[Question] Why --checkpoint argument is not in the train.py of skrl?

[celestialdr4g0n]

Question

I’m using MAPPO in skrl, but the --checkpoint option seems to be missing. Is there a way to resume training from a checkpoint?

[RandomOakForest]

Thanks for bringing this to our attention. We will add the option for an upcoming release. In the meantime, you could modify your train.py to load a checkpoint as done in the skrl/play.py script here, just before calling runner.run() in skrl/train.py.