fix entropy_sum after merge

ml-agents/mlagents/trainers/torch/utils.py

             all_probs = None
         else:
             all_probs = torch.cat(all_probs_list, dim=-1)
-        entropy_sum = torch.sum(entropies, dim=1)
-        return log_probs, entropy_sum, all_probs
+        return log_probs, entropies, all_probs
 
     @staticmethod
     def masked_mean(tensor: torch.Tensor, masks: torch.Tensor) -> torch.Tensor: