yonad2008/predict_gnn_llama3_8b

GNN Jailbreak Prediction Model (llama3:8b)

Homogeneous GNN classifier for unsafe/jailbreak likelihood in multi-turn conversations.

Evaluation Results

Metric	Value
F1	0.9627
PR-AUC	0.9923
ROC-AUC	0.9920
Precision	0.9700
Recall	0.9580
Best Threshold	0.430

Training Details

• Target model: llama3:8b
• Datasets: harmbench
• Split column: goal
• Seed: 42
• Sentence model: sentence-transformers/all-MiniLM-L6-v2
• Hidden channels: 128
• Num layers: 2
• Dropout: 0.3

Dataset Size (training samples)

Prepared turn-level samples: 522