[RLLIB] restore_1_of_n_agents_from_checkpoint does not actually seem to work

[Phirefly9]

What happened + What you expected to happen

My team tried to use the example code
https://github.com/ray-project/ray/blob/ray-2.2.0/rllib/examples/restore_1_of_n_agents_from_checkpoint.py

as a basis for loading a pre-trained policy, however we think the example callback is missing a call to sync the worker weights

once we did that everything worked fine, but without the sync_weights the policy was random

Versions / Dependencies

ray/rllib 2.2
python 3.10.8

Reproduction script

the adjusted callback in the example should be:

 class RestoreWeightsCallback(DefaultCallbacks):
        def on_algorithm_init(self, *, algorithm: "Algorithm", **kwargs) -> None:
            algorithm.set_weights({"policy_0": restored_policy_0.get_weights()})
            algorithm.workers.sync_weights()

Issue Severity

Low: It annoys or frustrates me.