Fix: torch >= 2.6 compatibility & multi-instance file contention

libero/libero/__init__.py

@@ -1,11 +1,14 @@
 import os
+import warnings
 import yaml
+from filelock import FileLock
 
 # This is a default path for localizing all the benchmark related files
 libero_config_path = os.environ.get(
     "LIBERO_CONFIG_PATH", os.path.expanduser("~/.libero")
 )
 config_file = os.path.join(libero_config_path, "config.yaml")
+config_file_lock = os.path.join(libero_config_path, ".config_lock")
 
 
 def get_default_path_dict(custom_location=None):
@@ -39,11 +42,6 @@ def get_libero_path(query_key):
     with open(config_file, "r") as f:
         config = dict(yaml.load(f.read(), Loader=yaml.FullLoader))
 
-    # Give warnings in case the user needs to access the paths
-    for key in config:
-        if not os.path.exists(config[key]):
-            print(f"[Warning]: {key} path {config[key]} does not exist!")
-
     assert (
         query_key in config
     ), f"Key {query_key} not found in config file {config_file}. You need to modify it. Available keys are: {config.keys()}"
@@ -52,45 +50,25 @@ def get_libero_path(query_key):
 
 def set_libero_default_path(custom_location=os.path.dirname(os.path.abspath(__file__))):
     print(
-        f"[Warning] You are changing the default path for Libero config. This will affect all the paths in the config file."
+        "[Warning] You are changing the default path for Libero config. This will affect all the paths in the config file."
     )
     new_config = get_default_path_dict(custom_location)
     with open(config_file, "w") as f:
         yaml.dump(new_config, f)
 
-
-if not os.path.exists(libero_config_path):
-    os.makedirs(libero_config_path)
+with FileLock(config_file_lock):
+    if not os.path.exists(libero_config_path):
+        os.makedirs(libero_config_path)
 
 if not os.path.exists(config_file):
     # Create a default config file
 
     default_path_dict = get_default_path_dict()
-    answer = input(
-        "Do you want to specify a custom path for the dataset folder? (Y/N): "
-    ).lower()
-    if answer == "y":
-        # If the user wants to specify a custom storage path, prompt them to enter it
-        custom_dataset_path = input(
-            "Enter the path where you want to store the datasets: "
-        )
-        full_custom_dataset_path = os.path.join(
-            os.path.abspath(os.path.expanduser(custom_dataset_path)), "datasets"
-        )
-        # Check if the custom storage path exists, and create if it doesn't
-
-        print("The full path of the custom storage path you entered is:")
-        print(full_custom_dataset_path)
-        print("Do you want to continue? (Y/N)")
-        confirm_answer = input().lower()
-        if confirm_answer == "y":
-            if not os.path.exists(full_custom_dataset_path):
-                os.makedirs(full_custom_dataset_path)
-            default_path_dict["datasets"] = full_custom_dataset_path
     print("Initializing the default config file...")
     print(f"The following information is stored in the config file: {config_file}")
     # write all the paths into a yaml file
-    with open(config_file, "w") as f:
-        yaml.dump(default_path_dict, f)
-    for key, value in default_path_dict.items():
-        print(f"{key}: {value}")
+    with FileLock(config_file_lock):
+        with open(config_file, "w") as f:
+            yaml.dump(default_path_dict, f)
+        for key, value in default_path_dict.items():
+            print(f"{key}: {value}")

libero/libero/benchmark/__init__.py

@@ -161,7 +161,7 @@ def get_task_init_states(self, i):
             self.tasks[i].problem_folder,
             self.tasks[i].init_states_file,
         )
-        init_states = torch.load(init_states_path)
+        init_states = torch.load(init_states_path, weights_only=False)
         return init_states
 
     def set_task_embs(self, task_embs):

libero/lifelong/evaluate.py

@@ -251,7 +251,7 @@ def main():
         init_states_path = os.path.join(
             cfg.init_states_folder, task.problem_folder, task.init_states_file
         )
-        init_states = torch.load(init_states_path)
+        init_states = torch.load(init_states_path, weights_only=False)
         indices = np.arange(env_num) % init_states.shape[0]
         init_states_ = init_states[indices]
 

libero/lifelong/metric.py

@@ -104,7 +104,7 @@ def evaluate_one_task_success(
         init_states_path = os.path.join(
             cfg.init_states_folder, task.problem_folder, task.init_states_file
         )
-        init_states = torch.load(init_states_path)
+        init_states = torch.load(init_states_path, weights_only=False)
         num_success = 0
         for i in range(eval_loop_num):
             env.reset()

libero/lifelong/utils.py

@@ -56,7 +56,7 @@ def torch_save_model(model, model_path, cfg=None, previous_masks=None):
 
 
 def torch_load_model(model_path, map_location=None):
-    model_dict = torch.load(model_path, map_location=map_location)
+    model_dict = torch.load(model_path, weights_only=False, map_location=map_location)
     cfg = None
     if "cfg" in model_dict:
         cfg = model_dict["cfg"]