Downloads · 30 days
14
9% of all-time downloads
predict_llama3.2_3b
yonad2008/predict_llama3.2_3b
predict_llama3.2_3b is a text classification model from yonad2008. Use it when you need a label for a piece of text.
Fine-tuned DeBERTa-v3-base for detecting unsafe/jailbreak prompts in multi-turn conversations.
All-time downloads
157
Public
Parameters
184M
2.2 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors738 MB · 99%
At a glance
- Task
- Text Classification
- Model type
- deberta-v2
- Access
- Public
- Created
- Mar 4, 2026
- Updated
- Mar 19, 2026
- SHA
- 87179021
Try a prompt
Model card
From the Hugging Face model README
Jailbreak Prediction Model: llama3.2:3b
Fine-tuned DeBERTa-v3-base for detecting unsafe/jailbreak prompts in multi-turn conversations.
Evaluation Results (best fold: 2)
| Metric | Value |
|---|---|
| F1 | 0.8687 |
| PR-AUC | 0.9193 |
| ROC-AUC | 0.9649 |
| Precision | 0.8431 |
| Recall | 0.8958 |
| Best Threshold | 0.35 |
Training Details
- Base model:
microsoft/deberta-v3-base - Target model:
llama3.2:3b - Datasets: HarmBench
- K-Folds: 5
- Epochs: 5
- Learning Rate: 2e-05
- Max Length: 512
- Input format: turns only
Dataset Size (before turn expansion)
Original rows (after cleaning and balancing): 1319 (unsafe: 220, safe: 1099)
- Task
- Text Classification
- Type
- deberta-v2
- Languages
- en
- Created
- Mar 4, 2026
- Updated
- Mar 19, 2026