**🧠 SMSDetection-DistilBERT-SMS**

A DistilBERT-based binary classifier fine-tuned on the SMS Spam Collection dataset. It classifies messages as either **spam** or **ham** (not spam). This model is suitable for real-world applications like mobile SMS spam filters, automated customer message triage, and telecom fraud detection.

---

✨ **Model Highlights**

- 📌  Based on `distilbert-base-uncased` 
- 🔍 Fine-tuned on the SMS Spam Collection dataset  
- ⚡ Supports binary classification: Spam vs Not Spam
- 💾 Lightweight and optimized for both CPU and GPU environments 

---

🧠 Intended Uses

- ✅ Mobile SMS spam filtering 
- ✅ Telecom customer service automation 
- ✅ Fraudulent message detection 
- ✅ User inbox categorization 
- ✅ Regulatory compliance monitoring
  
---
- 🚫 Limitations

- ❌ Trained on English SMS messages only  
- ❌ May underperform on emails, social media texts, or non-English content  
- ❌ Not designed for multilingual datasets 
- ❌ Slight performance dip expected for long messages (>128 tokens)  

---

🏋️‍♂️ Training Details

| Field          | Value                          |
| -------------- | ------------------------------ |
| **Base Model** | `distilbert-base-uncased`      |
| **Dataset**    |SMS Spam Collection (UCI)       |
| **Framework**  | PyTorch with 🤗 Transformers   |
| **Epochs**     | 3                              |
| **Batch Size** | 16                             |
| **Max Length** | 128 tokens                     |
| **Optimizer**  | AdamW                          |
| **Loss**       | CrossEntropyLoss (token-level) |
| **Device**     | Trained on CUDA-enabled GPU    |

---

📊 Evaluation Metrics

| Metric                                          | Score |
| ----------------------------------------------- | ----- |
| Accuracy                                        | 0.99  |
| F1-Score                                        | 0.96  |
| Precision                                       | 0.98  |
| Recall                                          | 0.93  |


---


---
🚀 Usage
```python
from transformers import BertTokenizerFast, BertForTokenClassification
from transformers import pipeline
import torch

model_name = "AventIQ-AI/SMS-Spam-Detection-Model"
tokenizer = BertTokenizerFast.from_pretrained(model_name)
model = BertForTokenClassification.from_pretrained(model_name)
model.eval()


# Inference
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

def predict_sms(text):
    inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=128)
    inputs = {k: v.to(device) for k, v in inputs.items()}
    with torch.no_grad():
        outputs = model(**inputs)
        logits = outputs.logits
        predicted = torch.argmax(logits, dim=1).item()
    return "spam" if predicted == 1 else "ham"

# Test example
print(predict_sms("You've won $1,000,000! Call now to claim your prize!"))

```
---

- 🧩 Quantization
- Post-training static quantization applied using PyTorch to reduce model size and accelerate inference on edge devices.

----

🗂 Repository Structure
```
.
├── model/               # Quantized model files
├── tokenizer_config/    # Tokenizer and vocab files
├── model.safensors/     # Fine-tuned model in safetensors format
├── README.md            # Model card

```
---
🤝 Contributing

Open to improvements and feedback! Feel free to submit a pull request or open an issue if you find any bugs or want to enhance the model.