Hotfix: revert loading with weights_only=True (#1913)

Author: araffin

docs/misc/changelog.rst

@@ -33,11 +33,23 @@ Others:
 Bug Fixes:
 ^^^^^^^^^^
 
+Documentation:
+^^^^^^^^^^^^^^
+
+Release 2.3.2 (2024-04-27)
+--------------------------
+
+Bug Fixes:
+^^^^^^^^^^
+- Reverted ``torch.load()`` to be called ``weights_only=False`` as it caused loading issue with old version of PyTorch.
+
+
 Documentation:
 ^^^^^^^^^^^^^^
 - Added ER-MRL to the project page (@corentinlger)
 - Updated Tensorboard Logging Videos documentation (@NickLucche)
 
+
 Release 2.3.1 (2024-04-22)
 --------------------------
 
@@ -55,6 +67,11 @@ Release 2.3.0 (2024-03-31)
 
 **New defaults hyperparameters for DDPG, TD3 and DQN**
 
+.. warning::
+
+  Because of ``weights_only=True``, this release breaks loading of policies when using PyTorch 1.13.
+  Please upgrade to PyTorch >= 2.0 or upgrade SB3 version (we reverted the change in SB3 2.3.2)
+
 
 Breaking Changes:
 ^^^^^^^^^^^^^^^^^

stable_baselines3/common/save_util.py

@@ -447,7 +447,8 @@ def load_from_zip_file(
                     file_content.seek(0)
                     # Load the parameters with the right ``map_location``.
                     # Remove ".pth" ending with splitext
-                    th_object = th.load(file_content, map_location=device, weights_only=True)
+                    # Note(antonin): we cannot use weights_only=True, as it breaks with PyTorch 1.13, see GH#1911
+                    th_object = th.load(file_content, map_location=device, weights_only=False)
                     # "tensors.pth" was renamed "pytorch_variables.pth" in v0.9.0, see PR #138
                     if file_path == "pytorch_variables.pth" or file_path == "tensors.pth":
                         # PyTorch variables (not state_dicts)

stable_baselines3/version.txt

@@ -1 +1 @@
-2.4.0a0
+2.3.2