This study explores the use of deep learning for classifying gastric cancer subtypes from whole slide images (WSIs). Given the challenge of class imbalance, the CutMix technique was integrated into the training process to enhance model performance. During inference, a patch-based voting mechanism was applied to determine the final WSI-level classification. The proposed approach was evaluated on both internal and external test datasets, achieving high accuracy (up to 0.99 and 0.94 with a threshold). These findings suggest that the method could aid in the automated classification of gastric cancer subtypes in clinical settings.