weights_only=True for all the occurences of torch.load

terratorch/io/file.py

@@ -34,10 +34,11 @@ def load_torch_weights(model:nn.Module=None, save_dir: str = None, name: str = N
                 torch.load(
                     os.path.join(save_dir, name),
                     map_location=torch.device(device),
+                    weights_only=True,
                 )
             )
         else:
-            model.load_state_dict(torch.load(os.path.join(save_dir, name), map_location='cpu'))
+            model.load_state_dict(torch.load(os.path.join(save_dir, name), map_location='cpu', weights_only=True))
 
     except Exception:
         print(

terratorch/models/backbones/clay_v1/embedder.py

@@ -67,7 +67,7 @@ def __init__(
 
     def load_clay_weights(self, ckpt_path):
         "Load the weights from the Clay model encoder."
-        ckpt = torch.load(ckpt_path)
+        ckpt = torch.load(ckpt_path, weights_only=True)
         state_dict = ckpt.get("state_dict")
         state_dict = {
             re.sub(r"^model\.encoder\.", "", name): param

terratorch/models/backbones/dofa_vit.py

@@ -141,7 +141,7 @@ def load_dofa_weights(model: nn.Module, ckpt_data: str | None = None,  weights:
             repo_id = ckpt_data.split("/resolve/")[0].replace("https://hf.co/", '')
             filename = ckpt_data.split("/")[-1]
             ckpt_data = huggingface_hub.hf_hub_download(repo_id=repo_id, filename=filename)
-        checkpoint_model = torch.load(ckpt_data, map_location="cpu")
+        checkpoint_model = torch.load(ckpt_data, map_location="cpu", weights_only=True)
 
         for k in ["head.weight", "head.bias"]:
                 if (

terratorch/models/backbones/prithvi_vit.py

@@ -215,7 +215,7 @@ def _create_prithvi(
 
     if ckpt_path is not None:
         # Load model from checkpoint
-        state_dict = torch.load(ckpt_path, map_location="cpu")
+        state_dict = torch.load(ckpt_path, map_location="cpu", weights_only=True)
         state_dict = checkpoint_filter_wrapper_fn(state_dict, model, pretrained_bands, model_bands)
         model.load_state_dict(state_dict, strict=False)
     elif pretrained:
@@ -225,7 +225,7 @@ def _create_prithvi(
             # Load model from Hugging Face
             pretrained_path = hf_hub_download(repo_id=pretrained_weights[variant]["hf_hub_id"],
                                               filename=pretrained_weights[variant]["hf_hub_filename"])
-            state_dict = torch.load(pretrained_path, map_location="cpu")
+            state_dict = torch.load(pretrained_path, map_location="cpu", weights_only=True)
             state_dict = checkpoint_filter_wrapper_fn(state_dict, model, pretrained_bands, model_bands)
             model.load_state_dict(state_dict, strict=True)
         except RuntimeError as e:

terratorch/models/backbones/scalemae.py

@@ -258,7 +258,7 @@ def vit_huge_patch14(**kwargs):
     return model
 
 def load_scalemae_weights(model: nn.Module, ckpt_data: str, model_bands: list[HLSBands], input_size: int = 224) -> nn.Module:
-    checkpoint_model = torch.load(ckpt_data, map_location="cpu")["model"]
+    checkpoint_model = torch.load(ckpt_data, map_location="cpu", weights_only=True)["model"]
     state_dict = model.state_dict()
 
     for k in ["head.weight", "head.bias"]:

terratorch/models/backbones/torchgeo_resnet.py

@@ -448,8 +448,8 @@ def load_resnet_weights(model: nn.Module, model_bands, ckpt_data: str, weights:
             repo_id = ckpt_data.split("/resolve/")[0].replace("https://hf.co/", '')
             filename = ckpt_data.split("/")[-1]
             ckpt_data = huggingface_hub.hf_hub_download(repo_id=repo_id, filename=filename)
-        # checkpoint_model = torch.load(ckpt_data, map_location="cpu")["model"]
-        checkpoint_model = torch.load(ckpt_data, map_location="cpu")
+
+        checkpoint_model = torch.load(ckpt_data, map_location="cpu", weights_only=True)
         state_dict = model.state_dict()
 
         for k in ["fc.weight", "fc.bias"]:

terratorch/models/backbones/torchgeo_swin_satlas.py

@@ -236,8 +236,8 @@ def load_swin_weights(model: nn.Module, model_bands, ckpt_data: str, weights: We
             repo_id = ckpt_data.split("/resolve/")[0].replace("https://hf.co/", '')
             filename = ckpt_data.split("/")[-1]
             ckpt_data = huggingface_hub.hf_hub_download(repo_id=repo_id, filename=filename)
-        # checkpoint_model = torch.load(ckpt_data, map_location="cpu")["model"]
-        checkpoint_model = torch.load(ckpt_data, map_location="cpu")
+
+        checkpoint_model = torch.load(ckpt_data, map_location="cpu", weights_only=True)
         state_dict = model.state_dict()
 
         for k in ["head.weight", "head.bias"]:

terratorch/models/backbones/torchgeo_vit.py

@@ -179,8 +179,8 @@ def load_vit_weights(model: nn.Module, model_bands, ckpt_data: str, weights: Wei
             repo_id = ckpt_data.split("/resolve/")[0].replace("https://hf.co/", '')
             filename = ckpt_data.split("/")[-1]
             ckpt_data = huggingface_hub.hf_hub_download(repo_id=repo_id, filename=filename)
-        # checkpoint_model = torch.load(ckpt_data, map_location="cpu")["model"]
-        checkpoint_model = torch.load(ckpt_data, map_location="cpu")
+
+        checkpoint_model = torch.load(ckpt_data, map_location="cpu", weights_only=True)
         state_dict = model.state_dict()
 
         for k in ["head.weight", "head.bias"]:

terratorch/models/satmae_model_factory.py

@@ -227,9 +227,9 @@ def build_model(
                 backbone: nn.Module = ModelWrapper(model=backbone_template(**backbone_kwargs), kind=backbone_kind)
 
                 if self.CPU_ONLY:
-                    model_dict = torch.load(checkpoint_path, map_location="cpu")
+                    model_dict = torch.load(checkpoint_path, map_location="cpu", weights_only=True)
                 else:
-                    model_dict = torch.load(checkpoint_path)
+                    model_dict = torch.load(checkpoint_path, weights_only=True)
 
 
                 # Filtering parameters from the model state_dict (when necessary)