Cleanup

Author: bmind7

ml-agents/mlagents/trainers/policy/torch_policy.py

@@ -109,7 +109,6 @@ def evaluate(
         if "log_probs" in run_out:
             run_out["log_probs"] = run_out["log_probs"].to_log_probs_tuple()
         if "entropy" in run_out:
-            # Ensure entropy is detached and moved to CPU before NumPy conversion
             run_out["entropy"] = ModelUtils.to_numpy(run_out["entropy"])
         if self.use_recurrent:
             run_out["memory_out"] = ModelUtils.to_numpy(memories).squeeze(0)

ml-agents/mlagents/trainers/trainer_controller.py

@@ -28,7 +28,6 @@
 from mlagents.trainers.behavior_id_utils import BehaviorIdentifiers
 from mlagents.trainers.agent_processor import AgentManager
 from mlagents import torch_utils
-from mlagents.trainers.settings import TorchSettings
 from mlagents.torch_utils.globals import get_rank
 
 
@@ -293,9 +292,6 @@ def join_threads(self, timeout_seconds: float = 1.0) -> None:
                     merge_gauges(thread_timer_stack.gauges)
 
     def trainer_update_func(self, trainer: Trainer) -> None:
-        # Note: Avoid calling torch.set_default_device in worker threads; it can
-        # interfere with PyTorch's global device context manager. The policy and
-        # optimizer code explicitly places tensors on the configured default_device().
         while not self.kill_trainers:
             with hierarchical_timer("trainer_advance"):
                 trainer.advance()