Sentiment analysis of Indonesian e-commerce reviews plays an important role in understanding user perceptions of products and services. Transformer-based models, particularly the Multilingual Text-to-Text Transfer Transformer (mT5), provide an effective text-to-text approach for sentiment classification tasks. However, model performance is highly influenced by the selection of training hyperparameters. Therefore, this study aims to analyze the effect of batch size variation and number of training epochs on the performance of the mT5 model in Tokopedia review sentiment classification. The dataset consists of Indonesian reviews categorized into two sentiment classes, positive and negative. The research stages include preprocessing, SentencePiece tokenization, mT5 model training, and evaluation using accuracy, precision, recall, and F1- score metrics. The experimental results show that a smaller batch size tends to produce more stable performance but requires longer training time, while a larger batch size requires more epochs to reach optimal convergence. The configuration using batch size 4 and 10 epochs achieves the best performance with a training accuracy of 0.963. These findings highlight the importance of proper hyperparameter selection in optimizing Transformerbased models for Indonesian sentiment analysis.