Fix StableDiffusionInstructPix2PixPipelineSingleFileSlowTests (huggingface#10557)

Author: hlky

src/diffusers/loaders/single_file_utils.py

@@ -186,6 +186,7 @@
     "inpainting": 512,
     "inpainting_v2": 512,
     "controlnet": 512,
+    "instruct-pix2pix": 512,
     "v2": 768,
     "v1": 512,
 }

tests/single_file/single_file_testing_utils.py

@@ -47,6 +47,8 @@ def download_diffusers_config(repo_id, tmpdir):
 
 
 class SDSingleFileTesterMixin:
+    single_file_kwargs = {}
+
     def _compare_component_configs(self, pipe, single_file_pipe):
         for param_name, param_value in single_file_pipe.text_encoder.config.to_dict().items():
             if param_name in ["torch_dtype", "architectures", "_name_or_path"]:
@@ -154,7 +156,7 @@ def test_single_file_components_with_original_config_local_files_only(
         self._compare_component_configs(pipe, single_file_pipe)
 
     def test_single_file_format_inference_is_same_as_pretrained(self, expected_max_diff=1e-4):
-        sf_pipe = self.pipeline_class.from_single_file(self.ckpt_path, safety_checker=None)
+        sf_pipe = self.pipeline_class.from_single_file(self.ckpt_path, safety_checker=None, **self.single_file_kwargs)
         sf_pipe.unet.set_attn_processor(AttnProcessor())
         sf_pipe.enable_model_cpu_offload(device=torch_device)
 
@@ -170,7 +172,7 @@ def test_single_file_format_inference_is_same_as_pretrained(self, expected_max_d
 
         max_diff = numpy_cosine_similarity_distance(image.flatten(), image_single_file.flatten())
 
-        assert max_diff < expected_max_diff
+        assert max_diff < expected_max_diff, f"{image.flatten()} != {image_single_file.flatten()}"
 
     def test_single_file_components_with_diffusers_config(
         self,

tests/single_file/test_stable_diffusion_single_file.py

@@ -132,6 +132,7 @@ class StableDiffusionInstructPix2PixPipelineSingleFileSlowTests(unittest.TestCas
         "https://raw.githubusercontent.com/timothybrooks/instruct-pix2pix/refs/heads/main/configs/generate.yaml"
     )
     repo_id = "timbrooks/instruct-pix2pix"
+    single_file_kwargs = {"extract_ema": True}
 
     def setUp(self):
         super().setUp()