# BERT-Base-Uncased Quantized Model for Sentiment Analysis for Whistleblower Report Sentiment 

This repository hosts a quantized version of the BERT model, fine-tuned for stock-market-analysis-sentiment-classification tasks. The model has been optimized for efficient deployment while maintaining high accuracy, making it suitable for resource-constrained environments.

## Model Details

- **Model Architecture:** BERT Base Uncased  
- **Task:** Sentiment Analysis for Whistleblower Report Sentiment 
- **Dataset:** Stanford Sentiment Treebank v2 (SST2)  
- **Quantization:** Float16  
- **Fine-tuning Framework:** Hugging Face Transformers  

## Usage

### Installation

```sh
pip install transformers torch
```


### Loading the Model

```python

from transformers import BertForSequenceClassification, BertTokenizer
import torch

# Load quantized model
quantized_model_path = "AventIQ-AI/sentiment-analysis-for-whistleblower-report-sentiment"
quantized_model = BertForSequenceClassification.from_pretrained(quantized_model_path)
quantized_model.eval()  # Set to evaluation mode
quantized_model.half()  # Convert model to FP16

# Load tokenizer
tokenizer = BertTokenizer.from_pretrained("bert-base-uncased")

# Define a test sentence
test_sentence = "Despite repeated warnings to upper management, safety protocols continued to be ignored at the facility. Employees expressed growing concerns, but supervisors dismissed them without proper review. While some improvements were promised, no tangible actions were taken. The general atmosphere became increasingly hostile, with staff fearing retaliation for speaking out. However, a few departments did show minor signs of improvement after anonymous complaints were made."

# Tokenize input
inputs = tokenizer(test_sentence, return_tensors="pt", padding=True, truncation=True, max_length=128)

# Ensure input tensors are in correct dtype
inputs["input_ids"] = inputs["input_ids"].long()  # Convert to long type
inputs["attention_mask"] = inputs["attention_mask"].long()  # Convert to long type

# Make prediction
with torch.no_grad():
    outputs = quantized_model(**inputs)

# Get predicted class
predicted_class = torch.argmax(outputs.logits, dim=1).item()
print(f"Predicted Class: {predicted_class}")


label_mapping = {0: "very_negative", 1: "negative", 2: "neutral", 3: "positive", 4: "very_positive"}  # Example

predicted_label = label_mapping[predicted_class]
print(f"Predicted Label: {predicted_label}")

```

## Performance Metrics

- **Accuracy:** 0.82 

## Fine-Tuning Details

### Dataset

The dataset is taken from Kaggle Stanford Sentiment Treebank v2 (SST2).

### Training

- Number of epochs: 3  
- Batch size: 8  
- Evaluation strategy: epoch  
- Learning rate: 2e-5  

### Quantization

Post-training quantization was applied using PyTorch's built-in quantization framework to reduce the model size and improve inference efficiency.

## Repository Structure

```
.
├── model/               # Contains the quantized model files
├── tokenizer_config/    # Tokenizer configuration and vocabulary files
├── model.safensors/     # Fine Tuned Model
├── README.md            # Model documentation
```

## Limitations

- The model may not generalize well to domains outside the fine-tuning dataset.  
- Quantization may result in minor accuracy degradation compared to full-precision models.  

## Contributing

Contributions are welcome! Feel free to open an issue or submit a pull request if you have suggestions or improvements.