The wear-levelling is performed by the SSD controller, and basically the controller is unaware of partitions and filesystems.
The important point is that there exists a mapping table between the drive logical sectors (the ones that the OS addresses) and the drive physical sectors in the NAND chips. The wear-levelling and garbage collector algorithms heavily rely on this mapping table.
The logical address space of the drive is possibly partitioned, but the physical address space is not.
For instance, the OS tells the controller to write some data to the logical sector #A which is close to the beginning of the drive from the OS point of view, but the controller can decide to use any physical sector anywhere on the drive. For instance the sector #B that is near the end of the drive if it has a low write count, and it updates the mapping table with a link (logical)#A --> (physical)#B. Then when the OS wants to read the logical sector #A, the controller actually returns the content of the physical sector #B.
Whether the logical sector #B is in the same partition than the logical sector #A or not is irrelevant.
Furthermore the controller can decide at any moment to move the content of the physical sector #B to any other physical sector #C, and he just has to update the mapping table #A --> #C. This is completely transparent for the OS.