excessive Agency](https://huggingface.co/ApiFort/excessive_agency)** | 60.90% | 68.80% | 53.80% | **96.50%** |
| **[System Prompt Leakage](https://huggingface.co/ApiFort/system_prompt_leakage)** | 79.40% | 79.70% | 78.70% | **98.08%** |
| **[Content Safety](https://huggingface.co/ApiFort/content_safety)** | 84.00% | 78.80% | 76.00% | **95.30%** |
| **[Unbounded Consumption](https://huggingface.co/ApiFort/unbounded_consumption)** | 63.00% | 63.30% | 55.00% | **99.79%** |

---

## 🗃️ Training & Validation Datasets

The adapters were trained and validated on the following dataset references:

| Category | Source Datasets / References |
|:---|:---|
| **[Prompt Injection](https://huggingface.co/ApiFort/jailbreak_content_injection)** | BIPIA, Deepset, Internal 1K Validation |
| **[PII Extraction](https://huggingface.co/ApiFort/pii)** | AI4Privacy PII Masking 300k (EN, TR, FR, DE, ES) |
| **[Code Security](https://huggingface.co/ApiFort/code_security)** | r2vul, securecode_web |
| **[Excessive Agency](https://huggingface.co/ApiFort/excessive_agency)** | jinjinyien/ToolSafety, minpeter/xlam-function-calling-60k-parsed |
| **[System Prompt Leakage](https://huggingface.co/ApiFort/system_prompt_leakage)** | S-Labs/prompt-injection-dataset, Synthetic data |
| **[Content Safety](https://huggingface.co/ApiFort/content_safety)** | NVIDIA Nemotron-3.5-Content-Safety-Dataset, Wildguardmix |
| **[Unbounded Consumption](https://huggingface.co/ApiFort/unbounded_consumption)** | neuralchemy/prompt-injection-Threat-Matrix, Lakera/mosscap_prompt_injection |