Llama-Phishsense-1B: AI-Powered Phishing Defense

The AcuteShrewdSecurity/Llama-Phishsense-1B is a 1 billion parameter model, fine-tuned from meta-llama/Llama-Guard-3-1B using Low-Rank Adaptation (LoRA). Its primary function is to classify emails as either phishing (TRUE) or non-phishing (FALSE), providing a lightweight yet powerful defense against evolving email scams.

Key Capabilities

• High Accuracy: Achieves an F1-score of 0.99 on balanced datasets and strong performance on custom expert-designed phishing cases, significantly outperforming the base model.
• Efficiency: LoRA fine-tuning ensures fast operation and minimal computational resource requirements, making it suitable for diverse deployment environments.
• Binary Classification: Specializes in clear TRUE/FALSE predictions for phishing detection.
• Custom Dataset Training: Fine-tuned on a balanced dataset of 30k real-world phishing and benign emails for robust applicability.

Good For

• Corporate Enterprise Phishing Protection: Integrates into corporate email systems to mitigate risks from targeted phishing attacks and prevent unauthorized access.
• Individual Email Security: Identifies phishing attempts in personal inboxes, offering peace of mind by securing private data.
• Security as a Service: Can be incorporated into existing cybersecurity offerings by security professionals and IT providers to enhance client protection.