Description: Dialoguify is a state-of-the-art dialogue summarization tool powered by a fine-tuned Large Language Model (LLM) to efficiently summarize conversations between two individuals. Leveraging advanced NLP techniques and reinforcement learning, Dialoguify ensures high-quality, concise summaries while maintaining ethical standards by reducing toxic content.
Key Features:
- Large Language Model (LLM) Fine-Tuning: Utilizes FLAN-T5 for accurate dialogue summarization.
- Parameter Efficient Fine Tuning (PEFT): Implements Lower Rank Adaptations (LoRA) to enhance model performance, achieving a ROUGE-1 score of 40.81%.
- Toxicity Mitigation: Employs Proximal Policy Optimization (PPO) to minimize toxicity in conversational content, ensuring a safe and user-friendly experience.
Technologies Used:
- Python
- PyTorch
- Hugging Face
Project Highlights:
- FLAN-T5 Fine-Tuning: Customized the FLAN-T5 model to effectively summarize dialogues.
- Lower Rank Adaptations (LoRA): Enhanced model efficiency and performance through PEFT techniques.
- Proximal Policy Optimization (PPO): Applied reinforcement learning to reduce toxicity and improve the ethical standards of generated summaries.
Dialoguify showcases the integration of advanced language models and reinforcement learning to deliver high-quality dialogue summarizations with a focus on ethical AI practices.