[CI] Fix CI

.github/unittest/install_dependencies.sh

@@ -1,12 +1,10 @@
 
 
 python -m pip install --upgrade pip
-python -m pip install flake8 pytest pytest-cov hydra-core tqdm torch_geometric
+python -m pip install flake8 pytest pytest-cov hydra-core tqdm torch torch_geometric
 
 if [ -f requirements.txt ]; then pip install -r requirements.txt; fi
 
-python -m pip install --pre torch --extra-index-url https://download.pytorch.org/whl/nightly/cpu --force-reinstall
-
 pip install torchrl
 
 cd ../BenchMARL

.github/unittest/install_dependencies_nightly.sh

@@ -5,7 +5,9 @@ python -m pip install flake8 pytest pytest-cov hydra-core tqdm torch_geometric
 
 if [ -f requirements.txt ]; then pip install -r requirements.txt; fi
 
-python -m pip install --pre torch --extra-index-url https://download.pytorch.org/whl/nightly/cpu --force-reinstall
+python -m pip install torch
+# Not using nightly torch
+# python -m pip install --pre torch --extra-index-url https://download.pytorch.org/whl/nightly/cpu --force-reinstall
 
 cd ..
 python -m pip install git+https://github.com/pytorch-labs/tensordict.git

benchmarl/algorithms/isac.py

@@ -18,13 +18,7 @@
     ProbabilisticActor,
     TanhNormal,
 )
-from torchrl.objectives import (
-    ClipPPOLoss,
-    DiscreteSACLoss,
-    LossModule,
-    SACLoss,
-    ValueEstimators,
-)
+from torchrl.objectives import DiscreteSACLoss, LossModule, SACLoss, ValueEstimators
 
 from benchmarl.algorithms.common import Algorithm, AlgorithmConfig
 from benchmarl.models.common import ModelConfig
@@ -149,7 +143,7 @@ def _get_loss(
         )
         return loss_module, True
 
-    def _get_parameters(self, group: str, loss: ClipPPOLoss) -> Dict[str, Iterable]:
+    def _get_parameters(self, group: str, loss: LossModule) -> Dict[str, Iterable]:
         items = {
             "loss_actor": list(loss.actor_network_params.flatten_keys().values()),
             "loss_qvalue": list(loss.qvalue_network_params.flatten_keys().values()),

benchmarl/algorithms/maddpg.py

@@ -220,7 +220,7 @@ def get_value_module(self, group: str) -> TensorDictModule:
             modules.append(
                 TensorDictModule(
                     lambda state, action: torch.cat(
-                        [state, action.view(*action.shape[:-2], -1)], dim=-1
+                        [state, action.reshape(*action.shape[:-2], -1)], dim=-1
                     ),
                     in_keys=["state", (group, "action")],
                     out_keys=["state_action"],

benchmarl/algorithms/masac.py

@@ -383,7 +383,7 @@ def get_continuous_value_module(self, group: str) -> TensorDictModule:
             modules.append(
                 TensorDictModule(
                     lambda state, action: torch.cat(
-                        [state, action.view(*action.shape[:-2], -1)], dim=-1
+                        [state, action.reshape(*action.shape[:-2], -1)], dim=-1
                     ),
                     in_keys=["state", (group, "action")],
                     out_keys=["state_action"],

test/test_vmas.py

@@ -109,7 +109,7 @@ def test_reloading_trainer(
         algo_config = algo_config.get_from_yaml()
         if isinstance(algo_config, VdnConfig):
             # There are some bugs currently in TorchRL https://github.com/pytorch/rl/issues/1593
-            return
+            pytest.skip()
         ExperimentUtils.check_experiment_loading(
             algo_config=algo_config,
             model_config=mlp_sequence_config,