# ModernBERT-BizClassifier - Business Relevance Classifier

## 📌 Overview
ModernBERT-BizClassifier is a **fine-tuned ModernBERT model** that classifies text from **any source** (social media, news, reports, etc.) and determines whether it is **useful for businesses**. 

The model distinguishes between:
- **Business_Action_Direct**: Clearly relevant business information (e.g., financial reports, earnings, funding rounds).
- **Business_Intelligence_Indirect**: Text that implies business relevance (e.g., regulatory news, economic trends, emerging markets).
- **Not_Business_Relevant**: Text that contains no useful business insights.

## 🛠 Model Details

| Hyperparameter         | Value |
|-----------------------|----------------|
| **Base Model**       | `ModernBERT-base` |
| **Max Context Length** | `8192 tokens` |
| **Batch Size**       | `32` |
| **Learning Rate**    | `1e-5` |
| **Weight Decay**     | `0.005` |
| **Epochs**          | `2.0` |
| **Warmup Ratio**     | `0.1` |
| **Precision Used**   | `BF16` if CUDA available |
| **Problem Type**     | `Multi-label classification` |

## 📊 Evaluation Results

| Metric                      | Score  |
|-----------------------------|--------|
| **Final Validation Loss**   | `0.2940` |
| **Micro F1 Score**          | `0.8049` |
| **Macro F1 Score**          | `0.8042` |
| **Evaluation Runtime**      | `3.48 seconds` |
| **Samples Processed Per Second** | `239.09` |
| **Steps Processed Per Second** | `7.48` |

## 🏗 Model Card
- **Tags**: `business-classification`, `BERT`, `financial-analysis`, `text-classification`

```python
{
    1: "Business_Action_Direct",
    2: "Business_Intelligence_Indirect",
    3: "Not_Business_Relevant"
}
```