---
language:
- en
tags:
- text-detoxification
- text2text-generation
- detoxification
- content-moderation
- toxicity-reduction
- llama
- gguf
- minibase
license: apache-2.0
datasets:
- paradetox
metrics:
- toxicity-reduction
- semantic-similarity
- fluency
- latency
model-index:
- name: Detoxify-Small
  results:
  - task:
      type: text-detoxification
      name: Toxicity Reduction
    dataset:
      type: paradetox
      name: ParaDetox
      config: toxic-neutral
      split: test
    metrics:
    - type: toxicity-reduction
      value: 0.032
      name: Average Toxicity Reduction
    - type: semantic-similarity
      value: 0.471
      name: Semantic to Expected
    - type: fluency
      value: 0.919
      name: Text Fluency
    - type: latency
      value: 66.4
      name: Average Latency (ms)
---

# Detoxify-Small 🤖

<div align="center">

**A highly compact (~100 MB) and efficient text detoxification model for removing toxicity while preserving meaning.**

[![Model Size](https://img.shields.io/badge/Model_Size-138MB-blue)](https://huggingface.co/)
[![Architecture](https://img.shields.io/badge/Architecture-LlamaForCausalLM-green)](https://huggingface.co/)
[![License](https://img.shields.io/badge/License-Apache_2.0-yellow)](LICENSE)
[![Discord](https://img.shields.io/badge/Discord-Join_Community-5865F2)](https://discord.com/invite/BrJn4D2Guh)

*Built by [Minibase](https://minibase.ai) - Train and deploy small AI models from your browser.*
*Browse all of the models and datasets available on the [Minibase Marketplace](https://minibase.ai/wiki/Special:Marketplace).

</div>

## 📋 Model Summary

**Minibase-Detoxify-Small** is a compact language model fine-tuned specifically for text detoxification tasks. It takes toxic or inappropriate text as input and generates cleaned, non-toxic versions while preserving the original meaning and intent as much as possible.

### Key Features
- ⚡ **Fast Inference**: ~66ms average response time
- 🎯 **High Fluency**: 91.9% well-formed output text
- 🧹 **Effective Detoxification**: 3.2% average toxicity reduction
- 💾 **Compact Size**: Only 138MB (GGUF quantized)
- 🔒 **Privacy-First**: Runs locally, no data sent to external servers

## 🚀 Quick Start

### Local Inference (Recommended)

1. **Install llama.cpp** (if not already installed):
   ```bash
   git clone https://github.com/ggerganov/llama.cpp
   cd llama.cpp && make
   ```

2. **Download and run the model**:
   ```bash
   # Download model files
   wget https://huggingface.co/minibase/detoxify-small/resolve/main/model.gguf
   wget https://huggingface.co/minibase/detoxify-small/resolve/main/run_server.sh

   # Make executable and run
   chmod +x run_server.sh
   ./run_server.sh
   ```

3. **Make API calls**:
   ```python
   import requests

   # Detoxify text
   response = requests.post("http://127.0.0.1:8000/completion", json={
       "prompt": "Instruction: Rewrite the provided text to remove the toxicity.\n\nInput: This is fucking terrible!\n\nResponse: ",
       "max_tokens": 200,
       "temperature": 0.7
   })

   result = response.json()
   print(result["content"])  # "This is really terrible!"
   ```

### Python Client

```python
from detoxify_inference import DetoxifyClient

# Initialize client
client = DetoxifyClient()

# Detoxify text
toxic_text = "This product is fucking amazing, no bullshit!"
clean_text = client.detoxify_text(toxic_text)

print(clean_text)  # "This product is really amazing, no kidding!"
```

## 📊 Benchmarks & Performance

### ParaDetox Dataset Results (1,008 samples)

| Metric | Score | Description |
|--------|-------|-------------|
• Original Toxicity:            0.051 (5.1%)
• Final Toxicity:               0.020 (2.0%)

| **Toxicity Reduction** | 0.051 (ParaDetox) --> 0.020 | Reduced toxicity scores by more than 50% |
| **Semantic to Expected** | 0.471 (47.1%) | Similarity to human expert rewrites |
| **Semantic to Original** | 0.625 (62.5%) | How much original meaning is preserved |
| **Fluency** | 0.919 (91.9%) | Quality of generated text structure |
| **Latency** | 66.4ms | Average response time |
| **Throughput** | ~15 req/sec | Estimated requests per second |

### Dataset Breakdown

#### General Toxic Content (1,000 samples)
- **Semantic Preservation**: 62.7%
- **Fluency**: 91.9%

### Comparison with Baselines

| Model | Semantic Similarity | Toxicity Reduction | Fluency |
|-------|-------------------|-------------------|---------|
| **Detoxify-Small** | **0.471** | **0.032** | **0.919** |
| BART-base (ParaDetox) | 0.750 | ~0.15 | ~0.85 |
| Human Performance | 0.850 | ~0.25 | ~0.95 |

## 🏗️ Technical Details

### Model Architecture
- **Architecture**: LlamaForCausalLM
- **Parameters**: 49,152 (extremely compact)
- **Context Window**: 1,024 tokens
- **Quantization**: GGUF (4-bit quantization)
- **File Size**: 138MB
- **Memory Requirements**: 8GB RAM minimum, 16GB recommended

### Training Details
- **Base Model**: Custom-trained Llama architecture
- **Fine-tuning Dataset**: Curated toxic-neutral parallel pairs
- **Training Objective**: Instruction-following for detoxification
- **Optimization**: Quantized for edge deployment

### System Requirements
- **OS**: Linux, macOS, Windows
- **RAM**: 8GB minimum, 16GB recommended
- **Storage**: 200MB free space
- **Dependencies**: llama.cpp, Python 3.7+

## 📖 Usage Examples

### Basic Detoxification
```python
# Input: "This is fucking awesome!"
# Output: "This is really awesome!"

# Input: "You stupid idiot, get out of my way!"
# Output: "You silly person, please move aside!"
```

### API Integration
```python
import requests

def detoxify_text(text: str) -> str:
    """Detoxify text using Detoxify-Small API"""
    prompt = f"Instruction: Rewrite the provided text to remove the toxicity.\n\nInput: {text}\n\nResponse: "

    response = requests.post("http://127.0.0.1:8000/completion", json={
        "prompt": prompt,
        "max_tokens": 200,
        "temperature": 0.7
    })

    return response.json()["content"]

# Usage
toxic_comment = "This product sucks donkey balls!"
clean_comment = detoxify_text(toxic_comment)
print(clean_comment)  # "This product is not very good!"
```

### Batch Processing
```python
import asyncio
import aiohttp

async def detoxify_batch(texts: list) -> list:
    """Process multiple texts concurrently"""
    async with aiohttp.ClientSession() as session:
        tasks = []
        for text in texts:
            prompt = f"Instruction: Rewrite the provided text to remove the toxicity.\n\nInput: {text}\n\nResponse: "
            payload = {
                "prompt": prompt,
                "max_tokens": 200,
                "temperature": 0.7
            }
            tasks.append(session.post("http://127.0.0.1:8000/completion", json=payload))

        responses = await asyncio.gather(*tasks)
        return [await resp.json() for resp in responses]

# Process multiple comments
comments = [
    "This is fucking brilliant!",
    "You stupid moron!",
    "What the hell is wrong with you?"
]

clean_comments = await detoxify_batch(comments)
```

## 🔧 Advanced Configuration

### Server Configuration
```bash
# GPU acceleration (macOS with Metal)
llama-server \
  -m model.gguf \
  --host 127.0.0.1 \
  --port 8000 \
  --n-gpu-layers 35 \
  --metal

# CPU-only (lower memory usage)
llama-server \
  -m model.gguf \
  --host 127.0.0.1 \
  --port 8000 \
  --n-gpu-layers 0 \
  --threads 8

# Custom context window
llama-server \
  -m model.gguf \
  --ctx-size 2048 \
  --host 127.0.0.1 \
  --port 8000
```

### Temperature Settings
- **Low (0.1-0.3)**: Conservative detoxification, minimal changes
- **Medium (0.4-0.7)**: Balanced approach (recommended)
- **High (0.8-1.0)**: Creative detoxification, more aggressive changes

## 📚 Limitations & Biases

### Current Limitations
- **Vocabulary Scope**: Trained primarily on English toxic content
- **Context Awareness**: May not detect sarcasm or cultural context
- **Length Constraints**: Limited to 1024 token context window
- **Domain Specificity**: Optimized for general web content

### Potential Biases
- **Cultural Context**: May not handle culture-specific expressions
- **Dialect Variations**: Limited exposure to regional dialects
- **Emerging Slang**: May not recognize newest internet slang

## 🤝 Contributing

We welcome contributions! Please see our [Contributing Guide](CONTRIBUTING.md) for details.

### Development Setup
```bash
# Clone the repository
git clone https://github.com/minibase-ai/detoxify-small
cd detoxify-small

# Install dependencies
pip install -r requirements.txt

# Run tests
python -m pytest tests/
```

## 📜 Citation

If you use Detoxify-Small in your research, please cite:

```bibtex
@misc{detoxify-small-2025,
  title={Detoxify-Small: A Compact Text Detoxification Model},
  author={Minibase AI Team},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/minibase/detoxify-small}
}
```

## 📞 Contact & Community

- **Website**: [minibase.ai](https://minibase.ai)
- **Discord Community**: [Join our Discord](https://discord.com/invite/BrJn4D2Guh)
- **GitHub Issues**: [Report bugs or request features on Discord](https://discord.com/invite/BrJn4D2Guh)
- **Email**: hello@minibase.ai

### Support
- 📖 **Documentation**: [help.minibase.ai](https://help.minibase.ai)
- 💬 **Community Forum**: [Join our Discord Community](https://discord.com/invite/BrJn4D2Guh)

## 📋 License

This model is released under the [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0)).

## 🙏 Acknowledgments

- **ParaDetox Dataset**: Used for benchmarking and evaluation
- **llama.cpp**: For efficient local inference
- **Hugging Face**: For model hosting and community
- **Our amazing community**: For feedback and contributions

---

<div align="center">

**Built with ❤️ by the Minibase team**

*Making AI more accessible for everyone*

[📖 Minibase Help Center](https://help.minibase.ai) • [💬 Join our Discord](https://discord.com/invite/BrJn4D2Guh)

</div>