Update final CPU offloading code for more diffusion pipelines (#5589)

* Update final model offload for more pipelines

Add test to ensure all pipeline components are returned to CPU after
execution with model offloading

* Add comment to explain early UNet offload in Text-to-Video pipeline

* Style

src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py

@@ -1604,9 +1604,8 @@ class StableDiffusionXLControlNetInpaintPipeline(
 
         image = self.image_processor.postprocess(image, output_type=output_type)
 
-        # Offload last model to CPU
-        if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
-            self.final_offload_hook.offload()
+        # Offload all models
+        self.maybe_free_model_hooks()
 
         if not return_dict:
             return (image,)

src/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl_img2img.py

@@ -1433,9 +1433,8 @@ class StableDiffusionXLControlNetImg2ImgPipeline(
 
         image = self.image_processor.postprocess(image, output_type=output_type)
 
-        # Offload last model to CPU
-        if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
-            self.final_offload_hook.offload()
+        # Offload all models
+        self.maybe_free_model_hooks()
 
         if not return_dict:
             return (image,)

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_gligen.py

@@ -864,9 +864,8 @@ class StableDiffusionGLIGENPipeline(DiffusionPipeline):
 
         image = self.image_processor.postprocess(image, output_type=output_type, do_denormalize=do_denormalize)
 
-        # Offload last model to CPU
-        if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
-            self.final_offload_hook.offload()
+        # Offload all models
+        self.maybe_free_model_hooks()
 
         if not return_dict:
             return (image, has_nsfw_concept)

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_gligen_text_image.py

@@ -1031,9 +1031,8 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline):
 
         image = self.image_processor.postprocess(image, output_type=output_type, do_denormalize=do_denormalize)
 
-        # Offload last model to CPU
-        if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
-            self.final_offload_hook.offload()
+        # Offload all models
+        self.maybe_free_model_hooks()
 
         if not return_dict:
             return (image, has_nsfw_concept)

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_upscale.py

@@ -820,9 +820,8 @@ class StableDiffusionUpscalePipeline(
         if output_type == "pil" and self.watermarker is not None:
             image = self.watermarker.apply_watermark(image)
 
-        # Offload last model to CPU
-        if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
-            self.final_offload_hook.offload()
+        # Offload all models
+        self.maybe_free_model_hooks()
 
         if not return_dict:
             return (image, has_nsfw_concept)

src/diffusers/pipelines/stable_diffusion/pipeline_stable_unclip.py

@@ -942,9 +942,8 @@ class StableUnCLIPPipeline(DiffusionPipeline, TextualInversionLoaderMixin, LoraL
 
         image = self.image_processor.postprocess(image, output_type=output_type)
 
-        # Offload last model to CPU
-        if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
-            self.final_offload_hook.offload()
+        # Offload all models
+        self.maybe_free_model_hooks()
 
         if not return_dict:
             return (image,)

src/diffusers/pipelines/stable_diffusion/pipeline_stable_unclip_img2img.py

@@ -839,9 +839,8 @@ class StableUnCLIPImg2ImgPipeline(DiffusionPipeline, TextualInversionLoaderMixin
 
         image = self.image_processor.postprocess(image, output_type=output_type)
 
-        # Offload last model to CPU
-        if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
-            self.final_offload_hook.offload()
+        # Offload all models
+        self.maybe_free_model_hooks()
 
         if not return_dict:
             return (image,)

src/diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_xl_adapter.py

@@ -1059,9 +1059,8 @@ class StableDiffusionXLAdapterPipeline(
 
         image = self.image_processor.postprocess(image, output_type=output_type)
 
-        # Offload last model to CPU
-        if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
-            self.final_offload_hook.offload()
+        # Offload all models
+        self.maybe_free_model_hooks()
 
         if not return_dict:
             return (image,)

src/diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_synth_img2img.py

@@ -777,6 +777,7 @@ class VideoToVideoSDPipeline(DiffusionPipeline, TextualInversionLoaderMixin, Lor
         if output_type == "latent":
             return TextToVideoSDPipelineOutput(frames=latents)
 
+        # manually for max memory savings
         if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
             self.unet.to("cpu")
 

tests/pipelines/test_pipelines_common.py

@@ -742,6 +742,14 @@ class PipelineTesterMixin:
 
         max_diff = np.abs(to_np(output_with_offload) - to_np(output_without_offload)).max()
         self.assertLess(max_diff, expected_max_diff, "CPU offloading should not affect the inference results")
+        self.assertTrue(
+            all(
+                v.device == "cpu"
+                for k, v in pipe.components.values()
+                if isinstance(v, torch.nn.Module) and k not in pipe._exclude_from_cpu_offload
+            ),
+            "CPU offloading should leave all pipeline components on the CPU after inference",
+        )
 
     @unittest.skipIf(
         torch_device != "cuda" or not is_xformers_available(),