Pangolin Guard Base

🚀 PangolinGuard-Base

LLM applications are confronted with critical security challenges such as prompt injections and jailbreaks. These issues may lead to models leaking sensitive data or deviating from their intended behavior. Existing safeguard models are not fully open - source and have limited context windows (e.g., LlamaGuard only has a 512 - token context window).

Pangolin Guard is a lightweight ModernBERT (Base) model designed to discriminate malicious prompts (i.e., prompt injection attacks).

🤗 Tech - Blog | GitHub Repo

✨ Features

• Adding a self - hosted, inexpensive defense mechanism against prompt injection attacks to AI agents and conversational interfaces.

📚 Documentation

Evaluation Data

The model is evaluated on unseen data from a subset of specialized benchmarks targeting prompt safety and malicious input detection, while also testing over - defense behavior:

• NotInject: This benchmark is designed to measure over - defense in prompt guard models by including benign inputs enriched with trigger words common in prompt injection attacks.
• BIPIA: It evaluates privacy invasion attempts and boundary - pushing queries through indirect prompt injection attacks.
• Wildguard - Benign: Represents legitimate but potentially ambiguous prompts.
• PINT: Evaluates particularly nuanced prompt injection, jailbreaks, and benign prompts that could be misidentified as malicious.

Inference

from transformers import pipeline

classifier = pipeline("text-classification", "dcarpintero/pangolin-guard-base")
text = "your input text"
output = classifier(text)

Training Procedure

Training Hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e - 05
• train_batch_size: 64
• eval_batch_size: 32
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon = 1e - 08 and optimizer_args = No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 2

Training Results

Framework versions

• Transformers 4.50.0
• Pytorch 2.6.0+cu124
• Datasets 3.4.1
• Tokenizers 0.21.1

📄 License

This project is licensed under the Apache - 2.0 license.

📦 Information Table