Merge branch 'main' into controlnet-test-fixes

* Add properties and `IPAdapterTesterMixin` tests for `StableDiffusionPanoramaPipeline`

* Fix variable name typo and update comments

* Update deprecated `output_type="numpy"` to "np" in test files

* Discard changes to src/diffusers/pipelines/stable_diffusion_panorama/pipeline_stable_diffusion_panorama.py

* Update test_stable_diffusion_panorama.py

* Update numbers in README.md

* Update get_guidance_scale_embedding method to use timesteps instead of w

* Update number of checkpoints in README.md

* Add type hints and fix var name

* Fix PyTorch's convention for inplace functions

* Fix a typo

* Revert "Fix PyTorch's convention for inplace functions"

This reverts commit 74350cf65b.

* Fix typos

* Indent

* Refactor get_guidance_scale_embedding method in LEditsPPPipelineStableDiffusionXL class

.github/workflows/pypi_publish.yaml

@@ -52,7 +52,7 @@ jobs:
       - name: Install dependencies
         run: |
           python -m pip install --upgrade pip
-          pip install -U setuptools wheel twine
+          pip install -U setuptools wheel twine torch
       
       - name: Build the dist files
         run: python setup.py bdist_wheel && python setup.py sdist

README.md

@@ -77,7 +77,7 @@ Please refer to the [How to use Stable Diffusion in Apple Silicon](https://huggi
 
 ## Quickstart
 
-Generating outputs is super easy with 🤗 Diffusers. To generate an image from text, use the `from_pretrained` method to load any pretrained diffusion model (browse the [Hub](https://huggingface.co/models?library=diffusers&sort=downloads) for 19000+ checkpoints):
+Generating outputs is super easy with 🤗 Diffusers. To generate an image from text, use the `from_pretrained` method to load any pretrained diffusion model (browse the [Hub](https://huggingface.co/models?library=diffusers&sort=downloads) for 22000+ checkpoints):
 
 ```python
 from diffusers import DiffusionPipeline
@@ -219,7 +219,7 @@ Also, say 👋 in our public Discord channel <a href="https://discord.gg/G7tWnz9
 - https://github.com/deep-floyd/IF
 - https://github.com/bentoml/BentoML
 - https://github.com/bmaltais/kohya_ss
-- +8000 other amazing GitHub repositories 💪
+- +9000 other amazing GitHub repositories 💪
 
 Thank you for using us ❤️.
 

examples/research_projects/onnxruntime/unconditional_image_generation/train_unconditional.py

@@ -637,7 +637,7 @@ def main(args):
                     generator=generator,
                     batch_size=args.eval_batch_size,
                     num_inference_steps=args.ddpm_num_inference_steps,
-                    output_type="numpy",
+                    output_type="np",
                 ).images
 
                 if args.use_ema:

examples/unconditional_image_generation/train_unconditional.py

@@ -648,7 +648,7 @@ def main(args):
                     generator=generator,
                     batch_size=args.eval_batch_size,
                     num_inference_steps=args.ddpm_num_inference_steps,
-                    output_type="numpy",
+                    output_type="np",
                 ).images
 
                 if args.use_ema:

src/diffusers/models/attention.py

@@ -293,7 +293,7 @@ class BasicTransformerBlock(nn.Module):
     ) -> torch.FloatTensor:
         if cross_attention_kwargs is not None:
             if cross_attention_kwargs.get("scale", None) is not None:
-                logger.warning("Passing `scale` to `cross_attention_kwargs` is depcrecated. `scale` will be ignored.")
+                logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
 
         # Notice that normalization is always applied before the real computation in the following blocks.
         # 0. Self-Attention

src/diffusers/models/transformers/transformer_2d.py

@@ -308,7 +308,7 @@ class Transformer2DModel(ModelMixin, ConfigMixin):
         """
         if cross_attention_kwargs is not None:
             if cross_attention_kwargs.get("scale", None) is not None:
-                logger.warning("Passing `scale` to `cross_attention_kwargs` is depcrecated. `scale` will be ignored.")
+                logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
         # ensure attention_mask is a bias, and give it a singleton query_tokens dimension.
         #   we may have done this conversion already, e.g. if we came here via UNet2DConditionModel#forward.
         #   we can tell by counting dims; if ndim == 2: it's a mask rather than a bias.

src/diffusers/models/unets/unet_2d_blocks.py

@@ -846,7 +846,7 @@ class UNetMidBlock2DCrossAttn(nn.Module):
     ) -> torch.FloatTensor:
         if cross_attention_kwargs is not None:
             if cross_attention_kwargs.get("scale", None) is not None:
-                logger.warning("Passing `scale` to `cross_attention_kwargs` is depcrecated. `scale` will be ignored.")
+                logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
 
         hidden_states = self.resnets[0](hidden_states, temb)
         for attn, resnet in zip(self.attentions, self.resnets[1:]):
@@ -986,7 +986,7 @@ class UNetMidBlock2DSimpleCrossAttn(nn.Module):
     ) -> torch.FloatTensor:
         cross_attention_kwargs = cross_attention_kwargs if cross_attention_kwargs is not None else {}
         if cross_attention_kwargs.get("scale", None) is not None:
-            logger.warning("Passing `scale` to `cross_attention_kwargs` is depcrecated. `scale` will be ignored.")
+            logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
 
         if attention_mask is None:
             # if encoder_hidden_states is defined: we are doing cross-attn, so we should use cross-attn mask.
@@ -1116,7 +1116,7 @@ class AttnDownBlock2D(nn.Module):
     ) -> Tuple[torch.FloatTensor, Tuple[torch.FloatTensor, ...]]:
         cross_attention_kwargs = cross_attention_kwargs if cross_attention_kwargs is not None else {}
         if cross_attention_kwargs.get("scale", None) is not None:
-            logger.warning("Passing `scale` to `cross_attention_kwargs` is depcrecated. `scale` will be ignored.")
+            logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
 
         output_states = ()
 
@@ -1241,7 +1241,7 @@ class CrossAttnDownBlock2D(nn.Module):
     ) -> Tuple[torch.FloatTensor, Tuple[torch.FloatTensor, ...]]:
         if cross_attention_kwargs is not None:
             if cross_attention_kwargs.get("scale", None) is not None:
-                logger.warning("Passing `scale` to `cross_attention_kwargs` is depcrecated. `scale` will be ignored.")
+                logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
 
         output_states = ()
 
@@ -1986,7 +1986,7 @@ class SimpleCrossAttnDownBlock2D(nn.Module):
     ) -> Tuple[torch.FloatTensor, Tuple[torch.FloatTensor, ...]]:
         cross_attention_kwargs = cross_attention_kwargs if cross_attention_kwargs is not None else {}
         if cross_attention_kwargs.get("scale", None) is not None:
-            logger.warning("Passing `scale` to `cross_attention_kwargs` is depcrecated. `scale` will be ignored.")
+            logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
 
         output_states = ()
 
@@ -2201,7 +2201,7 @@ class KCrossAttnDownBlock2D(nn.Module):
     ) -> Tuple[torch.FloatTensor, Tuple[torch.FloatTensor, ...]]:
         cross_attention_kwargs = cross_attention_kwargs if cross_attention_kwargs is not None else {}
         if cross_attention_kwargs.get("scale", None) is not None:
-            logger.warning("Passing `scale` to `cross_attention_kwargs` is depcrecated. `scale` will be ignored.")
+            logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
 
         output_states = ()
 
@@ -2483,7 +2483,7 @@ class CrossAttnUpBlock2D(nn.Module):
     ) -> torch.FloatTensor:
         if cross_attention_kwargs is not None:
             if cross_attention_kwargs.get("scale", None) is not None:
-                logger.warning("Passing `scale` to `cross_attention_kwargs` is depcrecated. `scale` will be ignored.")
+                logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
 
         is_freeu_enabled = (
             getattr(self, "s1", None)
@@ -3312,7 +3312,7 @@ class SimpleCrossAttnUpBlock2D(nn.Module):
     ) -> torch.FloatTensor:
         cross_attention_kwargs = cross_attention_kwargs if cross_attention_kwargs is not None else {}
         if cross_attention_kwargs.get("scale", None) is not None:
-            logger.warning("Passing `scale` to `cross_attention_kwargs` is depcrecated. `scale` will be ignored.")
+            logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
 
         if attention_mask is None:
             # if encoder_hidden_states is defined: we are doing cross-attn, so we should use cross-attn mask.
@@ -3694,7 +3694,7 @@ class KAttentionBlock(nn.Module):
     ) -> torch.FloatTensor:
         cross_attention_kwargs = cross_attention_kwargs if cross_attention_kwargs is not None else {}
         if cross_attention_kwargs.get("scale", None) is not None:
-            logger.warning("Passing `scale` to `cross_attention_kwargs` is depcrecated. `scale` will be ignored.")
+            logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
 
         # 1. Self-Attention
         if self.add_self_attention:

src/diffusers/models/unets/unet_3d_blocks.py

@@ -1183,7 +1183,7 @@ class CrossAttnDownBlockMotion(nn.Module):
     ):
         if cross_attention_kwargs is not None:
             if cross_attention_kwargs.get("scale", None) is not None:
-                logger.warning("Passing `scale` to `cross_attention_kwargs` is depcrecated. `scale` will be ignored.")
+                logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
 
         output_states = ()
 
@@ -1367,7 +1367,7 @@ class CrossAttnUpBlockMotion(nn.Module):
     ) -> torch.FloatTensor:
         if cross_attention_kwargs is not None:
             if cross_attention_kwargs.get("scale", None) is not None:
-                logger.warning("Passing `scale` to `cross_attention_kwargs` is depcrecated. `scale` will be ignored.")
+                logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
 
         is_freeu_enabled = (
             getattr(self, "s1", None)
@@ -1707,7 +1707,7 @@ class UNetMidBlockCrossAttnMotion(nn.Module):
     ) -> torch.FloatTensor:
         if cross_attention_kwargs is not None:
             if cross_attention_kwargs.get("scale", None) is not None:
-                logger.warning("Passing `scale` to `cross_attention_kwargs` is depcrecated. `scale` will be ignored.")
+                logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
 
         hidden_states = self.resnets[0](hidden_states, temb)
 

src/diffusers/pipelines/amused/pipeline_amused_img2img.py

@@ -127,7 +127,7 @@ class AmusedImg2ImgPipeline(DiffusionPipeline):
                 on the amount of noise initially added. When `strength` is 1, added noise is maximum and the denoising
                 process runs for the full number of iterations specified in `num_inference_steps`. A value of 1
                 essentially ignores `image`.
-            num_inference_steps (`int`, *optional*, defaults to 16):
+            num_inference_steps (`int`, *optional*, defaults to 12):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the
                 expense of slower inference.
             guidance_scale (`float`, *optional*, defaults to 10.0):
@@ -191,7 +191,7 @@ class AmusedImg2ImgPipeline(DiffusionPipeline):
             negative_prompt_embeds is None and negative_encoder_hidden_states is not None
         ):
             raise ValueError(
-                "pass either both `negatve_prompt_embeds` and `negative_encoder_hidden_states` or neither"
+                "pass either both `negative_prompt_embeds` and `negative_encoder_hidden_states` or neither"
             )
 
         if (prompt is None and prompt_embeds is None) or (prompt is not None and prompt_embeds is not None):

src/diffusers/pipelines/controlnet/pipeline_controlnet.py

@@ -824,20 +824,22 @@ class StableDiffusionControlNetPipeline(
         return latents
 
     # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
-    def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
+    def get_guidance_scale_embedding(
+        self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+    ) -> torch.FloatTensor:
         """
         See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
 
         Args:
-            timesteps (`torch.Tensor`):
+            w (`torch.Tensor`):
-                generate embedding vectors at these timesteps
+                Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
             embedding_dim (`int`, *optional*, defaults to 512):
-                dimension of the embeddings to generate
+                Dimension of the embeddings to generate.
-            dtype:
+            dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
-                data type of the generated embeddings
+                Data type of the generated embeddings.
 
         Returns:
-            `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+            `torch.FloatTensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
         """
         assert len(w.shape) == 1
         w = w * 1000.0

src/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py

@@ -869,20 +869,22 @@ class StableDiffusionXLControlNetPipeline(
             self.vae.decoder.mid_block.to(dtype)
 
     # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
-    def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
+    def get_guidance_scale_embedding(
+        self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+    ) -> torch.FloatTensor:
         """
         See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
 
         Args:
-            timesteps (`torch.Tensor`):
+            w (`torch.Tensor`):
-                generate embedding vectors at these timesteps
+                Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
             embedding_dim (`int`, *optional*, defaults to 512):
-                dimension of the embeddings to generate
+                Dimension of the embeddings to generate.
-            dtype:
+            dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
-                data type of the generated embeddings
+                Data type of the generated embeddings.
 
         Returns:
-            `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+            `torch.FloatTensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
         """
         assert len(w.shape) == 1
         w = w * 1000.0

src/diffusers/pipelines/deprecated/spectrogram_diffusion/pipeline_spectrogram_diffusion.py

@@ -133,7 +133,7 @@ class SpectrogramDiffusionPipeline(DiffusionPipeline):
         generator: Optional[torch.Generator] = None,
         num_inference_steps: int = 100,
         return_dict: bool = True,
-        output_type: str = "numpy",
+        output_type: str = "np",
         callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
         callback_steps: int = 1,
     ) -> Union[AudioPipelineOutput, Tuple]:
@@ -157,7 +157,7 @@ class SpectrogramDiffusionPipeline(DiffusionPipeline):
                 expense of slower inference.
             return_dict (`bool`, *optional*, defaults to `True`):
                 Whether or not to return a [`~pipelines.AudioPipelineOutput`] instead of a plain tuple.
-            output_type (`str`, *optional*, defaults to `"numpy"`):
+            output_type (`str`, *optional*, defaults to `"np"`):
                 The output format of the generated audio.
             callback (`Callable`, *optional*):
                 A function that calls every `callback_steps` steps during inference. The function is called with the
@@ -249,16 +249,16 @@ class SpectrogramDiffusionPipeline(DiffusionPipeline):
 
             logger.info("Generated segment", i)
 
-        if output_type == "numpy" and not is_onnx_available():
+        if output_type == "np" and not is_onnx_available():
             raise ValueError(
                 "Cannot return output in 'np' format if ONNX is not available. Make sure to have ONNX installed or set 'output_type' to 'mel'."
             )
-        elif output_type == "numpy" and self.melgan is None:
+        elif output_type == "np" and self.melgan is None:
             raise ValueError(
                 "Cannot return output in 'np' format if melgan component is not defined. Make sure to define `self.melgan` or set 'output_type' to 'mel'."
             )
 
-        if output_type == "numpy":
+        if output_type == "np":
             output = self.melgan(input_features=full_pred_mel.astype(np.float32))
         else:
             output = full_pred_mel

src/diffusers/pipelines/deprecated/versatile_diffusion/modeling_text_unet.py

@@ -2004,7 +2004,7 @@ class CrossAttnUpBlockFlat(nn.Module):
     ) -> torch.FloatTensor:
         if cross_attention_kwargs is not None:
             if cross_attention_kwargs.get("scale", None) is not None:
-                logger.warning("Passing `scale` to `cross_attention_kwargs` is depcrecated. `scale` will be ignored.")
+                logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
 
         is_freeu_enabled = (
             getattr(self, "s1", None)
@@ -2338,7 +2338,7 @@ class UNetMidBlockFlatCrossAttn(nn.Module):
     ) -> torch.FloatTensor:
         if cross_attention_kwargs is not None:
             if cross_attention_kwargs.get("scale", None) is not None:
-                logger.warning("Passing `scale` to `cross_attention_kwargs` is depcrecated. `scale` will be ignored.")
+                logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
 
         hidden_states = self.resnets[0](hidden_states, temb)
         for attn, resnet in zip(self.attentions, self.resnets[1:]):
@@ -2479,7 +2479,7 @@ class UNetMidBlockFlatSimpleCrossAttn(nn.Module):
     ) -> torch.FloatTensor:
         cross_attention_kwargs = cross_attention_kwargs if cross_attention_kwargs is not None else {}
         if cross_attention_kwargs.get("scale", None) is not None:
-            logger.warning("Passing `scale` to `cross_attention_kwargs` is depcrecated. `scale` will be ignored.")
+            logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
 
         if attention_mask is None:
             # if encoder_hidden_states is defined: we are doing cross-attn, so we should use cross-attn mask.

src/diffusers/pipelines/latent_consistency_models/pipeline_latent_consistency_img2img.py

@@ -548,20 +548,22 @@ class LatentConsistencyModelImg2ImgPipeline(
         return latents
 
     # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
-    def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
+    def get_guidance_scale_embedding(
+        self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+    ) -> torch.FloatTensor:
         """
         See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
 
         Args:
-            timesteps (`torch.Tensor`):
+            w (`torch.Tensor`):
-                generate embedding vectors at these timesteps
+                Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
             embedding_dim (`int`, *optional*, defaults to 512):
-                dimension of the embeddings to generate
+                Dimension of the embeddings to generate.
-            dtype:
+            dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
-                data type of the generated embeddings
+                Data type of the generated embeddings.
 
         Returns:
-            `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+            `torch.FloatTensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
         """
         assert len(w.shape) == 1
         w = w * 1000.0

src/diffusers/pipelines/latent_consistency_models/pipeline_latent_consistency_text2img.py

@@ -490,20 +490,22 @@ class LatentConsistencyModelPipeline(
         latents = latents * self.scheduler.init_noise_sigma
         return latents
 
-    def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
+    def get_guidance_scale_embedding(
+        self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+    ) -> torch.FloatTensor:
         """
         See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
 
         Args:
-            timesteps (`torch.Tensor`):
+            w (`torch.Tensor`):
-                generate embedding vectors at these timesteps
+                Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
             embedding_dim (`int`, *optional*, defaults to 512):
-                dimension of the embeddings to generate
+                Dimension of the embeddings to generate.
-            dtype:
+            dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
-                data type of the generated embeddings
+                Data type of the generated embeddings.
 
         Returns:
-            `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+            `torch.FloatTensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
         """
         assert len(w.shape) == 1
         w = w * 1000.0

src/diffusers/pipelines/ledits_pp/pipeline_leditspp_stable_diffusion_xl.py

@@ -713,20 +713,22 @@ class LEditsPPPipelineStableDiffusionXL(
             self.vae.decoder.mid_block.to(dtype)
 
     # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
-    def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
+    def get_guidance_scale_embedding(
+        self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+    ) -> torch.FloatTensor:
         """
         See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
 
         Args:
-            timesteps (`torch.Tensor`):
+            w (`torch.Tensor`):
-                generate embedding vectors at these timesteps
+                Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
             embedding_dim (`int`, *optional*, defaults to 512):
-                dimension of the embeddings to generate
+                Dimension of the embeddings to generate.
-            dtype:
+            dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
-                data type of the generated embeddings
+                Data type of the generated embeddings.
 
         Returns:
-            `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+            `torch.FloatTensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
         """
         assert len(w.shape) == 1
         w = w * 1000.0

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py

@@ -669,20 +669,22 @@ class StableDiffusionPipeline(
         return latents
 
     # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
-    def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
+    def get_guidance_scale_embedding(
+        self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+    ) -> torch.FloatTensor:
         """
         See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
 
         Args:
-            timesteps (`torch.Tensor`):
+            w (`torch.Tensor`):
-                generate embedding vectors at these timesteps
+                Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
             embedding_dim (`int`, *optional*, defaults to 512):
-                dimension of the embeddings to generate
+                Dimension of the embeddings to generate.
-            dtype:
+            dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
-                data type of the generated embeddings
+                Data type of the generated embeddings.
 
         Returns:
-            `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+            `torch.FloatTensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
         """
         assert len(w.shape) == 1
         w = w * 1000.0

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_img2img.py

@@ -767,20 +767,22 @@ class StableDiffusionImg2ImgPipeline(
         return latents
 
     # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
-    def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
+    def get_guidance_scale_embedding(
+        self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+    ) -> torch.FloatTensor:
         """
         See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
 
         Args:
-            timesteps (`torch.Tensor`):
+            w (`torch.Tensor`):
-                generate embedding vectors at these timesteps
+                Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
             embedding_dim (`int`, *optional*, defaults to 512):
-                dimension of the embeddings to generate
+                Dimension of the embeddings to generate.
-            dtype:
+            dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
-                data type of the generated embeddings
+                Data type of the generated embeddings.
 
         Returns:
-            `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+            `torch.FloatTensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
         """
         assert len(w.shape) == 1
         w = w * 1000.0

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint.py

@@ -909,20 +909,22 @@ class StableDiffusionInpaintPipeline(
         return timesteps, num_inference_steps - t_start
 
     # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
-    def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
+    def get_guidance_scale_embedding(
+        self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+    ) -> torch.FloatTensor:
         """
         See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
 
         Args:
-            timesteps (`torch.Tensor`):
+            w (`torch.Tensor`):
-                generate embedding vectors at these timesteps
+                Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
             embedding_dim (`int`, *optional*, defaults to 512):
-                dimension of the embeddings to generate
+                Dimension of the embeddings to generate.
-            dtype:
+            dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
-                data type of the generated embeddings
+                Data type of the generated embeddings.
 
         Returns:
-            `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+            `torch.FloatTensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
         """
         assert len(w.shape) == 1
         w = w * 1000.0

src/diffusers/pipelines/stable_diffusion_diffedit/pipeline_stable_diffusion_diffedit.py

@@ -1304,7 +1304,7 @@ class StableDiffusionDiffEditPipeline(
         callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
         callback_steps: int = 1,
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
-        clip_ckip: int = None,
+        clip_skip: int = None,
     ):
         r"""
         The call function to the pipeline for generation.
@@ -1426,7 +1426,7 @@ class StableDiffusionDiffEditPipeline(
             prompt_embeds=prompt_embeds,
             negative_prompt_embeds=negative_prompt_embeds,
             lora_scale=text_encoder_lora_scale,
-            clip_skip=clip_ckip,
+            clip_skip=clip_skip,
         )
         # For classifier free guidance, we need to do two forward passes.
         # Here we concatenate the unconditional and text embeddings into a single batch

src/diffusers/pipelines/stable_diffusion_ldm3d/pipeline_stable_diffusion_ldm3d.py

@@ -644,20 +644,22 @@ class StableDiffusionLDM3DPipeline(
         return latents
 
     # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
-    def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
+    def get_guidance_scale_embedding(
+        self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+    ) -> torch.FloatTensor:
         """
         See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
 
         Args:
-            timesteps (`torch.Tensor`):
+            w (`torch.Tensor`):
-                generate embedding vectors at these timesteps
+                Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
             embedding_dim (`int`, *optional*, defaults to 512):
-                dimension of the embeddings to generate
+                Dimension of the embeddings to generate.
-            dtype:
+            dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
-                data type of the generated embeddings
+                Data type of the generated embeddings.
 
         Returns:
-            `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+            `torch.FloatTensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
         """
         assert len(w.shape) == 1
         w = w * 1000.0

src/diffusers/pipelines/stable_diffusion_sag/pipeline_stable_diffusion_sag.py

@@ -632,7 +632,7 @@ class StableDiffusionSAGPipeline(DiffusionPipeline, StableDiffusionMixin, Textua
         # corresponds to doing no classifier free guidance.
         do_classifier_free_guidance = guidance_scale > 1.0
         # and `sag_scale` is` `s` of equation (16)
-        # of the self-attentnion guidance paper: https://arxiv.org/pdf/2210.00939.pdf
+        # of the self-attention guidance paper: https://arxiv.org/pdf/2210.00939.pdf
         # `sag_scale = 0` means no self-attention guidance
         do_self_attention_guidance = sag_scale > 0.0
 
@@ -667,7 +667,7 @@ class StableDiffusionSAGPipeline(DiffusionPipeline, StableDiffusionMixin, Textua
 
         if timesteps.dtype not in [torch.int16, torch.int32, torch.int64]:
             raise ValueError(
-                f"{self.__class__.__name__} does not support using a scheduler of type {self.scheduler.__class__.__name__}. Please make sure to use one of 'DDIMScheduler, PNDMScheduler, DDPMScheduler, DEISMultistepScheduler, UniPCMultistepScheduler, DPMSolverMultistepScheduler, DPMSolverSinlgestepScheduler'."
+                f"{self.__class__.__name__} does not support using a scheduler of type {self.scheduler.__class__.__name__}. Please make sure to use one of 'DDIMScheduler, PNDMScheduler, DDPMScheduler, DEISMultistepScheduler, UniPCMultistepScheduler, DPMSolverMultistepScheduler, DPMSolverSinglestepScheduler'."
             )
 
         # 5. Prepare latent variables
@@ -723,7 +723,7 @@ class StableDiffusionSAGPipeline(DiffusionPipeline, StableDiffusionMixin, Textua
                         noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
                         noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
 
-                    # perform self-attention guidance with the stored self-attentnion map
+                    # perform self-attention guidance with the stored self-attention map
                     if do_self_attention_guidance:
                         # classifier-free guidance produces two chunks of attention map
                         # and we only use unconditional one according to equation (25)

src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py

@@ -740,20 +740,22 @@ class StableDiffusionXLPipeline(
             self.vae.decoder.mid_block.to(dtype)
 
     # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
-    def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
+    def get_guidance_scale_embedding(
+        self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+    ) -> torch.FloatTensor:
         """
         See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
 
         Args:
-            timesteps (`torch.Tensor`):
+            w (`torch.Tensor`):
-                generate embedding vectors at these timesteps
+                Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
             embedding_dim (`int`, *optional*, defaults to 512):
-                dimension of the embeddings to generate
+                Dimension of the embeddings to generate.
-            dtype:
+            dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
-                data type of the generated embeddings
+                Data type of the generated embeddings.
 
         Returns:
-            `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+            `torch.FloatTensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
         """
         assert len(w.shape) == 1
         w = w * 1000.0

src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py

@@ -874,20 +874,22 @@ class StableDiffusionXLImg2ImgPipeline(
             self.vae.decoder.mid_block.to(dtype)
 
     # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
-    def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
+    def get_guidance_scale_embedding(
+        self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+    ) -> torch.FloatTensor:
         """
         See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
 
         Args:
-            timesteps (`torch.Tensor`):
+            w (`torch.Tensor`):
-                generate embedding vectors at these timesteps
+                Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
             embedding_dim (`int`, *optional*, defaults to 512):
-                dimension of the embeddings to generate
+                Dimension of the embeddings to generate.
-            dtype:
+            dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
-                data type of the generated embeddings
+                Data type of the generated embeddings.
 
         Returns:
-            `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+            `torch.FloatTensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
         """
         assert len(w.shape) == 1
         w = w * 1000.0

src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py

@@ -1110,20 +1110,22 @@ class StableDiffusionXLInpaintPipeline(
             self.vae.decoder.mid_block.to(dtype)
 
     # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
-    def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
+    def get_guidance_scale_embedding(
+        self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+    ) -> torch.FloatTensor:
         """
         See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
 
         Args:
-            timesteps (`torch.Tensor`):
+            w (`torch.Tensor`):
-                generate embedding vectors at these timesteps
+                Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
             embedding_dim (`int`, *optional*, defaults to 512):
-                dimension of the embeddings to generate
+                Dimension of the embeddings to generate.
-            dtype:
+            dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
-                data type of the generated embeddings
+                Data type of the generated embeddings.
 
         Returns:
-            `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+            `torch.FloatTensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
         """
         assert len(w.shape) == 1
         w = w * 1000.0

src/diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_adapter.py

@@ -613,20 +613,22 @@ class StableDiffusionAdapterPipeline(DiffusionPipeline, StableDiffusionMixin):
         return height, width
 
     # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
-    def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
+    def get_guidance_scale_embedding(
+        self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+    ) -> torch.FloatTensor:
         """
         See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
 
         Args:
-            timesteps (`torch.Tensor`):
+            w (`torch.Tensor`):
-                generate embedding vectors at these timesteps
+                Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
             embedding_dim (`int`, *optional*, defaults to 512):
-                dimension of the embeddings to generate
+                Dimension of the embeddings to generate.
-            dtype:
+            dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
-                data type of the generated embeddings
+                Data type of the generated embeddings.
 
         Returns:
-            `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+            `torch.FloatTensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
         """
         assert len(w.shape) == 1
         w = w * 1000.0

src/diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_xl_adapter.py

@@ -784,20 +784,22 @@ class StableDiffusionXLAdapterPipeline(
         return height, width
 
     # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
-    def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
+    def get_guidance_scale_embedding(
+        self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+    ) -> torch.FloatTensor:
         """
         See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
 
         Args:
-            timesteps (`torch.Tensor`):
+            w (`torch.Tensor`):
-                generate embedding vectors at these timesteps
+                Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
             embedding_dim (`int`, *optional*, defaults to 512):
-                dimension of the embeddings to generate
+                Dimension of the embeddings to generate.
-            dtype:
+            dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
-                data type of the generated embeddings
+                Data type of the generated embeddings.
 
         Returns:
-            `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+            `torch.FloatTensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
         """
         assert len(w.shape) == 1
         w = w * 1000.0

src/diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_zero.py

@@ -575,8 +575,8 @@ class TextToVideoZeroPipeline(DiffusionPipeline, StableDiffusionMixin, TextualIn
                 Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for video
                 generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
                 tensor is generated by sampling using the supplied random `generator`.
-            output_type (`str`, *optional*, defaults to `"numpy"`):
+            output_type (`str`, *optional*, defaults to `"np"`):
-                The output format of the generated video. Choose between `"latent"` and `"numpy"`.
+                The output format of the generated video. Choose between `"latent"` and `"np"`.
             return_dict (`bool`, *optional*, defaults to `True`):
                 Whether or not to return a
                 [`~pipelines.text_to_video_synthesis.pipeline_text_to_video_zero.TextToVideoPipelineOutput`] instead of

tests/pipelines/controlnet/test_controlnet.py

@@ -211,7 +211,7 @@ class ControlNetPipelineFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
             "image": image,
         }
 
@@ -402,7 +402,7 @@ class StableDiffusionMultiControlNetPipelineFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
             "image": images,
         }
 
@@ -602,7 +602,7 @@ class StableDiffusionMultiControlNetOneModelPipelineFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
             "image": images,
         }
 
@@ -1092,6 +1092,13 @@ class ControlNetPipelineSlowTests(unittest.TestCase):
         for param_name, param_value in single_file_pipe.controlnet.config.items():
             if param_name in PARAMS_TO_IGNORE:
                 continue
+
+            # This parameter doesn't appear to be loaded from the config.
+            # So when it is registered to config, it remains a tuple as this is the default in the class definition
+            # from_pretrained, does load from config and converts to a list when registering to config
+            if param_name == "conditioning_embedding_out_channels" and isinstance(param_value, tuple):
+                param_value = list(param_value)
+
             assert (
                 pipe.controlnet.config[param_name] == param_value
             ), f"{param_name} differs between single file loading and pretrained loading"

tests/pipelines/controlnet/test_controlnet_img2img.py

@@ -164,7 +164,7 @@ class ControlNetImg2ImgPipelineFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
             "image": image,
             "control_image": control_image,
         }
@@ -313,7 +313,7 @@ class StableDiffusionMultiControlNetPipelineFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
             "image": image,
             "control_image": control_image,
         }

tests/pipelines/controlnet/test_controlnet_inpaint.py

@@ -155,7 +155,7 @@ class ControlNetInpaintPipelineFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
             "image": image,
             "mask_image": mask_image,
             "control_image": control_image,
@@ -375,7 +375,7 @@ class MultiControlNetInpaintPipelineFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
             "image": image,
             "mask_image": mask_image,
             "control_image": control_image,

tests/pipelines/controlnet/test_controlnet_inpaint_sdxl.py

@@ -172,7 +172,7 @@ class ControlNetPipelineSDXLFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
             "image": init_image,
             "mask_image": mask_image,
             "control_image": control_image,

tests/pipelines/controlnet/test_controlnet_sdxl.py

@@ -1002,6 +1002,11 @@ class ControlNetSDXLPipelineSlowTests(unittest.TestCase):
         for param_name, param_value in single_file_pipe.unet.config.items():
             if param_name in PARAMS_TO_IGNORE:
                 continue
+
+            # Upcast attention might be set to None in a config file, which is incorrect. It should default to False in the model
+            if param_name == "upcast_attention" and pipe.unet.config[param_name] is None:
+                pipe.unet.config[param_name] = False
+
             assert (
                 pipe.unet.config[param_name] == param_value
             ), f"{param_name} differs between single file loading and pretrained loading"

tests/pipelines/controlnet/test_controlnet_sdxl_img2img.py

@@ -163,7 +163,7 @@ class ControlNetPipelineSDXLImg2ImgFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
             "image": image,
             "control_image": image,
         }

tests/pipelines/ddim/test_ddim.py

@@ -63,7 +63,7 @@ class DDIMPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
             "batch_size": 1,
             "generator": generator,
             "num_inference_steps": 2,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -113,7 +113,7 @@ class DDIMPipelineIntegrationTests(unittest.TestCase):
         ddim.set_progress_bar_config(disable=None)
 
         generator = torch.manual_seed(0)
-        image = ddim(generator=generator, eta=0.0, output_type="numpy").images
+        image = ddim(generator=generator, eta=0.0, output_type="np").images
 
         image_slice = image[0, -3:, -3:, -1]
 
@@ -133,7 +133,7 @@ class DDIMPipelineIntegrationTests(unittest.TestCase):
         ddpm.set_progress_bar_config(disable=None)
 
         generator = torch.manual_seed(0)
-        image = ddpm(generator=generator, output_type="numpy").images
+        image = ddpm(generator=generator, output_type="np").images
 
         image_slice = image[0, -3:, -3:, -1]
 

tests/pipelines/ddpm/test_ddpm.py

@@ -50,10 +50,10 @@ class DDPMPipelineFastTests(unittest.TestCase):
         ddpm.set_progress_bar_config(disable=None)
 
         generator = torch.Generator(device=device).manual_seed(0)
-        image = ddpm(generator=generator, num_inference_steps=2, output_type="numpy").images
+        image = ddpm(generator=generator, num_inference_steps=2, output_type="np").images
 
         generator = torch.Generator(device=device).manual_seed(0)
-        image_from_tuple = ddpm(generator=generator, num_inference_steps=2, output_type="numpy", return_dict=False)[0]
+        image_from_tuple = ddpm(generator=generator, num_inference_steps=2, output_type="np", return_dict=False)[0]
 
         image_slice = image[0, -3:, -3:, -1]
         image_from_tuple_slice = image_from_tuple[0, -3:, -3:, -1]
@@ -75,10 +75,10 @@ class DDPMPipelineFastTests(unittest.TestCase):
         ddpm.set_progress_bar_config(disable=None)
 
         generator = torch.manual_seed(0)
-        image = ddpm(generator=generator, num_inference_steps=2, output_type="numpy").images
+        image = ddpm(generator=generator, num_inference_steps=2, output_type="np").images
 
         generator = torch.manual_seed(0)
-        image_eps = ddpm(generator=generator, num_inference_steps=2, output_type="numpy")[0]
+        image_eps = ddpm(generator=generator, num_inference_steps=2, output_type="np")[0]
 
         image_slice = image[0, -3:, -3:, -1]
         image_eps_slice = image_eps[0, -3:, -3:, -1]
@@ -102,7 +102,7 @@ class DDPMPipelineIntegrationTests(unittest.TestCase):
         ddpm.set_progress_bar_config(disable=None)
 
         generator = torch.manual_seed(0)
-        image = ddpm(generator=generator, output_type="numpy").images
+        image = ddpm(generator=generator, output_type="np").images
 
         image_slice = image[0, -3:, -3:, -1]
 

tests/pipelines/deepfloyd_if/test_if.py

@@ -50,7 +50,7 @@ class IFPipelineFastTests(PipelineTesterMixin, IFPipelineTesterMixin, unittest.T
             "prompt": "A painting of a squirrel eating a burger",
             "generator": generator,
             "num_inference_steps": 2,
-            "output_type": "numpy",
+            "output_type": "np",
         }
 
         return inputs

tests/pipelines/deepfloyd_if/test_if_img2img.py

@@ -55,7 +55,7 @@ class IFImg2ImgPipelineFastTests(PipelineTesterMixin, IFPipelineTesterMixin, uni
             "image": image,
             "generator": generator,
             "num_inference_steps": 2,
-            "output_type": "numpy",
+            "output_type": "np",
         }
 
         return inputs

tests/pipelines/deepfloyd_if/test_if_img2img_superresolution.py

@@ -57,7 +57,7 @@ class IFImg2ImgSuperResolutionPipelineFastTests(PipelineTesterMixin, IFPipelineT
             "original_image": original_image,
             "generator": generator,
             "num_inference_steps": 2,
-            "output_type": "numpy",
+            "output_type": "np",
         }
 
         return inputs

tests/pipelines/deepfloyd_if/test_if_inpainting.py

@@ -57,7 +57,7 @@ class IFInpaintingPipelineFastTests(PipelineTesterMixin, IFPipelineTesterMixin,
             "mask_image": mask_image,
             "generator": generator,
             "num_inference_steps": 2,
-            "output_type": "numpy",
+            "output_type": "np",
         }
 
         return inputs

tests/pipelines/deepfloyd_if/test_if_inpainting_superresolution.py

@@ -59,7 +59,7 @@ class IFInpaintingSuperResolutionPipelineFastTests(PipelineTesterMixin, IFPipeli
             "mask_image": mask_image,
             "generator": generator,
             "num_inference_steps": 2,
-            "output_type": "numpy",
+            "output_type": "np",
         }
 
         return inputs

tests/pipelines/deepfloyd_if/test_if_superresolution.py

@@ -52,7 +52,7 @@ class IFSuperResolutionPipelineFastTests(PipelineTesterMixin, IFPipelineTesterMi
             "image": image,
             "generator": generator,
             "num_inference_steps": 2,
-            "output_type": "numpy",
+            "output_type": "np",
         }
 
         return inputs

tests/pipelines/dit/test_dit.py

@@ -74,7 +74,7 @@ class DiTPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
             "class_labels": [1],
             "generator": generator,
             "num_inference_steps": 2,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/latent_diffusion/test_latent_diffusion.py

@@ -113,7 +113,7 @@ class LDMTextToImagePipelineFastTests(PipelineTesterMixin, unittest.TestCase):
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -153,7 +153,7 @@ class LDMTextToImagePipelineSlowTests(unittest.TestCase):
             "generator": generator,
             "num_inference_steps": 3,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -189,7 +189,7 @@ class LDMTextToImagePipelineNightlyTests(unittest.TestCase):
             "generator": generator,
             "num_inference_steps": 50,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/latent_diffusion/test_latent_diffusion_superresolution.py

@@ -84,7 +84,7 @@ class LDMSuperResolutionPipelineFastTests(unittest.TestCase):
         init_image = self.dummy_image.to(device)
 
         generator = torch.Generator(device=device).manual_seed(0)
-        image = ldm(image=init_image, generator=generator, num_inference_steps=2, output_type="numpy").images
+        image = ldm(image=init_image, generator=generator, num_inference_steps=2, output_type="np").images
 
         image_slice = image[0, -3:, -3:, -1]
 
@@ -109,7 +109,7 @@ class LDMSuperResolutionPipelineFastTests(unittest.TestCase):
 
         init_image = self.dummy_image.to(torch_device)
 
-        image = ldm(init_image, num_inference_steps=2, output_type="numpy").images
+        image = ldm(init_image, num_inference_steps=2, output_type="np").images
 
         assert image.shape == (1, 64, 64, 3)
 
@@ -128,7 +128,7 @@ class LDMSuperResolutionPipelineIntegrationTests(unittest.TestCase):
         ldm.set_progress_bar_config(disable=None)
 
         generator = torch.manual_seed(0)
-        image = ldm(image=init_image, generator=generator, num_inference_steps=20, output_type="numpy").images
+        image = ldm(image=init_image, generator=generator, num_inference_steps=20, output_type="np").images
 
         image_slice = image[0, -3:, -3:, -1]
 

tests/pipelines/paint_by_example/test_paint_by_example.py

@@ -117,7 +117,7 @@ class PaintByExamplePipelineFastTests(PipelineTesterMixin, unittest.TestCase):
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/pndm/test_pndm.py

@@ -49,10 +49,10 @@ class PNDMPipelineFastTests(unittest.TestCase):
         pndm.set_progress_bar_config(disable=None)
 
         generator = torch.manual_seed(0)
-        image = pndm(generator=generator, num_inference_steps=20, output_type="numpy").images
+        image = pndm(generator=generator, num_inference_steps=20, output_type="np").images
 
         generator = torch.manual_seed(0)
-        image_from_tuple = pndm(generator=generator, num_inference_steps=20, output_type="numpy", return_dict=False)[0]
+        image_from_tuple = pndm(generator=generator, num_inference_steps=20, output_type="np", return_dict=False)[0]
 
         image_slice = image[0, -3:, -3:, -1]
         image_from_tuple_slice = image_from_tuple[0, -3:, -3:, -1]
@@ -77,7 +77,7 @@ class PNDMPipelineIntegrationTests(unittest.TestCase):
         pndm.to(torch_device)
         pndm.set_progress_bar_config(disable=None)
         generator = torch.manual_seed(0)
-        image = pndm(generator=generator, output_type="numpy").images
+        image = pndm(generator=generator, output_type="np").images
 
         image_slice = image[0, -3:, -3:, -1]
 

tests/pipelines/stable_diffusion/test_onnx_stable_diffusion.py

@@ -46,7 +46,7 @@ class OnnxStableDiffusionPipelineFastTests(OnnxPipelineTesterMixin, unittest.Tes
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 7.5,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/stable_diffusion/test_onnx_stable_diffusion_img2img.py

@@ -55,7 +55,7 @@ class OnnxStableDiffusionImg2ImgPipelineFastTests(OnnxPipelineTesterMixin, unitt
             "num_inference_steps": 3,
             "strength": 0.75,
             "guidance_scale": 7.5,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/stable_diffusion/test_onnx_stable_diffusion_upscale.py

@@ -55,7 +55,7 @@ class OnnxStableDiffusionUpscalePipelineFastTests(OnnxPipelineTesterMixin, unitt
             "generator": generator,
             "num_inference_steps": 3,
             "guidance_scale": 7.5,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/stable_diffusion/test_stable_diffusion.py

@@ -775,7 +775,7 @@ class StableDiffusionPipelineSlowTests(unittest.TestCase):
             "generator": generator,
             "num_inference_steps": 3,
             "guidance_scale": 7.5,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -950,7 +950,7 @@ class StableDiffusionPipelineSlowTests(unittest.TestCase):
             generator=generator,
             guidance_scale=7.5,
             num_inference_steps=2,
-            output_type="numpy",
+            output_type="np",
         )
         image_chunked = output_chunked.images
 
@@ -966,7 +966,7 @@ class StableDiffusionPipelineSlowTests(unittest.TestCase):
             generator=generator,
             guidance_scale=7.5,
             num_inference_steps=2,
-            output_type="numpy",
+            output_type="np",
         )
         image = output.images
 

tests/pipelines/stable_diffusion/test_stable_diffusion_img2img.py

@@ -179,7 +179,7 @@ class StableDiffusionImg2ImgPipelineFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/stable_diffusion/test_stable_diffusion_inpaint.py

@@ -199,7 +199,7 @@ class StableDiffusionInpaintPipelineFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -470,7 +470,7 @@ class StableDiffusionSimpleInpaintPipelineFastTests(StableDiffusionInpaintPipeli
             "generator": [generator1, generator2],
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -586,7 +586,7 @@ class StableDiffusionInpaintPipelineSlowTests(unittest.TestCase):
             "generator": generator,
             "num_inference_steps": 3,
             "guidance_scale": 7.5,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -847,7 +847,7 @@ class StableDiffusionInpaintPipelineAsymmetricAutoencoderKLSlowTests(unittest.Te
             "generator": generator,
             "num_inference_steps": 3,
             "guidance_scale": 7.5,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -1072,7 +1072,7 @@ class StableDiffusionInpaintPipelineNightlyTests(unittest.TestCase):
             "generator": generator,
             "num_inference_steps": 50,
             "guidance_scale": 7.5,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/stable_diffusion/test_stable_diffusion_instruction_pix2pix.py

@@ -131,7 +131,7 @@ class StableDiffusionInstructPix2PixPipelineFastTests(
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
             "image_guidance_scale": 1,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -288,7 +288,7 @@ class StableDiffusionInstructPix2PixPipelineSlowTests(unittest.TestCase):
             "num_inference_steps": 3,
             "guidance_scale": 7.5,
             "image_guidance_scale": 1.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/stable_diffusion_2/test_stable_diffusion.py

@@ -151,7 +151,7 @@ class StableDiffusion2PipelineFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -336,7 +336,7 @@ class StableDiffusion2PipelineSlowTests(unittest.TestCase):
             "generator": generator,
             "num_inference_steps": 3,
             "guidance_scale": 7.5,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -557,7 +557,7 @@ class StableDiffusion2PipelineNightlyTests(unittest.TestCase):
             "generator": generator,
             "num_inference_steps": 50,
             "guidance_scale": 7.5,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/stable_diffusion_2/test_stable_diffusion_attend_and_excite.py

@@ -138,7 +138,7 @@ class StableDiffusionAttendAndExcitePipelineFastTests(
             "generator": generator,
             "num_inference_steps": 1,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
             "max_iter_to_alter": 2,
             "thresholds": {0: 0.7},
         }
@@ -225,7 +225,7 @@ class StableDiffusionAttendAndExcitePipelineIntegrationTests(unittest.TestCase):
             generator=generator,
             num_inference_steps=5,
             max_iter_to_alter=5,
-            output_type="numpy",
+            output_type="np",
         ).images[0]
 
         expected_image = load_numpy(

tests/pipelines/stable_diffusion_2/test_stable_diffusion_depth.py

@@ -174,7 +174,7 @@ class StableDiffusionDepth2ImgPipelineFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -395,7 +395,7 @@ class StableDiffusionDepth2ImgPipelineSlowTests(unittest.TestCase):
             "num_inference_steps": 3,
             "strength": 0.75,
             "guidance_scale": 7.5,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -534,7 +534,7 @@ class StableDiffusionImg2ImgPipelineNightlyTests(unittest.TestCase):
             "num_inference_steps": 3,
             "strength": 0.75,
             "guidance_scale": 7.5,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/stable_diffusion_2/test_stable_diffusion_diffedit.py

@@ -143,7 +143,7 @@ class StableDiffusionDiffEditPipelineFastTests(PipelineLatentTesterMixin, Pipeli
             "num_inference_steps": 2,
             "inpaint_strength": 1.0,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
 
         return inputs
@@ -165,7 +165,7 @@ class StableDiffusionDiffEditPipelineFastTests(PipelineLatentTesterMixin, Pipeli
             "num_maps_per_mask": 2,
             "mask_encode_strength": 1.0,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
 
         return inputs
@@ -186,7 +186,7 @@ class StableDiffusionDiffEditPipelineFastTests(PipelineLatentTesterMixin, Pipeli
             "inpaint_strength": 1.0,
             "guidance_scale": 6.0,
             "decode_latents": True,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -417,7 +417,7 @@ class StableDiffusionDiffEditPipelineNightlyTests(unittest.TestCase):
             negative_prompt=source_prompt,
             inpaint_strength=0.7,
             num_inference_steps=25,
-            output_type="numpy",
+            output_type="np",
         ).images[0]
 
         expected_image = (

tests/pipelines/stable_diffusion_2/test_stable_diffusion_inpaint.py

@@ -129,7 +129,7 @@ class StableDiffusion2InpaintPipelineFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/stable_diffusion_2/test_stable_diffusion_latent_upscale.py

@@ -155,7 +155,7 @@ class StableDiffusionLatentUpscalePipelineFastTests(
             "image": self.dummy_image.cpu(),
             "generator": generator,
             "num_inference_steps": 2,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/stable_diffusion_2/test_stable_diffusion_v_pred.py

@@ -308,7 +308,7 @@ class StableDiffusion2VPredictionPipelineIntegrationTests(unittest.TestCase):
         prompt = "A painting of a squirrel eating a burger"
         generator = torch.manual_seed(0)
 
-        output = sd_pipe([prompt], generator=generator, num_inference_steps=5, output_type="numpy")
+        output = sd_pipe([prompt], generator=generator, num_inference_steps=5, output_type="np")
         image = output.images
 
         image_slice = image[0, 253:256, 253:256, -1]
@@ -335,7 +335,7 @@ class StableDiffusion2VPredictionPipelineIntegrationTests(unittest.TestCase):
         prompt = "a photograph of an astronaut riding a horse"
         generator = torch.manual_seed(0)
         image = sd_pipe(
-            [prompt], generator=generator, guidance_scale=7.5, num_inference_steps=5, output_type="numpy"
+            [prompt], generator=generator, guidance_scale=7.5, num_inference_steps=5, output_type="np"
         ).images
 
         image_slice = image[0, 253:256, 253:256, -1]
@@ -357,7 +357,7 @@ class StableDiffusion2VPredictionPipelineIntegrationTests(unittest.TestCase):
         pipe.enable_attention_slicing()
         generator = torch.manual_seed(0)
         output_chunked = pipe(
-            [prompt], generator=generator, guidance_scale=7.5, num_inference_steps=10, output_type="numpy"
+            [prompt], generator=generator, guidance_scale=7.5, num_inference_steps=10, output_type="np"
         )
         image_chunked = output_chunked.images
 
@@ -369,7 +369,7 @@ class StableDiffusion2VPredictionPipelineIntegrationTests(unittest.TestCase):
         # disable slicing
         pipe.disable_attention_slicing()
         generator = torch.manual_seed(0)
-        output = pipe([prompt], generator=generator, guidance_scale=7.5, num_inference_steps=10, output_type="numpy")
+        output = pipe([prompt], generator=generator, guidance_scale=7.5, num_inference_steps=10, output_type="np")
         image = output.images
 
         # make sure that more than 3.0 GB is allocated

tests/pipelines/stable_diffusion_adapter/test_stable_diffusion_adapter.py

@@ -246,7 +246,7 @@ class AdapterTests:
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/stable_diffusion_image_variation/test_stable_diffusion_image_variation.py

@@ -117,7 +117,7 @@ class StableDiffusionImageVariationPipelineFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -293,7 +293,7 @@ class StableDiffusionImageVariationPipelineNightlyTests(unittest.TestCase):
             "generator": generator,
             "num_inference_steps": 50,
             "guidance_scale": 7.5,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/stable_diffusion_ldm3d/test_stable_diffusion_ldm3d.py

@@ -107,7 +107,7 @@ class StableDiffusionLDM3DPipelineFastTests(unittest.TestCase):
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -222,7 +222,7 @@ class StableDiffusionLDM3DPipelineSlowTests(unittest.TestCase):
             "generator": generator,
             "num_inference_steps": 3,
             "guidance_scale": 7.5,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -268,7 +268,7 @@ class StableDiffusionPipelineNightlyTests(unittest.TestCase):
             "generator": generator,
             "num_inference_steps": 50,
             "guidance_scale": 7.5,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/stable_diffusion_panorama/test_stable_diffusion_panorama.py

@@ -105,7 +105,7 @@ class StableDiffusionPanoramaPipelineFastTests(PipelineLatentTesterMixin, Pipeli
             "width": None,
             "num_inference_steps": 1,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -263,7 +263,7 @@ class StableDiffusionPanoramaNightlyTests(unittest.TestCase):
             "generator": generator,
             "num_inference_steps": 3,
             "guidance_scale": 7.5,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/stable_diffusion_xl/test_stable_diffusion_xl_adapter.py

@@ -290,7 +290,7 @@ class StableDiffusionXLAdapterPipelineFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 5.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/stable_diffusion_xl/test_stable_diffusion_xl_instruction_pix2pix.py

@@ -143,7 +143,7 @@ class StableDiffusionXLInstructPix2PixPipelineFastTests(
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
             "image_guidance_scale": 1,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/stable_unclip/test_stable_unclip.py

@@ -168,7 +168,7 @@ class StableUnCLIPPipelineFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "prior_num_inference_steps": 2,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/test_pipelines.py

@@ -117,10 +117,10 @@ def _test_from_save_pretrained_dynamo(in_queue, out_queue, timeout):
             new_ddpm.to(torch_device)
 
         generator = torch.Generator(device=torch_device).manual_seed(0)
-        image = ddpm(generator=generator, num_inference_steps=5, output_type="numpy").images
+        image = ddpm(generator=generator, num_inference_steps=5, output_type="np").images
 
         generator = torch.Generator(device=torch_device).manual_seed(0)
-        new_image = new_ddpm(generator=generator, num_inference_steps=5, output_type="numpy").images
+        new_image = new_ddpm(generator=generator, num_inference_steps=5, output_type="np").images
 
         assert np.abs(image - new_image).max() < 1e-5, "Models don't give the same forward pass"
     except Exception:
@@ -363,12 +363,12 @@ class DownloadTests(unittest.TestCase):
         )
         pipe = pipe.to(torch_device)
         generator = torch.manual_seed(0)
-        out = pipe(prompt, num_inference_steps=2, generator=generator, output_type="numpy").images
+        out = pipe(prompt, num_inference_steps=2, generator=generator, output_type="np").images
 
         pipe_2 = StableDiffusionPipeline.from_pretrained("hf-internal-testing/tiny-stable-diffusion-torch")
         pipe_2 = pipe_2.to(torch_device)
         generator = torch.manual_seed(0)
-        out_2 = pipe_2(prompt, num_inference_steps=2, generator=generator, output_type="numpy").images
+        out_2 = pipe_2(prompt, num_inference_steps=2, generator=generator, output_type="np").images
 
         assert np.max(np.abs(out - out_2)) < 1e-3
 
@@ -379,7 +379,7 @@ class DownloadTests(unittest.TestCase):
         )
         pipe = pipe.to(torch_device)
         generator = torch.manual_seed(0)
-        out = pipe(prompt, num_inference_steps=2, generator=generator, output_type="numpy").images
+        out = pipe(prompt, num_inference_steps=2, generator=generator, output_type="np").images
 
         with tempfile.TemporaryDirectory() as tmpdirname:
             pipe.save_pretrained(tmpdirname)
@@ -388,7 +388,7 @@ class DownloadTests(unittest.TestCase):
 
             generator = torch.manual_seed(0)
 
-            out_2 = pipe_2(prompt, num_inference_steps=2, generator=generator, output_type="numpy").images
+            out_2 = pipe_2(prompt, num_inference_steps=2, generator=generator, output_type="np").images
 
         assert np.max(np.abs(out - out_2)) < 1e-3
 
@@ -398,7 +398,7 @@ class DownloadTests(unittest.TestCase):
         pipe = pipe.to(torch_device)
 
         generator = torch.manual_seed(0)
-        out = pipe(prompt, num_inference_steps=2, generator=generator, output_type="numpy").images
+        out = pipe(prompt, num_inference_steps=2, generator=generator, output_type="np").images
 
         with tempfile.TemporaryDirectory() as tmpdirname:
             pipe.save_pretrained(tmpdirname)
@@ -407,7 +407,7 @@ class DownloadTests(unittest.TestCase):
 
             generator = torch.manual_seed(0)
 
-            out_2 = pipe_2(prompt, num_inference_steps=2, generator=generator, output_type="numpy").images
+            out_2 = pipe_2(prompt, num_inference_steps=2, generator=generator, output_type="np").images
 
         assert np.max(np.abs(out - out_2)) < 1e-3
 
@@ -590,7 +590,7 @@ class DownloadTests(unittest.TestCase):
                 )
                 pipe = pipe.to(torch_device)
                 generator = torch.manual_seed(0)
-                out = pipe(prompt, num_inference_steps=2, generator=generator, output_type="numpy").images
+                out = pipe(prompt, num_inference_steps=2, generator=generator, output_type="np").images
 
                 with tempfile.TemporaryDirectory() as tmpdirname:
                     pipe.save_pretrained(tmpdirname)
@@ -601,7 +601,7 @@ class DownloadTests(unittest.TestCase):
 
                 generator = torch.manual_seed(0)
 
-                out_2 = pipe_2(prompt, num_inference_steps=2, generator=generator, output_type="numpy").images
+                out_2 = pipe_2(prompt, num_inference_steps=2, generator=generator, output_type="np").images
 
                 assert np.max(np.abs(out - out_2)) < 1e-3
 
@@ -626,7 +626,7 @@ class DownloadTests(unittest.TestCase):
             assert pipe._maybe_convert_prompt("<*>", pipe.tokenizer) == "<*>"
 
             prompt = "hey <*>"
-            out = pipe(prompt, num_inference_steps=1, output_type="numpy").images
+            out = pipe(prompt, num_inference_steps=1, output_type="np").images
             assert out.shape == (1, 128, 128, 3)
 
         # single token load local with weight name
@@ -642,7 +642,7 @@ class DownloadTests(unittest.TestCase):
             assert pipe._maybe_convert_prompt("<**>", pipe.tokenizer) == "<**>"
 
             prompt = "hey <**>"
-            out = pipe(prompt, num_inference_steps=1, output_type="numpy").images
+            out = pipe(prompt, num_inference_steps=1, output_type="np").images
             assert out.shape == (1, 128, 128, 3)
 
         # multi token load
@@ -665,7 +665,7 @@ class DownloadTests(unittest.TestCase):
             assert pipe._maybe_convert_prompt("<***>", pipe.tokenizer) == "<***> <***>_1 <***>_2"
 
             prompt = "hey <***>"
-            out = pipe(prompt, num_inference_steps=1, output_type="numpy").images
+            out = pipe(prompt, num_inference_steps=1, output_type="np").images
             assert out.shape == (1, 128, 128, 3)
 
         # multi token load a1111
@@ -693,7 +693,7 @@ class DownloadTests(unittest.TestCase):
             assert pipe._maybe_convert_prompt("<****>", pipe.tokenizer) == "<****> <****>_1 <****>_2"
 
             prompt = "hey <****>"
-            out = pipe(prompt, num_inference_steps=1, output_type="numpy").images
+            out = pipe(prompt, num_inference_steps=1, output_type="np").images
             assert out.shape == (1, 128, 128, 3)
 
         # multi embedding load
@@ -718,7 +718,7 @@ class DownloadTests(unittest.TestCase):
                 assert pipe._maybe_convert_prompt("<******>", pipe.tokenizer) == "<******>"
 
                 prompt = "hey <*****> <******>"
-                out = pipe(prompt, num_inference_steps=1, output_type="numpy").images
+                out = pipe(prompt, num_inference_steps=1, output_type="np").images
                 assert out.shape == (1, 128, 128, 3)
 
         # single token state dict load
@@ -731,7 +731,7 @@ class DownloadTests(unittest.TestCase):
         assert pipe._maybe_convert_prompt("<x>", pipe.tokenizer) == "<x>"
 
         prompt = "hey <x>"
-        out = pipe(prompt, num_inference_steps=1, output_type="numpy").images
+        out = pipe(prompt, num_inference_steps=1, output_type="np").images
         assert out.shape == (1, 128, 128, 3)
 
         # multi embedding state dict load
@@ -751,7 +751,7 @@ class DownloadTests(unittest.TestCase):
         assert pipe._maybe_convert_prompt("<xxxxxx>", pipe.tokenizer) == "<xxxxxx>"
 
         prompt = "hey <xxxxx> <xxxxxx>"
-        out = pipe(prompt, num_inference_steps=1, output_type="numpy").images
+        out = pipe(prompt, num_inference_steps=1, output_type="np").images
         assert out.shape == (1, 128, 128, 3)
 
         # auto1111 multi-token state dict load
@@ -777,7 +777,7 @@ class DownloadTests(unittest.TestCase):
         assert pipe._maybe_convert_prompt("<xxxx>", pipe.tokenizer) == "<xxxx> <xxxx>_1 <xxxx>_2"
 
         prompt = "hey <xxxx>"
-        out = pipe(prompt, num_inference_steps=1, output_type="numpy").images
+        out = pipe(prompt, num_inference_steps=1, output_type="np").images
         assert out.shape == (1, 128, 128, 3)
 
         # multiple references to multi embedding
@@ -789,7 +789,7 @@ class DownloadTests(unittest.TestCase):
         )
 
         prompt = "hey <cat> <cat>"
-        out = pipe(prompt, num_inference_steps=1, output_type="numpy").images
+        out = pipe(prompt, num_inference_steps=1, output_type="np").images
         assert out.shape == (1, 128, 128, 3)
 
     def test_text_inversion_multi_tokens(self):
@@ -1739,10 +1739,10 @@ class PipelineSlowTests(unittest.TestCase):
             new_ddpm.to(torch_device)
 
         generator = torch.Generator(device=torch_device).manual_seed(0)
-        image = ddpm(generator=generator, num_inference_steps=5, output_type="numpy").images
+        image = ddpm(generator=generator, num_inference_steps=5, output_type="np").images
 
         generator = torch.Generator(device=torch_device).manual_seed(0)
-        new_image = new_ddpm(generator=generator, num_inference_steps=5, output_type="numpy").images
+        new_image = new_ddpm(generator=generator, num_inference_steps=5, output_type="np").images
 
         assert np.abs(image - new_image).max() < 1e-5, "Models don't give the same forward pass"
 
@@ -1765,10 +1765,10 @@ class PipelineSlowTests(unittest.TestCase):
         ddpm_from_hub.set_progress_bar_config(disable=None)
 
         generator = torch.Generator(device=torch_device).manual_seed(0)
-        image = ddpm(generator=generator, num_inference_steps=5, output_type="numpy").images
+        image = ddpm(generator=generator, num_inference_steps=5, output_type="np").images
 
         generator = torch.Generator(device=torch_device).manual_seed(0)
-        new_image = ddpm_from_hub(generator=generator, num_inference_steps=5, output_type="numpy").images
+        new_image = ddpm_from_hub(generator=generator, num_inference_steps=5, output_type="np").images
 
         assert np.abs(image - new_image).max() < 1e-5, "Models don't give the same forward pass"
 
@@ -1788,10 +1788,10 @@ class PipelineSlowTests(unittest.TestCase):
         ddpm_from_hub_custom_model.set_progress_bar_config(disable=None)
 
         generator = torch.Generator(device=torch_device).manual_seed(0)
-        image = ddpm_from_hub_custom_model(generator=generator, num_inference_steps=5, output_type="numpy").images
+        image = ddpm_from_hub_custom_model(generator=generator, num_inference_steps=5, output_type="np").images
 
         generator = torch.Generator(device=torch_device).manual_seed(0)
-        new_image = ddpm_from_hub(generator=generator, num_inference_steps=5, output_type="numpy").images
+        new_image = ddpm_from_hub(generator=generator, num_inference_steps=5, output_type="np").images
 
         assert np.abs(image - new_image).max() < 1e-5, "Models don't give the same forward pass"
 
@@ -1803,7 +1803,7 @@ class PipelineSlowTests(unittest.TestCase):
         pipe.to(torch_device)
         pipe.set_progress_bar_config(disable=None)
 
-        images = pipe(output_type="numpy").images
+        images = pipe(output_type="np").images
         assert images.shape == (1, 32, 32, 3)
         assert isinstance(images, np.ndarray)
 
@@ -1878,7 +1878,7 @@ class PipelineSlowTests(unittest.TestCase):
         generator = [torch.Generator(device="cpu").manual_seed(33) for _ in range(prompt_embeds.shape[0])]
 
         images = pipe(
-            prompt_embeds=prompt_embeds, generator=generator, num_inference_steps=20, output_type="numpy"
+            prompt_embeds=prompt_embeds, generator=generator, num_inference_steps=20, output_type="np"
         ).images
 
         for i, image in enumerate(images):
@@ -1916,7 +1916,7 @@ class PipelineNightlyTests(unittest.TestCase):
         ddim.set_progress_bar_config(disable=None)
 
         generator = torch.Generator(device=torch_device).manual_seed(seed)
-        ddpm_images = ddpm(batch_size=2, generator=generator, output_type="numpy").images
+        ddpm_images = ddpm(batch_size=2, generator=generator, output_type="np").images
 
         generator = torch.Generator(device=torch_device).manual_seed(seed)
         ddim_images = ddim(
@@ -1924,7 +1924,7 @@ class PipelineNightlyTests(unittest.TestCase):
             generator=generator,
             num_inference_steps=1000,
             eta=1.0,
-            output_type="numpy",
+            output_type="np",
             use_clipped_model_output=True,  # Need this to make DDIM match DDPM
         ).images
 

tests/pipelines/unclip/test_unclip.py

@@ -233,7 +233,7 @@ class UnCLIPPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
             "prior_num_inference_steps": 2,
             "decoder_num_inference_steps": 2,
             "super_res_num_inference_steps": 2,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 

tests/pipelines/unidiffuser/test_unidiffuser.py

@@ -158,7 +158,7 @@ class UniDiffuserPipelineFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         return inputs
 
@@ -199,7 +199,7 @@ class UniDiffuserPipelineFastTests(
             "generator": generator,
             "num_inference_steps": 2,
             "guidance_scale": 6.0,
-            "output_type": "numpy",
+            "output_type": "np",
             "prompt_latents": latents.get("prompt_latents"),
             "vae_latents": latents.get("vae_latents"),
             "clip_latents": latents.get("clip_latents"),
@@ -590,7 +590,7 @@ class UniDiffuserPipelineSlowTests(unittest.TestCase):
             "generator": generator,
             "num_inference_steps": 3,
             "guidance_scale": 8.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         if generate_latents:
             latents = self.get_fixed_latents(device, seed=seed)
@@ -706,7 +706,7 @@ class UniDiffuserPipelineNightlyTests(unittest.TestCase):
             "generator": generator,
             "num_inference_steps": 3,
             "guidance_scale": 8.0,
-            "output_type": "numpy",
+            "output_type": "np",
         }
         if generate_latents:
             latents = self.get_fixed_latents(device, seed=seed)