Reinforcement Learning Guided Fine-Tuning Protocols for Transformer-Based Large Language Models
International Journal of Scientific Research in Science, Engineering and Technology · 30.03.2025 · 12 (2) · 895-903
DOI: 10.32628/ijsrset251222018 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Scientific Research in Science, Engineering and Technology, online erschienen am 30.03.2025. Verfasst von Navya Veginati. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Transformer-based Large Language Models (LLMs) have become a phenomenal success in the field of natural language processing; nevertheless, they require effective strategies to fine-tune their performance heavily. The article suggests a reinforcement learning-based fine-tuning (RL-FT) model to improve adaptability of models, quality of their output, and their compliance with task-related goals. The suggested methodology combines supervised learning with reinforcement learning through the application of the optimization based on rewards to optimize the responses of the models iteratively. A reward model is utilized to measure the outputs generated by the model through relevance, coherence and accuracy, and policy optimization techniques are utilized to optimize the model parameters. Also, there is a KL-divergence constraint that helps to stabilize training and avoid overfitting with the original model. Simulation-based experimental analysis shows that the suggested RL-FT model is more accurate, converges faster, and is more stable than baseline transformer and knowledge distillation-based models. The results highlight the effectiveness of combining reinforcement learning with fine-tuning for improving the performance of transformer-based LLMs.
Zitieren
APA
Veginati, N. (2025). Reinforcement Learning Guided Fine-Tuning Protocols for Transformer-Based Large Language Models. International Journal of Scientific Research in Science, Engineering and Technology, 12(2), 895-903. https://doi.org/10.32628/ijsrset251222018
Vancouver
Veginati N. Reinforcement Learning Guided Fine-Tuning Protocols for Transformer-Based Large Language Models. International Journal of Scientific Research in Science, Engineering and Technology. 2025;12(2):895-903. doi:10.32628/ijsrset251222018
BibTeX
@article{Veginati2025,
author = {Veginati, Navya},
title = {Reinforcement Learning Guided Fine-Tuning Protocols for Transformer-Based Large Language Models},
journal = {International Journal of Scientific Research in Science, Engineering and Technology},
year = {2025},
volume = {12},
number = {2},
pages = {895--903},
doi = {10.32628/ijsrset251222018}
}
EUVIMED