It's usually desirable to further trainings based on previously trained models. What checkpoint models or states from previous training should be loaded to continue the training? Policy, critic, and world model? Should the replay buffer be saved and loaded too for continuous training?
Hello,
It's usually desirable to further trainings based on previously trained models. What checkpoint models or states from previous training should be loaded to continue the training? Policy, critic, and world model? Should the replay buffer be saved and loaded too for continuous training?
Thanks!