|
|
--- |
|
|
title: Toxic Eye |
|
|
emoji: 👀 |
|
|
colorFrom: yellow |
|
|
colorTo: purple |
|
|
sdk: gradio |
|
|
sdk_version: 5.23.2 |
|
|
app_file: app.py |
|
|
pinned: false |
|
|
--- |
|
|
|
|
|
# Toxic Eye: Multi-Model Toxicity Evaluation Platform |
|
|
|
|
|
## Overview |
|
|
Toxic Eye is a comprehensive platform that evaluates text toxicity using multiple language models and classifiers. This platform provides a unique approach by combining both generative and classification models to analyze potentially toxic content. |
|
|
|
|
|
## Features |
|
|
|
|
|
### 1. Text Generation Models |
|
|
Our platform utilizes four state-of-the-art language models: |
|
|
- **Zephyr-7B**: Specialized in understanding context and nuance |
|
|
- **Llama-2**: Known for its robust performance in content analysis |
|
|
- **Mistral-7B**: Offers precise and detailed text evaluation |
|
|
- **Claude-2**: Provides comprehensive toxicity assessment |
|
|
|
|
|
### 2. Classification Models |
|
|
We employ four specialized classification models: |
|
|
- **Toxic-BERT**: Fine-tuned for toxic content detection |
|
|
- **RoBERTa-Toxic**: Advanced toxic pattern recognition |
|
|
- **DistilBERT-Toxic**: Efficient toxicity classification |
|
|
- **XLM-RoBERTa-Toxic**: Multilingual toxicity detection |
|
|
|
|
|
### 3. Community Integration |
|
|
Access to community insights and discussions about similar content patterns and toxicity analysis. |
|
|
|
|
|
## Technical Details |
|
|
|
|
|
### Model Architecture |
|
|
Each model in our platform is carefully selected to provide complementary analysis: |
|
|
```python |
|
|
def analyze_toxicity(text): |
|
|
# Multiple model evaluation |
|
|
llm_results = text_generation_models(text) |
|
|
classification_results = toxicity_classifiers(text) |
|
|
community_insights = fetch_community_data(text) |
|
|
return combined_analysis(llm_results, classification_results, community_insights) |
|
|
``` |
|
|
|
|
|
### Performance Considerations |
|
|
- Real-time analysis capabilities |
|
|
- Efficient multi-model parallel processing |
|
|
- Optimized response generation |
|
|
|
|
|
## Usage Guidelines |
|
|
|
|
|
1. Enter the text you want to analyze in the input box |
|
|
2. Review results from multiple models |
|
|
3. Compare different model perspectives |
|
|
4. Check community insights for context |
|
|
|
|
|
## References |
|
|
|
|
|
- [Hugging Face Models](https://huggingface.co/models) |
|
|
- [Toxicity Classification Research](https://arxiv.org/abs/2103.00153) |
|
|
- [Language Model Evaluation Methods](https://arxiv.org/abs/2009.07118) |
|
|
|
|
|
## Citation |
|
|
|
|
|
If you use this platform in your research, please cite: |
|
|
```bibtex |
|
|
@software{toxic_eye, |
|
|
title = {Toxic Eye: Multi-Model Toxicity Evaluation Platform}, |
|
|
year = {2024}, |
|
|
publisher = {Hugging Face}, |
|
|
url = {https://huggingface.co/spaces/[your-username]/toxic-eye} |
|
|
} |
|
|
``` |
