Jailbreak Prediction Model: llama2:7b (fine-tuned on harmful_behaviors_30)

Evaluation Results

Fine-Tuning Details

• Base model: yonad2008/predict_llama2_7b
• Target model: llama2:7b
• Dataset: harmful_behaviors_30.csv
• Frozen layers: 9/12 transformer layers
• Epochs: 3
• Learning Rate: 5e-06
• Max Length: 512
• Input format: turns only

Dataset Size (before turn expansion)