Th Real Joker comments

Repositories
Issues
Comments

Results 1 comments of


                                            Th Real Joker

[BUG] GAE parameters (gamma, lmbda) seemed to get changed by ClipPPOLoss, advantage module does not calculate loss_critic

I found the root of issue, that's because when I connect the policy_forward to collector (`train_kwargs["policy"] = self.policy_forward`), for every iteration (step), the collector spawns another process and runs the...