[fbsync] make weights deepcopyable (#6883)

Summary:
* make weights deepcopyable

* add test

* test enum member instead of whole enum

Reviewed By: datumbox

Differential Revision: D41020543

fbshipit-source-id: c96c35ad9a87cf40b5c65f5a1c11f842c25dc273

Author: NicolasHug

test/test_extended_models.py

@@ -1,3 +1,4 @@
+import copy
 import os
 
 import pytest
@@ -59,6 +60,25 @@ def test_get_model_weights(name, weight):
     assert models.get_model_weights(name) == weight
 
 
+@pytest.mark.parametrize("copy_fn", [copy.copy, copy.deepcopy])
+@pytest.mark.parametrize(
+    "name",
+    [
+        "resnet50",
+        "retinanet_resnet50_fpn_v2",
+        "raft_large",
+        "quantized_resnet50",
+        "lraspp_mobilenet_v3_large",
+        "mvit_v1_b",
+    ],
+)
+def test_weights_copyable(copy_fn, name):
+    model_weights = models.get_model_weights(name)
+    for weights in list(model_weights):
+        copied_weights = copy_fn(weights)
+        assert copied_weights is weights
+
+
 @pytest.mark.parametrize(
     "module", [models, models.detection, models.quantization, models.segmentation, models.video, models.optical_flow]
 )

torchvision/models/_api.py

@@ -75,6 +75,9 @@ def __getattr__(self, name):
                 return object.__getattribute__(self.value, name)
         return super().__getattr__(name)
 
+    def __deepcopy__(self, memodict=None):
+        return self
+
 
 def get_weight(name: str) -> WeightsEnum:
     """