from_tortoise_orm generated redundant queries

The code is as follows:

query = User.get_or_none(id=id).prefetch_related('roles')
result = await User_Pydantic.from_tortoise_orm(await query)

** User and Role is M2M relationship.

From the logs, You can see that it executed 3 queries, with the third one being obviously redundant:

# 1st
SELECT `id`,`name`,`phone` FROM `user` WHERE `id`=%s ASC LIMIT %s : [1, 2]

# 2nd
SELECT `sq0`.`_backward_relation_key` `_backward_relation_key`,`role`.`name` `name`,`role`.`code` `code`,`role`.`id` `id`
FROM `role` JOIN (SELECT `user_id` `_backward_relation_key`,`role_id` `_forward_relation_key` FROM `role_user`
WHERE `user_id` IN (%s)) `sq0` ON `sq0`.`_forward_relation_key`=`role`.`id` : [1]

# 3rd
SELECT `sq0`.`_backward_relation_key` `_backward_relation_key`,`role`.`name` `name`,`role`.`code` `code`,`role`.`id` `id`
FROM `role` JOIN (SELECT `user_id` `_backward_relation_key`,`role_id` `_forward_relation_key` FROM `role_user`
WHERE `user_id` IN (%s)) `sq0` ON `sq0`.`_forward_relation_key`=`role`.`id` : [1]

If I return the queryset directly instead of executing from_tortoise_orm, the SQL will be executed twice (properly).
Did I write something incorrectly there?

Of course, the above issue is not a bug, just an efficiency problem.

The report identifies prefetch_related and User_Pydantic.from_tortoise_orm as the relevant entry points; start by reproducing the example and tracing how the prefetched roles are handled during serialization. Done means the same relation is not queried redundantly while from_tortoise_orm still returns the expected related data.