Categorica/XLM-R-Indic-Hardened

Defended Indic Fact-Checking Model

This model has been fine-tuned and defended against adversarial attacks using the Unified Adversarial Testing Framework.

Model Details

• Language: multilingual
• Task: Fact-Checking / Claim Verification
• Defenses Applied: Product of Experts (PoE), Debiased Focal Loss (DFL)

Intended Use

Designed to verify claims in regional Indian languages with high robustness against adversarial manipulations (e.g., character swapping, multi-hop temporal logic).