What is the difference between alg.save and alg.save_checkpoint()

Can anyone tell the difference between of the two function. I prefer to use save_checkpoint().

when we save the checkpoint, doest it means we also the model states which include everything about the model state and training config setups?

I figured this out. algo.save is usually used if inference in needed.

What exactly do you mean by “inference” and which circumstances must be true so that it is needed?