Hey @Roller44 , thanks for the question. This looks more like a Ray Tune problem, but I’m assuming, you have to implement the save_checkpoint method for your custom Trainable. In that implementation, you could call save() on your wrapped DDPGTrainer. That being said, I’m not exactly sure why you even have to wrap a DDPGTrainer into a custom trainable in your case. Couldn’t you just use a DDPGTrainer directly in tune? Like tune.run("DDPG", ...)? But I guess you would like to add more customizations, in which case, this is totally fine.