LoRA-LiteE: A Computationally Efficient Framework for Chatbot Preference-Tuning
It addresses the scalability and accessibility issues of RLHF methods for chatbot alignment, making preference tuning more feasible in resource-constrained environments, though it is incremental as it builds on existing techniques like LoRA and ensemble learning.
This study tackled the problem of computationally intensive preference tuning for chatbots by introducing LoRA-LiteE, a framework combining SFT, LoRA, and ensemble learning, which achieved comparable performance to un-finetuned GPT-4 and outperformed single larger-scale models under limited resources.
Effective preference tuning is pivotal in aligning chatbot responses with human expectations, enhancing user satisfaction and engagement. Traditional approaches, notably Reinforcement Learning from Human Feedback (RLHF) as employed in advanced models like GPT-4, have demonstrated considerable success in this domain. However, RLHF methods are often computationally intensive and resource-demanding, limiting their scalability and accessibility for broader applications. To address these challenges, this study introduces LoRA-Lite Ensemble (LoRA-LiteE), an innovative framework that combines Supervised Fine-tuning (SFT) with Low-Rank Adaptation (LoRA) and Ensemble Learning techniques to effectively aggregate predictions of lightweight models, which aim to achieve a balance between the performance and computational cost. Utilizing the Chatbot Arena benchmark dataset, we conduct a comprehensive comparative analysis among our LoRA-LiteE model, corresponding base models at different scales, and GPT-4 trained with RLHF. Our empirical results demonstrate that the proposed LoRA-LiteE model achieves comparable performance to un-finetuned GPT-4 and outperforms the single larger-scale models under limited resource constraints. These findings highlight that our LoRA-LiteE provides a feasible and efficient methodology for human preference prediction in chatbot systems, enhancing scalability and accessibility, and thereby broadening the applicability of preference-tuned chatbots in resource-constrained environments.