fix the in-place modification in unet condition when using controlnet (#2586)

* fix the in-place modification in unet condition when using controlnet, which will cause backprop errors when training

* add clone to mid block

* fix-copies

---------

Co-authored-by: Patrick von Platen <[email protected]>
Co-authored-by: William Berman <[email protected]>

Author: 3 people

src/diffusers/models/unet_2d_condition.py

@@ -598,7 +598,7 @@ def forward(
             for down_block_res_sample, down_block_additional_residual in zip(
                 down_block_res_samples, down_block_additional_residuals
             ):
-                down_block_res_sample += down_block_additional_residual
+                down_block_res_sample = down_block_res_sample + down_block_additional_residual
                 new_down_block_res_samples += (down_block_res_sample,)
 
             down_block_res_samples = new_down_block_res_samples
@@ -614,7 +614,7 @@ def forward(
             )
 
         if mid_block_additional_residual is not None:
-            sample += mid_block_additional_residual
+            sample = sample + mid_block_additional_residual
 
         # 5. up
         for i, upsample_block in enumerate(self.up_blocks):

src/diffusers/pipelines/versatile_diffusion/modeling_text_unet.py

@@ -688,7 +688,7 @@ def forward(
             for down_block_res_sample, down_block_additional_residual in zip(
                 down_block_res_samples, down_block_additional_residuals
             ):
-                down_block_res_sample += down_block_additional_residual
+                down_block_res_sample = down_block_res_sample + down_block_additional_residual
                 new_down_block_res_samples += (down_block_res_sample,)
 
             down_block_res_samples = new_down_block_res_samples
@@ -704,7 +704,7 @@ def forward(
             )
 
         if mid_block_additional_residual is not None:
-            sample += mid_block_additional_residual
+            sample = sample + mid_block_additional_residual
 
         # 5. up
         for i, upsample_block in enumerate(self.up_blocks):