yonad2008's picture
Upload README.md with huggingface_hub
7532e82 verified
metadata
language: en
tags:
  - jailbreak-detection
  - deberta-v3
  - text-classification
  - fine-tuned
model-index:
  - name: predict_zephyr_7b-beta_harmful_behaviors_1
    results:
      - task:
          type: text-classification
          name: Jailbreak Detection
        metrics:
          - name: F1
            type: f1
            value: 0.2699
          - name: PR-AUC
            type: pr_auc
            value: 0.3947
          - name: ROC-AUC
            type: roc_auc
            value: 0.6368
          - name: Precision
            type: precision
            value: 0.1618
          - name: Recall
            type: recall
            value: 0.8148

Jailbreak Prediction Model: zephyr:7b-beta (fine-tuned on harmful_behaviors_1)

Incrementally fine-tuned from yonad2008/predict_zephyr_7b-beta on data from harmful_behaviors_1.csv.

Evaluation Results

Metric Value
F1 0.2699
PR-AUC 0.3947
ROC-AUC 0.6368
Precision 0.1618
Recall 0.8148
Best Threshold 0.35

Fine-Tuning Details

  • Base model: yonad2008/predict_zephyr_7b-beta
  • Target model: zephyr:7b-beta
  • Dataset: harmful_behaviors_1.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): 296 (unsafe: 74, safe: 222)