Document output_type="pt" in the agent docs and drop per-file comments

.ai/references/testing.md

@@ -24,7 +24,7 @@ Follow the style introduced in [#14113](https://github.com/huggingface/diffusers
   - Set `pipeline_class`, `required_input_params_in_call_signature` (params that must appear in `__call__`'s signature), and `batch_input_params` (params that get batched). Use the canonical sets in `..pipeline_params` where one fits, or an inline `frozenset([...])`.
   - Set `output_shape` — the per-sample output shape for `get_dummy_inputs()`, i.e. `(channels, height, width)` for an image pipeline and `(num_frames, channels, height, width)` for a video one. Assert against `self.output_shape` in pipeline-specific tests instead of repeating the literal.
   - Implement `get_dummy_components(...)` — build every sub-module from the **real classes** at tiny config, each preceded by `torch.manual_seed(0)`.
-  - Implement `get_dummy_inputs()` — **no `device` / `seed` arguments** (unlike the old style). Use `self.get_generator(0)` for the generator, keep sizes tiny, and set `output_type="pt"` so tests compare torch tensors directly with `assert_tensors_close` (no numpy round-trip). Remember `"pt"` images are `(batch, channels, height, width)`.
+  - Implement `get_dummy_inputs()` — **no `device` / `seed` arguments** (unlike the old style). Use `self.get_generator(0)` for the generator, keep sizes tiny, and set `output_type="pt"` so tests compare torch tensors directly with `assert_tensors_close` (no numpy round-trip). Remember `"pt"` images are `(batch, channels, height, width)` and videos `(batch, num_frames, channels, height, width)`, unlike `"np"` outputs which are `(batch, height, width, channels)`.
 - **Compose the config with one mixin per concern**, one test class each, named `Test<Pipeline>...`. Add only the mixins that apply:
   - `PipelineTesterMixin` — core save/load, dict-vs-tuple equivalence, batching, dtype/device, callbacks. Put pipeline-specific tests as methods on this class.
   - `MemoryTesterMixin` — CPU offload, group offload, layerwise casting.

tests/pipelines/ace_step/test_ace_step.py

@@ -237,7 +237,6 @@ def get_dummy_inputs(self):
             "num_inference_steps": 2,
             "generator": self.get_generator(0),
             "max_text_length": 32,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/allegro/test_allegro.py

@@ -129,7 +129,6 @@ def get_dummy_inputs(self):
             "width": 16,
             "num_frames": 8,
             "max_sequence_length": 16,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/animatediff/test_animatediff.py

@@ -117,7 +117,6 @@ def get_dummy_inputs(self):
             "generator": self.get_generator(0),
             "num_inference_steps": 2,
             "guidance_scale": 7.5,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/animatediff/test_animatediff_controlnet.py

@@ -126,7 +126,6 @@ def get_dummy_inputs(self, num_frames: int = 2):
             "num_inference_steps": 2,
             "num_frames": num_frames,
             "guidance_scale": 7.5,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/animatediff/test_animatediff_sdxl.py

@@ -112,7 +112,6 @@ def get_dummy_inputs(self):
             "generator": self.get_generator(0),
             "num_inference_steps": 2,
             "guidance_scale": 7.5,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/animatediff/test_animatediff_sparsectrl.py

@@ -128,7 +128,6 @@ def get_dummy_inputs(self, num_frames: int = 2):
             "num_inference_steps": 2,
             "num_frames": num_frames,
             "guidance_scale": 7.5,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/animatediff/test_animatediff_video2video.py

@@ -115,7 +115,6 @@ def get_dummy_inputs(self, num_frames: int = 2):
             "generator": self.get_generator(0),
             "num_inference_steps": 2,
             "guidance_scale": 7.5,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/animatediff/test_animatediff_video2video_controlnet.py

@@ -129,7 +129,6 @@ def get_dummy_inputs(self, num_frames: int = 2):
             "generator": self.get_generator(0),
             "num_inference_steps": 2,
             "guidance_scale": 7.5,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/anyflow/test_anyflow.py

@@ -100,7 +100,6 @@ def get_dummy_inputs(self):
             "width": 16,
             "num_frames": 9,
             "max_sequence_length": 16,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/anyflow/test_anyflow_far.py

@@ -111,7 +111,6 @@ def get_dummy_inputs(self):
             "width": 16,
             "num_frames": 9,
             "max_sequence_length": 16,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/audioldm2/test_audioldm2.py

@@ -201,7 +201,6 @@ def get_dummy_inputs(self):
             "generator": self.get_generator(0),
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/aura_flow/test_pipeline_aura_flow.py

@@ -72,8 +72,6 @@ def get_dummy_inputs(self):
             "guidance_scale": 5.0,
             "height": None,
             "width": None,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
-            # Note `"pt"` images are `(batch, channels, height, width)`, unlike `"np"` (`(batch, h, w, c)`).
             "output_type": "pt",
         }
 

tests/pipelines/bria/test_pipeline_bria.py

@@ -108,7 +108,6 @@ def get_dummy_inputs(self):
             "height": 16,
             "width": 16,
             "max_sequence_length": 48,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/bria_fibo/test_pipeline_bria_fibo.py

@@ -101,7 +101,6 @@ def get_dummy_inputs(self):
             "guidance_scale": 5.0,
             "height": 32,
             "width": 32,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/bria_fibo_edit/test_pipeline_bria_fibo_edit.py

@@ -105,7 +105,6 @@ def get_dummy_inputs(self):
             "guidance_scale": 5.0,
             "height": 192,
             "width": 336,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
         inputs["image"] = Image.new("RGB", (336, 192), (255, 255, 255))

tests/pipelines/cogvideo/test_cogvideox.py

@@ -124,7 +124,6 @@ def get_dummy_inputs(self):
             "width": 16,
             "num_frames": 8,
             "max_sequence_length": 16,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/cogvideo/test_cogvideox_fun_control.py

@@ -116,7 +116,6 @@ def get_dummy_inputs(self):
             "height": height,
             "width": width,
             "max_sequence_length": 16,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/cogvideo/test_cogvideox_image2video.py

@@ -141,7 +141,6 @@ def get_dummy_inputs(self):
             "width": image_width,
             "num_frames": 8,
             "max_sequence_length": 16,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
         return inputs

tests/pipelines/cogview4/test_cogview4.py

@@ -99,7 +99,6 @@ def get_dummy_inputs(self):
             "height": 16,
             "width": 16,
             "max_sequence_length": 16,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/ddim/test_ddim.py

@@ -61,7 +61,6 @@ def get_dummy_inputs(self):
             "batch_size": 1,
             "generator": self.get_generator(0),
             "num_inference_steps": 2,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/ddpm/test_ddpm.py

@@ -61,7 +61,6 @@ def get_dummy_inputs(self):
             "batch_size": 1,
             "generator": self.get_generator(0),
             "num_inference_steps": 2,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/deepfloyd_if/test_if.py

@@ -117,7 +117,6 @@ def get_dummy_inputs(self):
             "prompt": "A painting of a squirrel eating a burger",
             "generator": self.get_generator(0),
             "num_inference_steps": 2,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/deepfloyd_if/test_if_img2img.py

@@ -120,7 +120,6 @@ def get_dummy_inputs(self):
             "image": image,
             "generator": self.get_generator(0),
             "num_inference_steps": 2,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/deepfloyd_if/test_if_img2img_superresolution.py

@@ -135,7 +135,6 @@ def get_dummy_inputs(self):
             "original_image": original_image,
             "generator": self.get_generator(0),
             "num_inference_steps": 2,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/deepfloyd_if/test_if_inpainting.py

@@ -130,7 +130,6 @@ def get_dummy_inputs(self):
             "mask_image": mask_image,
             "generator": self.get_generator(0),
             "num_inference_steps": 2,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/deepfloyd_if/test_if_inpainting_superresolution.py

@@ -145,7 +145,6 @@ def get_dummy_inputs(self):
             "mask_image": mask_image,
             "generator": self.get_generator(0),
             "num_inference_steps": 2,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/deepfloyd_if/test_if_superresolution.py

@@ -133,7 +133,6 @@ def get_dummy_inputs(self):
             "image": image,
             "generator": self.get_generator(0),
             "num_inference_steps": 2,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/dit/test_dit.py

@@ -75,7 +75,6 @@ def get_dummy_inputs(self):
             "class_labels": [1],
             "generator": self.get_generator(0),
             "num_inference_steps": 2,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/dreamlite/test_pipeline_dreamlite.py

@@ -73,7 +73,6 @@ def get_dummy_inputs(self):
             "height": 64,
             "width": 64,
             "max_sequence_length": 16,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/dreamlite/test_pipeline_dreamlite_mobile.py

@@ -63,7 +63,6 @@ def get_dummy_inputs(self):
             "height": 64,
             "width": 64,
             "max_sequence_length": 16,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/easyanimate/test_easyanimate.py

@@ -110,7 +110,6 @@ def get_dummy_inputs(self):
             "height": 16,
             "width": 16,
             "num_frames": 5,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/flux/test_pipeline_flux.py

@@ -129,8 +129,6 @@ def get_dummy_inputs(self):
             "height": 8,
             "width": 8,
             "max_sequence_length": 48,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
-            # Note `"pt"` images are `(batch, channels, height, width)`, unlike `"np"` (`(batch, h, w, c)`).
             "output_type": "pt",
         }
         return inputs

tests/pipelines/flux/test_pipeline_flux_control.py

@@ -120,8 +120,6 @@ def get_dummy_inputs(self):
             "height": 8,
             "width": 8,
             "max_sequence_length": 48,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
-            # Note `"pt"` images are `(batch, channels, height, width)`, unlike `"np"` (`(batch, h, w, c)`).
             "output_type": "pt",
         }
         return inputs

tests/pipelines/flux/test_pipeline_flux_control_img2img.py

@@ -101,8 +101,6 @@ def get_dummy_inputs(self):
             "width": 8,
             "max_sequence_length": 48,
             "strength": 0.8,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
-            # Note `"pt"` images are `(batch, channels, height, width)`, unlike `"np"` (`(batch, h, w, c)`).
             "output_type": "pt",
         }
         return inputs

tests/pipelines/flux/test_pipeline_flux_control_inpaint.py

@@ -108,8 +108,6 @@ def get_dummy_inputs(self):
             "height": 8,
             "width": 8,
             "max_sequence_length": 48,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
-            # Note `"pt"` images are `(batch, channels, height, width)`, unlike `"np"` (`(batch, h, w, c)`).
             "output_type": "pt",
         }
         return inputs

tests/pipelines/flux/test_pipeline_flux_fill.py

@@ -96,8 +96,6 @@ def get_dummy_inputs(self):
             "height": 32,
             "width": 32,
             "max_sequence_length": 48,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
-            # Note `"pt"` images are `(batch, channels, height, width)`, unlike `"np"` (`(batch, h, w, c)`).
             "output_type": "pt",
         }
         return inputs

tests/pipelines/flux/test_pipeline_flux_img2img.py

@@ -97,8 +97,6 @@ def get_dummy_inputs(self):
             "width": 8,
             "max_sequence_length": 48,
             "strength": 0.8,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
-            # Note `"pt"` images are `(batch, channels, height, width)`, unlike `"np"` (`(batch, h, w, c)`).
             "output_type": "pt",
         }
         return inputs

tests/pipelines/flux/test_pipeline_flux_inpaint.py

@@ -99,8 +99,6 @@ def get_dummy_inputs(self):
             "width": 32,
             "max_sequence_length": 48,
             "strength": 0.8,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
-            # Note `"pt"` images are `(batch, channels, height, width)`, unlike `"np"` (`(batch, h, w, c)`).
             "output_type": "pt",
         }
         return inputs

tests/pipelines/flux/test_pipeline_flux_kontext.py

@@ -106,8 +106,6 @@ def get_dummy_inputs(self):
             "width": 8,
             "max_area": 8 * 8,
             "max_sequence_length": 48,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
-            # Note `"pt"` images are `(batch, channels, height, width)`, unlike `"np"` (`(batch, h, w, c)`).
             "output_type": "pt",
             "_auto_resize": False,
         }

tests/pipelines/flux/test_pipeline_flux_kontext_inpaint.py

@@ -110,8 +110,6 @@ def get_dummy_inputs(self):
             "width": 32,
             "max_sequence_length": 48,
             "strength": 0.8,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
-            # Note `"pt"` images are `(batch, channels, height, width)`, unlike `"np"` (`(batch, h, w, c)`).
             "output_type": "pt",
             "_auto_resize": False,
         }

tests/pipelines/flux/testing_utils.py

@@ -51,7 +51,6 @@ def _modify_inputs_for_ip_adapter_test(self, inputs: dict[str, Any]):
         inputs["negative_prompt"] = ""
         if "true_cfg_scale" in inspect.signature(self.pipeline_class.__call__).parameters:
             inputs["true_cfg_scale"] = 4.0
-        # Request torch outputs so comparisons run on torch tensors directly (see `BasePipelineTesterConfig`).
         inputs["output_type"] = "pt"
         inputs["return_dict"] = False
         return inputs

tests/pipelines/flux2/test_pipeline_flux2.py

@@ -116,8 +116,6 @@ def get_dummy_inputs(self):
             "height": 8,
             "width": 8,
             "max_sequence_length": 8,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
-            # Note `"pt"` images are `(batch, channels, height, width)`, unlike `"np"` (`(batch, h, w, c)`).
             "output_type": "pt",
             "text_encoder_out_layers": (1,),
         }

tests/pipelines/flux2/test_pipeline_flux2_klein.py

@@ -104,8 +104,6 @@ def get_dummy_inputs(self):
             "height": 8,
             "width": 8,
             "max_sequence_length": 64,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
-            # Note `"pt"` images are `(batch, channels, height, width)`, unlike `"np"` (`(batch, h, w, c)`).
             "output_type": "pt",
             "text_encoder_out_layers": (1,),
         }

tests/pipelines/flux2/test_pipeline_flux2_klein_inpaint.py

@@ -97,7 +97,6 @@ def get_dummy_inputs(self):
             "max_sequence_length": 64,
             "strength": 0.8,
             "text_encoder_out_layers": (1,),
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
         return inputs

tests/pipelines/flux2/test_pipeline_flux2_klein_kv.py

@@ -93,7 +93,6 @@ def get_dummy_inputs(self):
             "width": 8,
             "max_sequence_length": 64,
             "text_encoder_out_layers": (1,),
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
         return inputs

tests/pipelines/glm_image/test_glm_image.py

@@ -129,7 +129,6 @@ def get_dummy_inputs(self):
             "height": 32,
             "width": 32,
             "max_sequence_length": 16,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }
 

tests/pipelines/helios/test_helios.py

@@ -107,7 +107,6 @@ def get_dummy_inputs(self):
             "width": 16,
             "num_frames": 9,
             "max_sequence_length": 16,
-            # Request torch outputs so tests compare torch tensors directly (see `BasePipelineTesterConfig`).
             "output_type": "pt",
         }