fix: fixed the issues with the alexnet, unet and resnet models due to…

… changes in the ivy.Module and transpose convolutions

ivy_models/alexnet/alexnet.py

@@ -85,5 +85,5 @@ def alexnet(pretrained=True, num_classes=1000, dropout=0, data_format="NCHW"):
         w_clean = ivy_models.helpers.load_torch_weights(
             url, model, custom_mapping=_alexnet_torch_weights_mapping
         )
-        model.v = w_clean
+        model._v = w_clean
     return model

ivy_models/resnet/resnet.py

@@ -182,7 +182,7 @@ def resnet_18(pretrained=True):
             raw_keys_to_prune=["num_batches_tracked"],
             custom_mapping=_resnet_torch_weights_mapping,
         )
-        model.v = w_clean
+        model._v = w_clean
     return model
 
 
@@ -197,7 +197,7 @@ def resnet_34(pretrained=True):
             raw_keys_to_prune=["num_batches_tracked"],
             custom_mapping=_resnet_torch_weights_mapping,
         )
-        model.v = w_clean
+        model._v = w_clean
     return model
 
 
@@ -212,7 +212,7 @@ def resnet_50(pretrained=True):
             raw_keys_to_prune=["num_batches_tracked"],
             custom_mapping=_resnet_torch_weights_mapping,
         )
-        model.v = w_clean
+        model._v = w_clean
     return model
 
 
@@ -227,7 +227,7 @@ def resnet_101(pretrained=True):
             raw_keys_to_prune=["num_batches_tracked"],
             custom_mapping=_resnet_torch_weights_mapping,
         )
-        model.v = w_clean
+        model._v = w_clean
     return model
 
 
@@ -242,5 +242,5 @@ def resnet_152(pretrained=True):
             raw_keys_to_prune=["num_batches_tracked"],
             custom_mapping=_resnet_torch_weights_mapping,
         )
-        model.v = w_clean
+        model._v = w_clean
     return model

ivy_models/unet/unet.py

@@ -2,6 +2,7 @@
 from ivy_models.unet.layers import UNetDoubleConv, UNetDown, UNetOutConv, UNetUp
 
 import builtins
+import re
 from ivy_models.helpers import load_torch_weights
 from ivy_models.base import BaseSpec, BaseModel
 
@@ -66,9 +67,9 @@ def _unet_torch_weights_mapping(old_key, new_key):
         "conv/weight",
     ]
 
-    if "up/weight" in old_key:
-        new_mapping = {"key_chain": new_key, "pattern": "b c h w -> h w b c"}
-    elif builtins.any([kc in old_key for kc in W_KEY]):
+    if builtins.any([kc in old_key for kc in W_KEY]) or re.match(
+        "up\d/((conv/double_conv/0)|(up))/weight", old_key
+    ):
         new_mapping = {"key_chain": new_key, "pattern": "b c h w -> h w c b"}
     elif "conv/bias" in old_key or "up/bias" in old_key:
         new_mapping = {"key_chain": new_key, "pattern": "h -> 1 1 1 h"}
@@ -86,7 +87,7 @@ def unet_carvana(n_channels=3, n_classes=2, v=None, pretrained=True):
             raw_keys_to_prune=["num_batches_tracked"],
             custom_mapping=_unet_torch_weights_mapping,
         )
-        model.v = w_clean
+        model._v = w_clean
     return model