Dospacite/xai-phishing-deepseek-r1-qwen-7b-merged
Dospacite/xai-phishing-deepseek-r1-qwen-7b-merged is a 7.6 billion parameter model, merged from a LoRA adapter into DeepSeek-R1-Distill-Qwen-7B. This model is specifically designed for classifying structured webpage evidence as phishing or benign. It provides an explanation for its classification, making it optimized for decision support in cybersecurity applications.
Loading preview...
Model Overview
The Dospacite/xai-phishing-deepseek-r1-qwen-7b-merged model is a specialized 7.6 billion parameter language model. It was created by merging the Dospacite/xai-phishing-deepseek-r1-qwen-7b LoRA adapter into the deepseek-ai/DeepSeek-R1-Distill-Qwen-7B base model. This standalone model includes complete weights, eliminating the need for PEFT or the adapter repository during inference.
Key Capabilities
- Phishing Detection: Classifies structured webpage evidence as either phishing or benign.
- Explainable AI (XAI): Provides explanations for its classification decisions, citing specific webpage features.
- Decision Support: Designed to offer insights that assist human review, rather than making autonomous decisions.
Intended Use Cases
This model is particularly well-suited for:
- Cybersecurity Analysis: Assisting security professionals in identifying and understanding potential phishing threats.
- Webpage Content Review: Automating initial screening of web content for malicious intent.
- Research in XAI for Security: Exploring how AI can provide transparent and interpretable outputs in critical security contexts.
It is crucial that the model's output, which serves as decision support, is always reviewed alongside the cited webpage features by a human expert.