Make UNet2DConditionOutput pickle-able (#3857)

* add default to unet output to prevent it from being a required arg

* add unit test

* make style

* adjust unit test

* mark as fast test

* adjust assert statement in test

---------

Co-authored-by: Prathik Rao <prathikrao@microsoft.com@orttrainingdev8.d32nl1ml4oruzj4qz3bqlggovf.px.internal.cloudapp.net>
Co-authored-by: root <root@orttrainingdev8.d32nl1ml4oruzj4qz3bqlggovf.px.internal.cloudapp.net>

src/diffusers/models/unet_2d_condition.py

@@ -57,7 +57,7 @@ class UNet2DConditionOutput(BaseOutput):
             The hidden states output conditioned on `encoder_hidden_states` input. Output of last layer of model.
     """
 
-    sample: torch.FloatTensor
+    sample: torch.FloatTensor = None
 
 
 class UNet2DConditionModel(ModelMixin, ConfigMixin, UNet2DConditionLoadersMixin):

tests/models/test_models_unet_2d_condition.py

@@ -13,6 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import copy
 import gc
 import os
 import tempfile
@@ -782,6 +783,22 @@ def test_custom_diffusion_xformers_on_off(self):
         assert (sample - on_sample).abs().max() < 1e-4
         assert (sample - off_sample).abs().max() < 1e-4
 
+    def test_pickle(self):
+        # enable deterministic behavior for gradient checkpointing
+        init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
+
+        init_dict["attention_head_dim"] = (8, 16)
+
+        model = self.model_class(**init_dict)
+        model.to(torch_device)
+
+        with torch.no_grad():
+            sample = model(**inputs_dict).sample
+
+        sample_copy = copy.copy(sample)
+
+        assert (sample - sample_copy).abs().max() < 1e-4
+
 
 @slow
 class UNet2DConditionModelIntegrationTests(unittest.TestCase):