yonad2008/predict_llama2_7b_harmful_behaviors_30

Jailbreak Prediction Model: llama2:7b (fine-tuned on harmfulbehaviors30)

Incrementally fine-tuned from `yonad2008/predict_llama2_7b` on data from harmful_behaviors_30.csv.

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)

Original rows (after cleaning and balancing): 108 (unsafe: 27, safe: 81)