Although researchers are working on proteomics development, protein function prediction (PFP) is still an open challenge that involves identifying the functional annotation of proteins. Even with advancements in deep learning approaches, existing models struggle to balance predictive accuracy and computational efficiency. The proposed solution in this regard helps balance both using a novel multi-attention enhanced lightweight CNN model termed “Attentive-LiteSeqCNN”. While dilated CNNs add the capability to capture both local and long-range dependencies in protein sequences, multi-attention helps refine the processing of dilated convolutions. The test results on the Data2017 dataset show that the proposed model performs much better than current methods, particularly in F-max, while still being efficient in terms of computing power. Best F-max obtained are: 0.535 for BP and 0.660 for MF.