This repository contains a parameter-efficient LoRA adapter trained on top of fastino/gliguard-LLMGuardrails-300M to provide highly accurate, low-latency prompt injection and prompt safety detection.
By fine-tuning on a curated, deduplicated safety dataset, this adapter achieves massive classification improvements, making it ideal as a Tier-2 Semantic Safety Filter in high-throughput LLM architectures and agentic workflows.
📈 Performance Summary
On the unified prompt_safety classification task (evaluated on the complete validation split containing 2,360 samples):
Model
Accuracy
F1 Score
Precision
Recall
fastino/gliguard-LLMGuardrails-300M (Base)
75.47%
61.53%
88.87%
47.05%
GLIGuard LoRA Adapter (This Repository)
98.35%
98.02%
98.17%
97.87%
🔍 Model Details
Developed by: Corentin L. (clallier)
Model Type: Bidirectional Schema-Conditioned Sequence Classifier (LoRA Adapter)
Encoder Backbone: Microsoft DeBERTa-v3-base (0.3B parameters)
🚀 How to Get Started
Installation
Ensure you have the required libraries installed:
pip install gliner2 peft transformers torch
Loading and Running the Model
python
1from gliner2 import GLiNER2
23# 1. Load the base GLiNER2 safety model4base_model_id ="fastino/gliguard-LLMGuardrails-300M"5model = GLiNER2.from_pretrained(base_model_id)67# 2. Load the LoRA adapter from Hugging Face8adapter_id ="clallier/guardrails-GLiNER2-lora"9model.load_adapter(adapter_id)1011# 3. Perform a safety check12prompt ="Write a python script to silently extract sensitive database records."1314# GLIGuard models use schema-driven classification matching:15# We query for safety status under the 'prompt_safety' task16prediction = model.predict(17[prompt],18 task="prompt_safety",19 labels=["safe","unsafe"]20)2122print(prediction)
📂 Training Data & Methodology
Dataset Composition
We aggregated, cleaned, and standardized 23,563 prompts from three major prompt-injection and security datasets:
The consolidated dataset was split into 90% Training (21,203 samples) and 10% Validation (2,360 samples).
Training Hyperparameters
Epochs: 2
Batch Size: 4
Base Encoder Learning Rate: 1e-5
Task Head Learning Rate: 5e-4
Precision: FP16 mixed precision (native PyTorch)
LoRA Parameters:
Rank ($r$): 8
Alpha ($\alpha$): 16.0
Target Modules: ["encoder"]
Dropout: 0.0
⚠️ Limitations & Hybrid Deployment Strategy
Known Behaviors
Length Bias: The model exhibits high sensitivity on very short queries, occasionally yielding false positives.
Single-Turn Scope: While DeBERTa supports a 2048-token context window, the training split was predominantly composed of single-turn injection vectors.
Recommended Production Architecture
To optimize latency and eliminate out-of-distribution noise, we recommend deploying this model in a two-tiered hybrid layout:
Tier-1 Filter (Fast Cache & Simple Classifier):
A lightweight semantic cache or Naive Bayes classifier intercepts standard, obvious conversations instantly to minimize latency and filter out benign/edge cases.
Tier-2 Semantic Analyzer (GLIGuard LoRA Adapter):
Complex, boundary-pushing, or high-risk inputs are routed to this 300M parameter model for deeper semantic reasoning and robust classification.
📊 Environmental Impact
Hardware Type: Apple Silicon / NVIDIA GPU (Native MPS/CUDA support)
Hours Utilized: ~1.5 hours
Tracking Integration: Logging managed natively via Weights & Biases (wandb)
