prompt_firewall_v2 / model_card.json
adithegod's picture
Upload 12 files
51bf803 verified
Raw History Blame Contribute Delete
1.83 kB
{
"version": "2.0",
"created_at": "2026-06-21T11:07:28.861636",
"architecture": "XGBoost + DeBERTa-v3-base ensemble",
"embedding": "sentence-transformers/all-MiniLM-L6-v2",
"transformer": "microsoft/deberta-v3-base",
"feature_dim": 396,
"ensemble": {
"w_xgb": 0.05,
"w_transformer": 0.95,
"w_faiss-delta": 0.1,
"threshold": 0.915
},
"xgb": {
"n_estimators": 300,
"max_depth": 5,
"threshold": 0.5499999999999998
},
"training_data": {
"adversarial": 12000,
"benign": 12000,
"families": [
"JAILBREAK_PROMPTS",
"ADVERSARIAL_ATTACKS",
"TECHNICAL_ATTACKS",
"ENHANCED_JAILBREAKS",
"CRITICAL_SCENARIOS",
"ADVANCED_TECHNIQUES",
"EXTREME_SCENARIOS",
"JAILBREAK_PROMPTS_Aditya",
"ARMAAN_KEY_ATTACK",
"MULTILINGUAL_PROMPTS",
"ARMAAN_MULTILINGUAL_PROMPTS",
"multi_lang",
"LANGUAGE_BASED_PROMPTS_JIYA",
"SKELETON_KEY_ATTACKS",
"COMBINATION_ATTACKS",
"REFUSAL_SUPPRESSION_ATTACKS",
"AFFIRMATIVE_FORCING_ATTACKS",
"ROLEPLAY_PERSONA_ATTACKS",
"MANY_SHOT_ATTACKS",
"BEST_OF_N_SEMANTIC_ATTACKS",
"JAILBREAK_ATTACKS",
"PROMPT_INJECTION_ATTACKS",
"LOGIC_TRAP_ATTACKS",
"ADVERSARIAL_CREATIVE_ATTACK",
"CHAIN_OF_THOUGHT_ATTACKS",
"SOCIAL_ENGINEERING_SIMULATION",
"AMBIGUOUS_LANGUAGE_SIMULATION",
"MULTI_TURN_PERSISTENCE_SIMULATION",
"ADVANCED_HIGH_FIDELITY_SIMULATION",
"PERSUASIVE_ARGUMENT_SIMULATION",
"PHILOSOPHICAL_TRAP_ATTACKS",
"CONSISTENCY_EXPLOIT_ATTACKS",
"FALSE_PREMISE_ATTACKS",
"REDUCTIO_ATTACKS",
"META_REASONING_ATTACKS",
"OBFUSCATION_ATTACKS",
"PERSONA_ATTACKS"
]
},
"test_metrics": {
"f1": 0.9808,
"fpr": 0.0096,
"tpr": 0.9716
}
}