Plagiarism is a pervasive challenge in academic and digital landscapes, necessitating effective detection mechanisms. This paper presents an innovative approach utilizing advanced NLP models, specifically RoBERTa and RoBERTa-large, to enhance plagiarism detection capabilities. The seven-step methodology addresses data processing, model training, evaluation, storage, and detection, with a focus on the unique attributes of the RoBERTa-large model. Additionally, the proposed solution adeptly manages token limitations in RoBERTa-large, providing an efficient strategy for handling extended texts in plagiarism detection tasks.  This paper introduces a solution to overcome token limitations in the RoBERTa-large model, enhancing plagiarism detection accuracy. The approach involves systematic text segmentation, tokenization, and result amalgamation for efficient processing of longer texts. It tackles RoBERTa-large’s constraints, ensuring optimal performance in plagiarism detection tasks.