seffyehl/BetterShield

---

language: en license: mit tags: - hate-speech-detection - text-classification - rationale-extraction datasets: - your-dataset-name metrics: - accuracy - f1 model-index: - name: hate-speech-rationale-model results: - task: type: text-classification name: Hate Speech Detection metrics: - type: accuracy value: 0.XX # Your test accuracy - type: f1 value: 0.XX # Your test F1

Hate Speech Detection with Rationale Extraction

This model detects hate speech in text and provides rationale explanations for its predictions.

Model Description

• Architecture: HateBERT + Rationale BERT + Multi-Scale CNN + Attention
• Training Data: [Describe your dataset]
• Performance:
  • Validation Loss: 0.27
  • Test Accuracy: XX%
  • Test F1: XX%

Usage

from huggingface_hub import hf_hub_download
import torch

# Download model
model_path = hf_hub_download(
    repo_id="seffyehl/BetterShield",
    filename="pytorch_model.pth"
)

# Load model (see full example in repository)
checkpoint = torch.load(model_path, map_location='cpu')
# ... (rest of loading code)

Training Details

• Batch Size: 8
• Learning Rate: 1e-5
• Weight Decay: 0.05
• Dropout: 0.5
• Epochs: Stopped early at epoch X

Limitations

[Describe any known limitations]

Citation

@misc{bettershield-2025,
  author = Orion,
  title = Hate Speech Detection with Rationale Extraction,
  year = 2025,
  publisher = HuggingFace,
  url = https://huggingface.co/seffyehl/BetterShield
}