|
|
--- |
|
|
license: apache-2.0 |
|
|
language: |
|
|
- en |
|
|
metrics: |
|
|
- accuracy |
|
|
- f1 |
|
|
- precision |
|
|
- recall |
|
|
base_model: FacebookAI/roberta-base |
|
|
pipeline_tag: text-classification |
|
|
library_name: transformers |
|
|
tags: |
|
|
- roberta |
|
|
- toxicity-detection |
|
|
- transformers |
|
|
- text-classification |
|
|
- custom-dataset |
|
|
eval_results: |
|
|
eval_accuracy: 0.94 |
|
|
eval_f1: 0.93 |
|
|
eval_precision: 0.95 |
|
|
eval_recall: 0.91 |
|
|
--- |
|
|
|
|
|
# 🛡️ Toxicity-RoBERTa-Base |
|
|
|
|
|
A fine-tuned transformer model built on top of [`roberta-base`](https://huggingface.co/roberta-base) to **detect toxic content** in text — including insults, threats, hate speech, and offensive language. |
|
|
The model is lightweight, accurate, and ideal for **real-time moderation** tasks. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🧩 Use Cases |
|
|
|
|
|
This model is designed to flag toxic messages in: |
|
|
|
|
|
- 🧵 Social media comments and posts |
|
|
- 🛠️ Developer forums and Discord/Slack bots |
|
|
- 🧠 LLM output moderation |
|
|
- 🧩 Community Q&A sites (like Reddit, Stack Overflow) |
|
|
- 🚨 User-generated content platforms (blogs, review sites, games) |
|
|
|
|
|
--- |
|
|
|
|
|
## 🔍 Model Summary |
|
|
|
|
|
| Attribute | Details | |
|
|
|----------------------|-------------------------------------| |
|
|
| Base Architecture | `roberta-base` | |
|
|
| Fine-tuned For | Toxic vs. Non-toxic classification | |
|
|
| Classes | `0 = Non-toxic`, `1 = Toxic` | |
|
|
| Language | English (`en`) | |
|
|
| Data Sources | Custom dataset (multi-domain) | |
|
|
| Framework | 🤗 Transformers | |
|
|
| Total Parameters | ~125M | |
|
|
|
|
|
--- |
|
|
|
|
|
## 📊 Performance |
|
|
|
|
|
| Metric | Result | |
|
|
|--------------|--------| |
|
|
| Accuracy | 94% | |
|
|
| F1 Score | 93% | |
|
|
| Precision | 95% | |
|
|
| Recall | 91% | |
|
|
|
|
|
--- |
|
|
|
|
|
## ⚙️ Quick Start |
|
|
|
|
|
### 💡 Install Required Libraries |
|
|
|
|
|
```bash |
|
|
pip install transformers torch |
|
|
|
