Dospacite/xai-phishing-qwen3-4b-merged

TEXT GENERATIONConcurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jun 10, 2026Architecture:Transformer Featherless Exclusive Cold

Dospacite/xai-phishing-qwen3-4b-merged is a 4 billion parameter Qwen3-based language model developed by Dospacite, specifically fine-tuned for explainable phishing detection. This model excels at classifying structured webpage evidence as phishing or benign and provides explanations for its decisions. It is designed for decision support in cybersecurity applications, offering insights into webpage features that indicate malicious intent.

Loading preview...

Explainable Phishing Detector Qwen3 4B

This model, developed by Dospacite, is a 4 billion parameter Qwen3-based language model created by merging a LoRA adapter into the base Qwen/Qwen3-4B model. It is designed to provide explainable phishing detection by analyzing structured webpage features.

Key Capabilities

  • Phishing Classification: Accurately classifies structured webpage evidence as either phishing or benign.
  • Explainable AI (XAI): Provides explanations for its classification decisions, detailing which webpage features contributed to the verdict.
  • Standalone Operation: Contains complete model weights, eliminating the need for PEFT or external adapter repositories during inference.

Intended Use

This model is primarily intended for decision support in cybersecurity contexts. Its output, which includes both a classification and an explanation, should be reviewed by human operators alongside the cited webpage features to make informed judgments about potential phishing threats. It expects a structured webpage feature prompt for optimal performance.

Finetunes
1 models