updates

tests/pipelines/animatediff/test_animatediff.py

@@ -547,6 +547,14 @@ def test_xformers_attention_forwardGenerator_pass(self):
     def test_vae_slicing(self):
         return super().test_vae_slicing(image_count=2)
 
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "num_images_per_prompt": 1,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)
+
 
 @slow
 @require_torch_accelerator

tests/pipelines/animatediff/test_animatediff_controlnet.py

@@ -517,3 +517,11 @@ def test_vae_slicing(self, video_count=2):
         output_2 = pipe(**inputs)
 
         assert np.abs(output_2[0].flatten() - output_1[0].flatten()).max() < 1e-2
+
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "num_images_per_prompt": 1,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)

tests/pipelines/animatediff/test_animatediff_sdxl.py

@@ -305,3 +305,7 @@ def test_xformers_attention_forwardGenerator_pass(self):
 
         max_diff = np.abs(to_np(output_with_offload) - to_np(output_without_offload)).max()
         self.assertLess(max_diff, 1e-4, "XFormers attention should not affect the inference results")
+
+    @unittest.skip("Test currently not supported.")
+    def test_encode_prompt_works_in_isolation(self):
+        pass

tests/pipelines/animatediff/test_animatediff_sparsectrl.py

@@ -484,3 +484,11 @@ def test_free_init_with_schedulers(self):
 
     def test_vae_slicing(self):
         return super().test_vae_slicing(image_count=2)
+
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "num_images_per_prompt": 1,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)

tests/pipelines/animatediff/test_animatediff_video2video.py

@@ -544,3 +544,11 @@ def test_free_noise_multi_prompt(self):
             inputs["strength"] = 0.5
             inputs["prompt"] = {0: "Caterpillar on a leaf", 10: "Butterfly on a leaf", 42: "Error on a leaf"}
             pipe(**inputs).frames[0]
+
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "num_images_per_prompt": 1,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)

tests/pipelines/animatediff/test_animatediff_video2video_controlnet.py

@@ -533,3 +533,11 @@ def test_free_noise_multi_prompt(self):
             inputs["strength"] = 0.5
             inputs["prompt"] = {0: "Caterpillar on a leaf", 10: "Butterfly on a leaf", 42: "Error on a leaf"}
             pipe(**inputs).frames[0]
+
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "num_images_per_prompt": 1,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)

tests/pipelines/blipdiffusion/test_blipdiffusion.py

@@ -198,3 +198,7 @@ def test_blipdiffusion(self):
         assert (
             np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
         ), f" expected_slice {image_slice.flatten()}, but got {image_slice.flatten()}"
+
+    @unittest.skip("Test not supported because of complexities in deriving query_embeds.")
+    def test_encode_prompt_works_in_isolation(self):
+        pass

tests/pipelines/controlnet/test_controlnet.py

@@ -287,6 +287,13 @@ def test_controlnet_lcm_custom_timesteps(self):
 
         assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
 
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)
+
 
 class StableDiffusionMultiControlNetPipelineFastTests(
     IPAdapterTesterMixin, PipelineTesterMixin, PipelineKarrasSchedulerTesterMixin, unittest.TestCase
@@ -521,6 +528,13 @@ def test_inference_multiple_prompt_input(self):
 
         assert image.shape == (4, 64, 64, 3)
 
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)
+
 
 class StableDiffusionMultiControlNetOneModelPipelineFastTests(
     IPAdapterTesterMixin, PipelineTesterMixin, PipelineKarrasSchedulerTesterMixin, unittest.TestCase
@@ -706,6 +720,13 @@ def test_save_pretrained_raise_not_implemented_exception(self):
             except NotImplementedError:
                 pass
 
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)
+
 
 @slow
 @require_torch_accelerator

tests/pipelines/controlnet/test_controlnet_blip_diffusion.py

@@ -222,3 +222,7 @@ def test_blipdiffusion_controlnet(self):
         assert (
             np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
         ), f" expected_slice {expected_slice}, but got {image_slice.flatten()}"
+
+    @unittest.skip("Test not supported because of complexities in deriving query_embeds.")
+    def test_encode_prompt_works_in_isolation(self):
+        pass

tests/pipelines/controlnet/test_controlnet_img2img.py

@@ -189,6 +189,13 @@ def test_xformers_attention_forwardGenerator_pass(self):
     def test_inference_batch_single_identical(self):
         self._test_inference_batch_single_identical(expected_max_diff=2e-3)
 
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)
+
 
 class StableDiffusionMultiControlNetPipelineFastTests(
     IPAdapterTesterMixin, PipelineTesterMixin, PipelineKarrasSchedulerTesterMixin, unittest.TestCase
@@ -391,6 +398,13 @@ def test_save_pretrained_raise_not_implemented_exception(self):
             except NotImplementedError:
                 pass
 
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)
+
 
 @slow
 @require_torch_accelerator

tests/pipelines/controlnet/test_controlnet_inpaint.py

@@ -450,6 +450,13 @@ def test_save_pretrained_raise_not_implemented_exception(self):
             except NotImplementedError:
                 pass
 
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)
+
 
 @slow
 @require_torch_accelerator

tests/pipelines/pag/test_pag_animatediff.py

@@ -553,3 +553,11 @@ def test_pag_applied_layers(self):
         pag_layers = ["motion_modules.42"]
         with self.assertRaises(ValueError):
             pipe._set_pag_attn_processor(pag_applied_layers=pag_layers, do_classifier_free_guidance=False)
+
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "num_images_per_prompt": 1,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)

tests/pipelines/pag/test_pag_controlnet_sd.py

@@ -28,9 +28,7 @@
     StableDiffusionControlNetPipeline,
     UNet2DConditionModel,
 )
-from diffusers.utils.testing_utils import (
-    enable_full_determinism,
-)
+from diffusers.utils.testing_utils import enable_full_determinism, torch_device
 from diffusers.utils.torch_utils import randn_tensor
 
 from ..pipeline_params import (
@@ -246,3 +244,10 @@ def test_pag_uncond(self):
 
         max_diff = np.abs(image_slice.flatten() - expected_slice).max()
         assert max_diff < 1e-3, f"output is different from expected, {image_slice.flatten()}"
+
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)

tests/pipelines/pag/test_pag_controlnet_sd_inpaint.py

@@ -32,10 +32,7 @@
     StableDiffusionControlNetPAGInpaintPipeline,
     UNet2DConditionModel,
 )
-from diffusers.utils.testing_utils import (
-    enable_full_determinism,
-    floats_tensor,
-)
+from diffusers.utils.testing_utils import enable_full_determinism, floats_tensor, torch_device
 from diffusers.utils.torch_utils import randn_tensor
 
 from ..pipeline_params import (
@@ -243,3 +240,10 @@ def test_pag_uncond(self):
 
         max_diff = np.abs(image_slice.flatten() - expected_slice).max()
         assert max_diff < 1e-3, f"output is different from expected, {image_slice.flatten()}"
+
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)

tests/pipelines/pag/test_pag_hunyuan_dit.py

@@ -356,3 +356,10 @@ def test_pag_applied_layers(self):
         pag_layers = ["blocks.0", r"blocks\.1"]
         pipe._set_pag_attn_processor(pag_applied_layers=pag_layers, do_classifier_free_guidance=False)
         assert len(pipe.pag_attn_processors) == 2
+
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)

tests/pipelines/pag/test_pag_kolors.py

@@ -252,3 +252,6 @@ def test_pag_inference(self):
 
     def test_inference_batch_single_identical(self):
         self._test_inference_batch_single_identical(expected_max_diff=3e-3)
+
+    def test_encode_prompt_works_in_isolation(self):
+        return super().test_encode_prompt_works_in_isolation(atol=1e-3, rtol=1e-3)

tests/pipelines/pag/test_pag_sd.py

@@ -278,6 +278,13 @@ def test_pag_inference(self):
         max_diff = np.abs(image_slice.flatten() - expected_slice).max()
         self.assertLessEqual(max_diff, 1e-3)
 
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)
+
 
 @slow
 @require_torch_gpu

tests/pipelines/pag/test_pag_sd_img2img.py

@@ -210,6 +210,13 @@ def test_pag_inference(self):
         max_diff = np.abs(image_slice.flatten() - expected_slice).max()
         self.assertLessEqual(max_diff, 1e-3)
 
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)
+
 
 @slow
 @require_torch_gpu

tests/pipelines/pag/test_pag_sd_inpaint.py

@@ -244,6 +244,13 @@ def test_pag_inference(self):
         max_diff = np.abs(image_slice.flatten() - expected_slice).max()
         assert max_diff < 1e-3, f"output is different from expected, {image_slice.flatten()}"
 
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict, atol=1e-3, rtol=1e-3)
+
 
 @slow
 @require_torch_gpu

tests/pipelines/stable_diffusion_adapter/test_stable_diffusion_adapter.py

@@ -336,6 +336,13 @@ def test_adapter_lcm_custom_timesteps(self):
 
         assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
 
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)
+
 
 class StableDiffusionFullAdapterPipelineFastTests(
     AdapterTests, PipelineTesterMixin, PipelineFromPipeTesterMixin, unittest.TestCase

tests/pipelines/stable_diffusion_gligen/test_stable_diffusion_gligen.py

@@ -169,3 +169,7 @@ def test_attention_slicing_forward_pass(self):
 
     def test_inference_batch_single_identical(self):
         super().test_inference_batch_single_identical(batch_size=3, expected_max_diff=3e-3)
+
+    @unittest.skip("Test not supported as tokenizer is used for parsing bounding boxes.")
+    def test_encode_prompt_works_in_isolation(self):
+        pass

tests/pipelines/stable_diffusion_panorama/test_stable_diffusion_panorama.py

@@ -258,6 +258,13 @@ def test_stable_diffusion_panorama_pndm(self):
 
         assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
 
+    def test_encode_prompt_works_in_isolation(self):
+        extra_required_param_value_dict = {
+            "device": torch.device(torch_device).type,
+            "do_classifier_free_guidance": self.get_dummy_inputs(device=torch_device).get("guidance_scale", 1.0) > 1.0,
+        }
+        return super().test_encode_prompt_works_in_isolation(extra_required_param_value_dict)
+
 
 @nightly
 @require_torch_gpu

tests/pipelines/test_pipelines_common.py

@@ -2080,13 +2080,18 @@ def test_encode_prompt_works_in_isolation(self, extra_required_param_value_dict=
             and pipe_call_parameters.get("prompt_embeds").default is None
         ):
             pipe_without_tes_inputs.update({"prompt": None})
+
         pipe_out = pipe_without_text_encoders(**pipe_without_tes_inputs)[0]
 
         # Compare against regular pipeline outputs.
         full_pipe = self.pipeline_class(**components).to(torch_device)
         inputs = self.get_dummy_inputs(torch_device)
         pipe_out_2 = full_pipe(**inputs)[0]
-        self.assertTrue(np.allclose(pipe_out, pipe_out_2, atol=atol, rtol=rtol))
+
+        if isinstance(pipe_out, np.ndarray) and isinstance(pipe_out_2, np.ndarray):
+            self.assertTrue(np.allclose(pipe_out, pipe_out_2, atol=atol, rtol=rtol))
+        elif isinstance(pipe_out, torch.Tensor) and isinstance(pipe_out_2, torch.Tensor):
+            self.assertTrue(torch.allclose(pipe_out, pipe_out_2, atol=atol, rtol=rtol))
 
     def test_StableDiffusionMixin_component(self):
         """Any pipeline that have LDMFuncMixin should have vae and unet components."""