diffusers 0.28.2__py3-none-any.whl → 0.29.1__py3-none-any.whl

diffusers/models/normalization.py

@@ -57,10 +57,12 @@ class AdaLayerNormZero(nn.Module):
         num_embeddings (`int`): The size of the embeddings dictionary.
     """
 
-    def __init__(self, embedding_dim: int, num_embeddings: int):
+    def __init__(self, embedding_dim: int, num_embeddings: Optional[int] = None):
         super().__init__()
-
-        self.emb = CombinedTimestepLabelEmbeddings(num_embeddings, embedding_dim)
+        if num_embeddings is not None:
+            self.emb = CombinedTimestepLabelEmbeddings(num_embeddings, embedding_dim)
+        else:
+            self.emb = None
 
         self.silu = nn.SiLU()
         self.linear = nn.Linear(embedding_dim, 6 * embedding_dim, bias=True)
@@ -69,11 +71,14 @@ class AdaLayerNormZero(nn.Module):
     def forward(
         self,
         x: torch.Tensor,
-        timestep: torch.Tensor,
-        class_labels: torch.LongTensor,
+        timestep: Optional[torch.Tensor] = None,
+        class_labels: Optional[torch.LongTensor] = None,
         hidden_dtype: Optional[torch.dtype] = None,
+        emb: Optional[torch.Tensor] = None,
     ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]:
-        emb = self.linear(self.silu(self.emb(timestep, class_labels, hidden_dtype=hidden_dtype)))
+        if self.emb is not None:
+            emb = self.emb(timestep, class_labels, hidden_dtype=hidden_dtype)
+        emb = self.linear(self.silu(emb))
         shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = emb.chunk(6, dim=1)
         x = self.norm(x) * (1 + scale_msa[:, None]) + shift_msa[:, None]
         return x, gate_msa, shift_mlp, scale_mlp, gate_mlp

diffusers/models/transformers/__init__.py

@@ -9,4 +9,5 @@ if is_torch_available():
     from .prior_transformer import PriorTransformer
     from .t5_film_transformer import T5FilmDecoder
     from .transformer_2d import Transformer2DModel
+    from .transformer_sd3 import SD3Transformer2DModel
     from .transformer_temporal import TransformerTemporalModel

diffusers/models/transformers/dual_transformer_2d.py

@@ -15,7 +15,8 @@ from typing import Optional
 
 from torch import nn
 
-from .transformer_2d import Transformer2DModel, Transformer2DModelOutput
+from ..modeling_outputs import Transformer2DModelOutput
+from .transformer_2d import Transformer2DModel
 
 
 class DualTransformer2DModel(nn.Module):
@@ -123,9 +124,9 @@ class DualTransformer2DModel(nn.Module):
                 tuple.
 
         Returns:
-            [`~models.transformer_2d.Transformer2DModelOutput`] or `tuple`:
-            [`~models.transformer_2d.Transformer2DModelOutput`] if `return_dict` is True, otherwise a `tuple`. When
-            returning a tuple, the first element is the sample tensor.
+            [`~models.transformers.transformer_2d.Transformer2DModelOutput`] or `tuple`:
+            [`~models.transformers.transformer_2d.Transformer2DModelOutput`] if `return_dict` is True, otherwise a
+            `tuple`. When returning a tuple, the first element is the sample tensor.
         """
         input_states = hidden_states
 

diffusers/models/transformers/hunyuan_transformer_2d.py

@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-from typing import Optional
+from typing import Dict, Optional, Union
 
 import torch
 import torch.nn.functional as F
@@ -21,7 +21,7 @@ from ...configuration_utils import ConfigMixin, register_to_config
 from ...utils import logging
 from ...utils.torch_utils import maybe_allow_in_graph
 from ..attention import FeedForward
-from ..attention_processor import Attention, HunyuanAttnProcessor2_0
+from ..attention_processor import Attention, AttentionProcessor, HunyuanAttnProcessor2_0
 from ..embeddings import (
     HunyuanCombinedTimestepTextSizeStyleEmbedding,
     PatchEmbed,
@@ -166,6 +166,7 @@ class HunyuanDiTBlock(nn.Module):
         self._chunk_size = None
         self._chunk_dim = 0
 
+    # Copied from diffusers.models.attention.BasicTransformerBlock.set_chunk_feed_forward
     def set_chunk_feed_forward(self, chunk_size: Optional[int], dim: int = 0):
         # Sets chunk feed-forward
         self._chunk_size = chunk_size
@@ -321,6 +322,110 @@ class HunyuanDiT2DModel(ModelMixin, ConfigMixin):
         self.norm_out = AdaLayerNormContinuous(self.inner_dim, self.inner_dim, elementwise_affine=False, eps=1e-6)
         self.proj_out = nn.Linear(self.inner_dim, patch_size * patch_size * self.out_channels, bias=True)
 
+    # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.fuse_qkv_projections
+    def fuse_qkv_projections(self):
+        """
+        Enables fused QKV projections. For self-attention modules, all projection matrices (i.e., query, key, value)
+        are fused. For cross-attention modules, key and value projection matrices are fused.
+
+        <Tip warning={true}>
+
+        This API is 🧪 experimental.
+
+        </Tip>
+        """
+        self.original_attn_processors = None
+
+        for _, attn_processor in self.attn_processors.items():
+            if "Added" in str(attn_processor.__class__.__name__):
+                raise ValueError("`fuse_qkv_projections()` is not supported for models having added KV projections.")
+
+        self.original_attn_processors = self.attn_processors
+
+        for module in self.modules():
+            if isinstance(module, Attention):
+                module.fuse_projections(fuse=True)
+
+    # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.unfuse_qkv_projections
+    def unfuse_qkv_projections(self):
+        """Disables the fused QKV projection if enabled.
+
+        <Tip warning={true}>
+
+        This API is 🧪 experimental.
+
+        </Tip>
+
+        """
+        if self.original_attn_processors is not None:
+            self.set_attn_processor(self.original_attn_processors)
+
+    @property
+    # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.attn_processors
+    def attn_processors(self) -> Dict[str, AttentionProcessor]:
+        r"""
+        Returns:
+            `dict` of attention processors: A dictionary containing all attention processors used in the model with
+            indexed by its weight name.
+        """
+        # set recursively
+        processors = {}
+
+        def fn_recursive_add_processors(name: str, module: torch.nn.Module, processors: Dict[str, AttentionProcessor]):
+            if hasattr(module, "get_processor"):
+                processors[f"{name}.processor"] = module.get_processor(return_deprecated_lora=True)
+
+            for sub_name, child in module.named_children():
+                fn_recursive_add_processors(f"{name}.{sub_name}", child, processors)
+
+            return processors
+
+        for name, module in self.named_children():
+            fn_recursive_add_processors(name, module, processors)
+
+        return processors
+
+    # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.set_attn_processor
+    def set_attn_processor(self, processor: Union[AttentionProcessor, Dict[str, AttentionProcessor]]):
+        r"""
+        Sets the attention processor to use to compute attention.
+
+        Parameters:
+            processor (`dict` of `AttentionProcessor` or only `AttentionProcessor`):
+                The instantiated processor class or a dictionary of processor classes that will be set as the processor
+                for **all** `Attention` layers.
+
+                If `processor` is a dict, the key needs to define the path to the corresponding cross attention
+                processor. This is strongly recommended when setting trainable attention processors.
+
+        """
+        count = len(self.attn_processors.keys())
+
+        if isinstance(processor, dict) and len(processor) != count:
+            raise ValueError(
+                f"A dict of processors was passed, but the number of processors {len(processor)} does not match the"
+                f" number of attention layers: {count}. Please make sure to pass {count} processor classes."
+            )
+
+        def fn_recursive_attn_processor(name: str, module: torch.nn.Module, processor):
+            if hasattr(module, "set_processor"):
+                if not isinstance(processor, dict):
+                    module.set_processor(processor)
+                else:
+                    module.set_processor(processor.pop(f"{name}.processor"))
+
+            for sub_name, child in module.named_children():
+                fn_recursive_attn_processor(f"{name}.{sub_name}", child, processor)
+
+        for name, module in self.named_children():
+            fn_recursive_attn_processor(name, module, processor)
+
+    def set_default_attn_processor(self):
+        """
+        Disables custom attention processors and sets the default attention implementation.
+        """
+        self.set_attn_processor(HunyuanAttnProcessor2_0())
+
     def forward(
         self,
         hidden_states,
@@ -425,3 +530,45 @@ class HunyuanDiT2DModel(ModelMixin, ConfigMixin):
         if not return_dict:
             return (output,)
         return Transformer2DModelOutput(sample=output)
+
+    # Copied from diffusers.models.unets.unet_3d_condition.UNet3DConditionModel.enable_forward_chunking
+    def enable_forward_chunking(self, chunk_size: Optional[int] = None, dim: int = 0) -> None:
+        """
+        Sets the attention processor to use [feed forward
+        chunking](https://huggingface.co/blog/reformer#2-chunked-feed-forward-layers).
+
+        Parameters:
+            chunk_size (`int`, *optional*):
+                The chunk size of the feed-forward layers. If not specified, will run feed-forward layer individually
+                over each tensor of dim=`dim`.
+            dim (`int`, *optional*, defaults to `0`):
+                The dimension over which the feed-forward computation should be chunked. Choose between dim=0 (batch)
+                or dim=1 (sequence length).
+        """
+        if dim not in [0, 1]:
+            raise ValueError(f"Make sure to set `dim` to either 0 or 1, not {dim}")
+
+        # By default chunk size is 1
+        chunk_size = chunk_size or 1
+
+        def fn_recursive_feed_forward(module: torch.nn.Module, chunk_size: int, dim: int):
+            if hasattr(module, "set_chunk_feed_forward"):
+                module.set_chunk_feed_forward(chunk_size=chunk_size, dim=dim)
+
+            for child in module.children():
+                fn_recursive_feed_forward(child, chunk_size, dim)
+
+        for module in self.children():
+            fn_recursive_feed_forward(module, chunk_size, dim)
+
+    # Copied from diffusers.models.unets.unet_3d_condition.UNet3DConditionModel.disable_forward_chunking
+    def disable_forward_chunking(self):
+        def fn_recursive_feed_forward(module: torch.nn.Module, chunk_size: int, dim: int):
+            if hasattr(module, "set_chunk_feed_forward"):
+                module.set_chunk_feed_forward(chunk_size=chunk_size, dim=dim)
+
+            for child in module.children():
+                fn_recursive_feed_forward(child, chunk_size, dim)
+
+        for module in self.children():
+            fn_recursive_feed_forward(module, None, 0)

diffusers/models/transformers/prior_transformer.py

@@ -266,13 +266,13 @@ class PriorTransformer(ModelMixin, ConfigMixin, UNet2DConditionLoadersMixin, Pef
             attention_mask (`torch.BoolTensor` of shape `(batch_size, num_embeddings)`):
                 Text mask for the text embeddings.
             return_dict (`bool`, *optional*, defaults to `True`):
-                Whether or not to return a [`~models.prior_transformer.PriorTransformerOutput`] instead of a plain
-                tuple.
+                Whether or not to return a [`~models.transformers.prior_transformer.PriorTransformerOutput`] instead of
+                a plain tuple.
 
         Returns:
-            [`~models.prior_transformer.PriorTransformerOutput`] or `tuple`:
-                If return_dict is True, a [`~models.prior_transformer.PriorTransformerOutput`] is returned, otherwise a
-                tuple is returned where the first element is the sample tensor.
+            [`~models.transformers.prior_transformer.PriorTransformerOutput`] or `tuple`:
+                If return_dict is True, a [`~models.transformers.prior_transformer.PriorTransformerOutput`] is
+                returned, otherwise a tuple is returned where the first element is the sample tensor.
         """
         batch_size = hidden_states.shape[0]
 

diffusers/models/transformers/transformer_2d.py

@@ -369,8 +369,8 @@ class Transformer2DModel(LegacyModelMixin, LegacyConfigMixin):
                 tuple.
 
         Returns:
-            If `return_dict` is True, an [`~models.transformer_2d.Transformer2DModelOutput`] is returned, otherwise a
-            `tuple` where the first element is the sample tensor.
+            If `return_dict` is True, an [`~models.transformers.transformer_2d.Transformer2DModelOutput`] is returned,
+            otherwise a `tuple` where the first element is the sample tensor.
         """
         if cross_attention_kwargs is not None:
             if cross_attention_kwargs.get("scale", None) is not None:

diffusers/models/transformers/transformer_sd3.py

@@ -0,0 +1,353 @@
+# Copyright 2024 Stability AI, The HuggingFace Team and The InstantX Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+from typing import Any, Dict, List, Optional, Union
+
+import torch
+import torch.nn as nn
+
+from ...configuration_utils import ConfigMixin, register_to_config
+from ...loaders import FromOriginalModelMixin, PeftAdapterMixin
+from ...models.attention import JointTransformerBlock
+from ...models.attention_processor import Attention, AttentionProcessor
+from ...models.modeling_utils import ModelMixin
+from ...models.normalization import AdaLayerNormContinuous
+from ...utils import USE_PEFT_BACKEND, is_torch_version, logging, scale_lora_layers, unscale_lora_layers
+from ..embeddings import CombinedTimestepTextProjEmbeddings, PatchEmbed
+from .transformer_2d import Transformer2DModelOutput
+
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+class SD3Transformer2DModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOriginalModelMixin):
+    """
+    The Transformer model introduced in Stable Diffusion 3.
+
+    Reference: https://arxiv.org/abs/2403.03206
+
+    Parameters:
+        sample_size (`int`): The width of the latent images. This is fixed during training since
+            it is used to learn a number of position embeddings.
+        patch_size (`int`): Patch size to turn the input data into small patches.
+        in_channels (`int`, *optional*, defaults to 16): The number of channels in the input.
+        num_layers (`int`, *optional*, defaults to 18): The number of layers of Transformer blocks to use.
+        attention_head_dim (`int`, *optional*, defaults to 64): The number of channels in each head.
+        num_attention_heads (`int`, *optional*, defaults to 18): The number of heads to use for multi-head attention.
+        cross_attention_dim (`int`, *optional*): The number of `encoder_hidden_states` dimensions to use.
+        caption_projection_dim (`int`): Number of dimensions to use when projecting the `encoder_hidden_states`.
+        pooled_projection_dim (`int`): Number of dimensions to use when projecting the `pooled_projections`.
+        out_channels (`int`, defaults to 16): Number of output channels.
+
+    """
+
+    _supports_gradient_checkpointing = True
+
+    @register_to_config
+    def __init__(
+        self,
+        sample_size: int = 128,
+        patch_size: int = 2,
+        in_channels: int = 16,
+        num_layers: int = 18,
+        attention_head_dim: int = 64,
+        num_attention_heads: int = 18,
+        joint_attention_dim: int = 4096,
+        caption_projection_dim: int = 1152,
+        pooled_projection_dim: int = 2048,
+        out_channels: int = 16,
+        pos_embed_max_size: int = 96,
+    ):
+        super().__init__()
+        default_out_channels = in_channels
+        self.out_channels = out_channels if out_channels is not None else default_out_channels
+        self.inner_dim = self.config.num_attention_heads * self.config.attention_head_dim
+
+        self.pos_embed = PatchEmbed(
+            height=self.config.sample_size,
+            width=self.config.sample_size,
+            patch_size=self.config.patch_size,
+            in_channels=self.config.in_channels,
+            embed_dim=self.inner_dim,
+            pos_embed_max_size=pos_embed_max_size,  # hard-code for now.
+        )
+        self.time_text_embed = CombinedTimestepTextProjEmbeddings(
+            embedding_dim=self.inner_dim, pooled_projection_dim=self.config.pooled_projection_dim
+        )
+        self.context_embedder = nn.Linear(self.config.joint_attention_dim, self.config.caption_projection_dim)
+
+        # `attention_head_dim` is doubled to account for the mixing.
+        # It needs to crafted when we get the actual checkpoints.
+        self.transformer_blocks = nn.ModuleList(
+            [
+                JointTransformerBlock(
+                    dim=self.inner_dim,
+                    num_attention_heads=self.config.num_attention_heads,
+                    attention_head_dim=self.inner_dim,
+                    context_pre_only=i == num_layers - 1,
+                )
+                for i in range(self.config.num_layers)
+            ]
+        )
+
+        self.norm_out = AdaLayerNormContinuous(self.inner_dim, self.inner_dim, elementwise_affine=False, eps=1e-6)
+        self.proj_out = nn.Linear(self.inner_dim, patch_size * patch_size * self.out_channels, bias=True)
+
+        self.gradient_checkpointing = False
+
+    # Copied from diffusers.models.unets.unet_3d_condition.UNet3DConditionModel.enable_forward_chunking
+    def enable_forward_chunking(self, chunk_size: Optional[int] = None, dim: int = 0) -> None:
+        """
+        Sets the attention processor to use [feed forward
+        chunking](https://huggingface.co/blog/reformer#2-chunked-feed-forward-layers).
+
+        Parameters:
+            chunk_size (`int`, *optional*):
+                The chunk size of the feed-forward layers. If not specified, will run feed-forward layer individually
+                over each tensor of dim=`dim`.
+            dim (`int`, *optional*, defaults to `0`):
+                The dimension over which the feed-forward computation should be chunked. Choose between dim=0 (batch)
+                or dim=1 (sequence length).
+        """
+        if dim not in [0, 1]:
+            raise ValueError(f"Make sure to set `dim` to either 0 or 1, not {dim}")
+
+        # By default chunk size is 1
+        chunk_size = chunk_size or 1
+
+        def fn_recursive_feed_forward(module: torch.nn.Module, chunk_size: int, dim: int):
+            if hasattr(module, "set_chunk_feed_forward"):
+                module.set_chunk_feed_forward(chunk_size=chunk_size, dim=dim)
+
+            for child in module.children():
+                fn_recursive_feed_forward(child, chunk_size, dim)
+
+        for module in self.children():
+            fn_recursive_feed_forward(module, chunk_size, dim)
+
+    @property
+    # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.attn_processors
+    def attn_processors(self) -> Dict[str, AttentionProcessor]:
+        r"""
+        Returns:
+            `dict` of attention processors: A dictionary containing all attention processors used in the model with
+            indexed by its weight name.
+        """
+        # set recursively
+        processors = {}
+
+        def fn_recursive_add_processors(name: str, module: torch.nn.Module, processors: Dict[str, AttentionProcessor]):
+            if hasattr(module, "get_processor"):
+                processors[f"{name}.processor"] = module.get_processor(return_deprecated_lora=True)
+
+            for sub_name, child in module.named_children():
+                fn_recursive_add_processors(f"{name}.{sub_name}", child, processors)
+
+            return processors
+
+        for name, module in self.named_children():
+            fn_recursive_add_processors(name, module, processors)
+
+        return processors
+
+    # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.set_attn_processor
+    def set_attn_processor(self, processor: Union[AttentionProcessor, Dict[str, AttentionProcessor]]):
+        r"""
+        Sets the attention processor to use to compute attention.
+
+        Parameters:
+            processor (`dict` of `AttentionProcessor` or only `AttentionProcessor`):
+                The instantiated processor class or a dictionary of processor classes that will be set as the processor
+                for **all** `Attention` layers.
+
+                If `processor` is a dict, the key needs to define the path to the corresponding cross attention
+                processor. This is strongly recommended when setting trainable attention processors.
+
+        """
+        count = len(self.attn_processors.keys())
+
+        if isinstance(processor, dict) and len(processor) != count:
+            raise ValueError(
+                f"A dict of processors was passed, but the number of processors {len(processor)} does not match the"
+                f" number of attention layers: {count}. Please make sure to pass {count} processor classes."
+            )
+
+        def fn_recursive_attn_processor(name: str, module: torch.nn.Module, processor):
+            if hasattr(module, "set_processor"):
+                if not isinstance(processor, dict):
+                    module.set_processor(processor)
+                else:
+                    module.set_processor(processor.pop(f"{name}.processor"))
+
+            for sub_name, child in module.named_children():
+                fn_recursive_attn_processor(f"{name}.{sub_name}", child, processor)
+
+        for name, module in self.named_children():
+            fn_recursive_attn_processor(name, module, processor)
+
+    # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.fuse_qkv_projections
+    def fuse_qkv_projections(self):
+        """
+        Enables fused QKV projections. For self-attention modules, all projection matrices (i.e., query, key, value)
+        are fused. For cross-attention modules, key and value projection matrices are fused.
+
+        <Tip warning={true}>
+
+        This API is 🧪 experimental.
+
+        </Tip>
+        """
+        self.original_attn_processors = None
+
+        for _, attn_processor in self.attn_processors.items():
+            if "Added" in str(attn_processor.__class__.__name__):
+                raise ValueError("`fuse_qkv_projections()` is not supported for models having added KV projections.")
+
+        self.original_attn_processors = self.attn_processors
+
+        for module in self.modules():
+            if isinstance(module, Attention):
+                module.fuse_projections(fuse=True)
+
+    # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.unfuse_qkv_projections
+    def unfuse_qkv_projections(self):
+        """Disables the fused QKV projection if enabled.
+
+        <Tip warning={true}>
+
+        This API is 🧪 experimental.
+
+        </Tip>
+
+        """
+        if self.original_attn_processors is not None:
+            self.set_attn_processor(self.original_attn_processors)
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if hasattr(module, "gradient_checkpointing"):
+            module.gradient_checkpointing = value
+
+    def forward(
+        self,
+        hidden_states: torch.FloatTensor,
+        encoder_hidden_states: torch.FloatTensor = None,
+        pooled_projections: torch.FloatTensor = None,
+        timestep: torch.LongTensor = None,
+        block_controlnet_hidden_states: List = None,
+        joint_attention_kwargs: Optional[Dict[str, Any]] = None,
+        return_dict: bool = True,
+    ) -> Union[torch.FloatTensor, Transformer2DModelOutput]:
+        """
+        The [`SD3Transformer2DModel`] forward method.
+
+        Args:
+            hidden_states (`torch.FloatTensor` of shape `(batch size, channel, height, width)`):
+                Input `hidden_states`.
+            encoder_hidden_states (`torch.FloatTensor` of shape `(batch size, sequence_len, embed_dims)`):
+                Conditional embeddings (embeddings computed from the input conditions such as prompts) to use.
+            pooled_projections (`torch.FloatTensor` of shape `(batch_size, projection_dim)`): Embeddings projected
+                from the embeddings of input conditions.
+            timestep ( `torch.LongTensor`):
+                Used to indicate denoising step.
+            block_controlnet_hidden_states: (`list` of `torch.Tensor`):
+                A list of tensors that if specified are added to the residuals of transformer blocks.
+            joint_attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+                `self.processor` in
+                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~models.transformer_2d.Transformer2DModelOutput`] instead of a plain
+                tuple.
+
+        Returns:
+            If `return_dict` is True, an [`~models.transformer_2d.Transformer2DModelOutput`] is returned, otherwise a
+            `tuple` where the first element is the sample tensor.
+        """
+        if joint_attention_kwargs is not None:
+            joint_attention_kwargs = joint_attention_kwargs.copy()
+            lora_scale = joint_attention_kwargs.pop("scale", 1.0)
+        else:
+            lora_scale = 1.0
+
+        if USE_PEFT_BACKEND:
+            # weight the lora layers by setting `lora_scale` for each PEFT layer
+            scale_lora_layers(self, lora_scale)
+        else:
+            if joint_attention_kwargs is not None and joint_attention_kwargs.get("scale", None) is not None:
+                logger.warning(
+                    "Passing `scale` via `joint_attention_kwargs` when not using the PEFT backend is ineffective."
+                )
+
+        height, width = hidden_states.shape[-2:]
+
+        hidden_states = self.pos_embed(hidden_states)  # takes care of adding positional embeddings too.
+        temb = self.time_text_embed(timestep, pooled_projections)
+        encoder_hidden_states = self.context_embedder(encoder_hidden_states)
+
+        for index_block, block in enumerate(self.transformer_blocks):
+            if self.training and self.gradient_checkpointing:
+
+                def create_custom_forward(module, return_dict=None):
+                    def custom_forward(*inputs):
+                        if return_dict is not None:
+                            return module(*inputs, return_dict=return_dict)
+                        else:
+                            return module(*inputs)
+
+                    return custom_forward
+
+                ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
+                encoder_hidden_states, hidden_states = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(block),
+                    hidden_states,
+                    encoder_hidden_states,
+                    temb,
+                    **ckpt_kwargs,
+                )
+
+            else:
+                encoder_hidden_states, hidden_states = block(
+                    hidden_states=hidden_states, encoder_hidden_states=encoder_hidden_states, temb=temb
+                )
+
+            # controlnet residual
+            if block_controlnet_hidden_states is not None and block.context_pre_only is False:
+                interval_control = len(self.transformer_blocks) // len(block_controlnet_hidden_states)
+                hidden_states = hidden_states + block_controlnet_hidden_states[index_block // interval_control]
+
+        hidden_states = self.norm_out(hidden_states, temb)
+        hidden_states = self.proj_out(hidden_states)
+
+        # unpatchify
+        patch_size = self.config.patch_size
+        height = height // patch_size
+        width = width // patch_size
+
+        hidden_states = hidden_states.reshape(
+            shape=(hidden_states.shape[0], height, width, patch_size, patch_size, self.out_channels)
+        )
+        hidden_states = torch.einsum("nhwpqc->nchpwq", hidden_states)
+        output = hidden_states.reshape(
+            shape=(hidden_states.shape[0], self.out_channels, height * patch_size, width * patch_size)
+        )
+
+        if USE_PEFT_BACKEND:
+            # remove `lora_scale` from each PEFT layer
+            unscale_lora_layers(self, lora_scale)
+
+        if not return_dict:
+            return (output,)
+
+        return Transformer2DModelOutput(sample=output)

diffusers/models/transformers/transformer_temporal.py

@@ -149,13 +149,14 @@ class TransformerTemporalModel(ModelMixin, ConfigMixin):
                 `self.processor` in
                 [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
             return_dict (`bool`, *optional*, defaults to `True`):
-                Whether or not to return a [`~models.unets.unet_2d_condition.UNet2DConditionOutput`] instead of a plain
-                tuple.
+                Whether or not to return a [`~models.transformers.transformer_temporal.TransformerTemporalModelOutput`]
+                instead of a plain tuple.
 
         Returns:
-            [`~models.transformer_temporal.TransformerTemporalModelOutput`] or `tuple`:
-                If `return_dict` is True, an [`~models.transformer_temporal.TransformerTemporalModelOutput`] is
-                returned, otherwise a `tuple` where the first element is the sample tensor.
+            [`~models.transformers.transformer_temporal.TransformerTemporalModelOutput`] or `tuple`:
+                If `return_dict` is True, an
+                [`~models.transformers.transformer_temporal.TransformerTemporalModelOutput`] is returned, otherwise a
+                `tuple` where the first element is the sample tensor.
         """
         # 1. Input
         batch_frames, channel, height, width = hidden_states.shape
@@ -294,13 +295,14 @@ class TransformerSpatioTemporalModel(nn.Module):
                 A tensor indicating whether the input contains only images. 1 indicates that the input contains only
                 images, 0 indicates that the input contains video frames.
             return_dict (`bool`, *optional*, defaults to `True`):
-                Whether or not to return a [`~models.transformer_temporal.TransformerTemporalModelOutput`] instead of a
-                plain tuple.
+                Whether or not to return a [`~models.transformers.transformer_temporal.TransformerTemporalModelOutput`]
+                instead of a plain tuple.
 
         Returns:
-            [`~models.transformer_temporal.TransformerTemporalModelOutput`] or `tuple`:
-                If `return_dict` is True, an [`~models.transformer_temporal.TransformerTemporalModelOutput`] is
-                returned, otherwise a `tuple` where the first element is the sample tensor.
+            [`~models.transformers.transformer_temporal.TransformerTemporalModelOutput`] or `tuple`:
+                If `return_dict` is True, an
+                [`~models.transformers.transformer_temporal.TransformerTemporalModelOutput`] is returned, otherwise a
+                `tuple` where the first element is the sample tensor.
         """
         # 1. Input
         batch_frames, _, height, width = hidden_states.shape