Gypsum board is an emerging wall material in the construction industry, but defects such as bubbles and slurry leakage during production pose serious risks to building safety. However, detecting these defects in real-time on edge devices remains a significant challenge due to the computational demands and limited research on lightweight networks specifically for gypsum board defect detection. To address these issues, we propose a lightweight defect detection method for gypsum board, termed RepViT-YOLO. Built upon the YOLOv8 framework, the approach incorporates a coordinate attention-enhanced RepViT (Reparameterization Vision Transformer) backbone to strengthen local feature perception and spatial localization while preserving lightweight efficiency. Furthermore, a novel C2f_RepCoABlock is embedded in the neck network to facilitate efficient multi-scale feature fusion. Compared with the YOLOv8 baseline, the proposed method achieves 0.9% and 1.0% improvements in mAP@50 and mAP@50:95, respectively, while reducing GFLOPs by 34.51%, model size by 42.67%, and parameters by 44.25%, demonstrating strong potential practical applicability.