Merge branch 'huggingface:main' into cogview4

Author: zRzRzRzRzRzRzR

examples/instruct_pix2pix/train_instruct_pix2pix.py

@@ -695,7 +695,7 @@ def preprocess_images(examples):
         )
         # We need to ensure that the original and the edited images undergo the same
         # augmentation transforms.
-        images = np.concatenate([original_images, edited_images])
+        images = np.stack([original_images, edited_images])
         images = torch.tensor(images)
         images = 2 * (images / 255) - 1
         return train_transforms(images)
@@ -706,7 +706,7 @@ def preprocess_train(examples):
         # Since the original and edited images were concatenated before
         # applying the transformations, we need to separate them and reshape
         # them accordingly.
-        original_images, edited_images = preprocessed_images.chunk(2)
+        original_images, edited_images = preprocessed_images
         original_images = original_images.reshape(-1, 3, args.resolution, args.resolution)
         edited_images = edited_images.reshape(-1, 3, args.resolution, args.resolution)
 

examples/instruct_pix2pix/train_instruct_pix2pix_sdxl.py

@@ -766,7 +766,7 @@ def preprocess_images(examples):
         )
         # We need to ensure that the original and the edited images undergo the same
         # augmentation transforms.
-        images = np.concatenate([original_images, edited_images])
+        images = np.stack([original_images, edited_images])
         images = torch.tensor(images)
         images = 2 * (images / 255) - 1
         return train_transforms(images)
@@ -906,7 +906,7 @@ def preprocess_train(examples):
         # Since the original and edited images were concatenated before
         # applying the transformations, we need to separate them and reshape
         # them accordingly.
-        original_images, edited_images = preprocessed_images.chunk(2)
+        original_images, edited_images = preprocessed_images
         original_images = original_images.reshape(-1, 3, args.resolution, args.resolution)
         edited_images = edited_images.reshape(-1, 3, args.resolution, args.resolution)
 

src/diffusers/models/attention_processor.py

@@ -405,11 +405,12 @@ def set_use_memory_efficient_attention_xformers(
             else:
                 try:
                     # Make sure we can run the memory efficient attention
-                    _ = xformers.ops.memory_efficient_attention(
-                        torch.randn((1, 2, 40), device="cuda"),
-                        torch.randn((1, 2, 40), device="cuda"),
-                        torch.randn((1, 2, 40), device="cuda"),
-                    )
+                    dtype = None
+                    if attention_op is not None:
+                        op_fw, op_bw = attention_op
+                        dtype, *_ = op_fw.SUPPORTED_DTYPES
+                    q = torch.randn((1, 2, 40), device="cuda", dtype=dtype)
+                    _ = xformers.ops.memory_efficient_attention(q, q, q)
                 except Exception as e:
                     raise e
 

src/diffusers/pipelines/aura_flow/pipeline_aura_flow.py

@@ -160,8 +160,10 @@ def check_inputs(
         prompt_attention_mask=None,
         negative_prompt_attention_mask=None,
     ):
-        if height % 8 != 0 or width % 8 != 0:
-            raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
+        if height % (self.vae_scale_factor * 2) != 0 or width % (self.vae_scale_factor * 2) != 0:
+            raise ValueError(
+                f"`height` and `width` have to be divisible by {self.vae_scale_factor * 2} but are {height} and {width}."
+            )
 
         if prompt is not None and prompt_embeds is not None:
             raise ValueError(

src/diffusers/pipelines/hunyuan_video/pipeline_hunyuan_video.py

@@ -348,7 +348,7 @@ def check_inputs(
         prompt_template=None,
     ):
         if height % 16 != 0 or width % 16 != 0:
-            raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
+            raise ValueError(f"`height` and `width` have to be divisible by 16 but are {height} and {width}.")
 
         if callback_on_step_end_tensor_inputs is not None and not all(
             k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs

src/diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_adapter.py

@@ -22,7 +22,7 @@
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer
 
 from ...image_processor import VaeImageProcessor
-from ...loaders import StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import FromSingleFileMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, MultiAdapter, T2IAdapter, UNet2DConditionModel
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import KarrasDiffusionSchedulers
@@ -188,7 +188,7 @@ def retrieve_timesteps(
     return timesteps, num_inference_steps
 
 
-class StableDiffusionAdapterPipeline(DiffusionPipeline, StableDiffusionMixin):
+class StableDiffusionAdapterPipeline(DiffusionPipeline, StableDiffusionMixin, FromSingleFileMixin):
     r"""
     Pipeline for text-to-image generation using Stable Diffusion augmented with T2I-Adapter
     https://arxiv.org/abs/2302.08453

tests/pipelines/hunyuan_video/test_hunyuan_video.py

@@ -132,7 +132,7 @@ def get_dummy_components(self, num_layers: int = 1, num_single_layers: int = 1):
 
         torch.manual_seed(0)
         text_encoder = LlamaModel(llama_text_encoder_config)
-        tokenizer = LlamaTokenizer.from_pretrained("hf-internal-testing/tiny-random-LlamaForCausalLM")
+        tokenizer = LlamaTokenizer.from_pretrained("finetrainers/dummy-hunyaunvideo", subfolder="tokenizer")
 
         torch.manual_seed(0)
         text_encoder_2 = CLIPTextModel(clip_text_encoder_config)
@@ -155,10 +155,8 @@ def get_dummy_inputs(self, device, seed=0):
         else:
             generator = torch.Generator(device=device).manual_seed(seed)
 
-        # Cannot test with dummy prompt because tokenizers are not configured correctly.
-        # TODO(aryan): create dummy tokenizers and using from hub
         inputs = {
-            "prompt": "",
+            "prompt": "dance monkey",
             "prompt_template": {
                 "template": "{}",
                 "crop_start": 0,