README.md

---
{}
---
# GPT-2 Fine-Tuning on Custom Dataset

## 📌 Overview
This project fine-tunes a **GPT-2** model on a custom dataset extracted from `EU_ACT.pdf`. The model is trained using **transfer learning** and uploaded to **Hugging Face Hub** for further deployment.

## 🚀 Features
- **Uses a pre-trained GPT-2 model** (Transfer Learning)
- **Processes text data from a PDF**
- **Tokenizes and fine-tunes the model**
- **Uploads the trained model to Hugging Face**
- **Automatically disables Weights & Biases logging**

## 📂 Project Structure
```
|-- fine_tune_gpt2.py   # Main script for training
|-- EU_ACT.pdf          # Custom dataset (input PDF)
|-- README.md           # Documentation
|-- image.png           # Hugging Face Metadata UI (optional)
```

## 🔧 Installation
Run the following command to install required libraries:
```bash
pip install transformers datasets torch tokenizers accelerate huggingface_hub
```

## 🔑 Hugging Face Authentication
Replace `'your-api-key-here'` with your actual Hugging Face token:
```python
from huggingface_hub import login
login(token='your-api-key-here')
```

## 🏋️‍♂️ Training the Model
Run the script to start training:
```bash
python fine_tune_gpt2.py
```

## 📤 Uploading the Model
After training, the model is automatically uploaded to **Hugging Face Hub**. Ensure you have set:
```python
hf_username = "your_hf_username"  # Replace with your HF username
repo_name = "gpt2-Transfer-euact"
```

## Pipeline
```python
from transformers import pipeline

# Load fine-tuned model from Hugging Face Hub
repo_name = "sssdddwd/gpt2-Transfer-euact"  # Replace with your actual repo
model_pipeline = pipeline("text-generation", model=repo_name)

# Generate text
prompt = "The European AI Act aims to"
output = model_pipeline(prompt, max_length=100, num_return_sequences=1)

# Print output
print(output[0]["generated_text"])
```
## 📊 Metadata UI Reference
![Hugging Face Metadata UI](image.png)

## 📌 License
You can add a license by modifying the Hugging Face `license` field.

## 📬 Contact
For issues or improvements, reach out via GitHub or Hugging Face discussions.