In future integrated sensing and communication (ISAC)-based vehicular networks, infrastructures such as road-side units (RSUs) are expected to simultaneously support high-efficiency communication and ubiquitous sensing functionalities for multiple vehicles. However, this integration introduces a critical challenge in multi-objective optimization (MOO) to achieve shared sensing and communication precoding (SSCP). To address this challenge, we propose a multi-task learning (MTL) scheme to solve the MOO problem. Specifically, we design a Transformer-based hybrid data-knowledge-driven network (Transformer-HDKDN), which enhances communication learning efficiency by incorporating domain knowledge. Furthermore, we introduce a gradient and uncertainty weighting (GUW) algorithm to dynamically balance the loss functions of the communication and sensing subtasks. Simulation results demonstrate that the proposed MTL scheme outperforms the evaluated schemes in terms of SSCP performance.