README.md

# GPT-2 for Storytelling

This repository hosts a quantized version of the GPT-2 model, fine-tuned for creative writing and storytelling tasks. The model has been optimized for efficient deployment while maintaining high coherence and creativity, making it suitable for resource-constrained environments.

## Model Details

- **Model Architecture:** gpt2-lmheadmodel-story-telling-model
- **Task:** Storytelling & Writing Prompts Generation  
- **Dataset:**  euclaise/writingprompts 
- **Quantization:** Float16  
- **Fine-tuning Framework:** Hugging Face Transformers  

## Usage

### Installation

```sh
pip install transformers torch
```

### Loading the Model

```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

device = "cuda" if torch.cuda.is_available() else "cpu"

model_name = "AventIQ-AI/gpt2-lmheadmodel-story-telling-model"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name).to(device)

import html

# Define test text
test_text = "Once upon a time, in a mystical land,"

# Tokenize input
inputs = tokenizer(test_text, return_tensors="pt").to(device)

# Generate response
with torch.no_grad():
    output_tokens = model.generate(
        **inputs,
        max_length=200, 
        num_beams=5,
        repetition_penalty=2.0, 
        temperature=0.7, 
        top_k=50, 
        top_p=0.9, 
        do_sample=True, 
        no_repeat_ngram_size=3, 
        num_return_sequences=1,  
        early_stopping=True, 
        length_penalty=1.2,  
        pad_token_id=tokenizer.eos_token_id, 
        eos_token_id=tokenizer.eos_token_id, 
        return_dict_in_generate=True,  
        output_scores=True  
    )

# Decode and clean response
generated_response = tokenizer.decode(output_tokens.sequences[0], skip_special_tokens=True)
cleaned_response = html.unescape(generated_response).replace("#39;", "'").replace("quot;", '"')

print("\nGenerated Response:\n", cleaned_response)
```

# 📊 ROUGE Evaluation Results

After fine-tuning the **GPT-2** model for storytelling, we obtained the following **ROUGE** scores:

| **Metric**  | **Score**  | **Meaning** |
|-------------|-----------|-------------|
| **ROUGE-1** | **0.7525** (~75%) | Measures overlap of **unigrams (single words)** between the reference and generated text. |
| **ROUGE-2** | **0.3552** (~35%) | Measures overlap of **bigrams (two-word phrases)**, indicating coherence and fluency. |
| **ROUGE-L** | **0.4904** (~49%) | Measures **longest matching word sequences**, testing sentence structure preservation. |
| **ROUGE-Lsum** | **0.5701** (~57%) | Similar to ROUGE-L but optimized for storytelling tasks. |


## Fine-Tuning Details

### Dataset

The Hugging Face euclaise/writingprompts dataset was used, containing creative writing prompts and responses.

### Training

- Number of epochs: 3  
- Batch size: 4  
- Evaluation strategy: epoch  
- Learning rate: 5e-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.safetensors/   # Quantized 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.