ModeSeq: Taming Sparse Multimodal Motion Prediction with Sequential Mode Modeling | Read Paper on Bytez