 The size of tensor a (1024) must match the size of tensor b (512), and LLuL does not have the capability to adjust the size