diffusers 0.23.1__py3-none-any.whl → 0.24.0__py3-none-any.whl

diffusers/pipelines/versatile_diffusion/modeling_text_unet.py

@@ -425,10 +425,7 @@ class UNetFlatConditionModel(ModelMixin, ConfigMixin):
 
         if num_attention_heads is not None:
             raise ValueError(
-                "At the moment it is not possible to define the number of attention heads via `num_attention_heads`"
-                " because of a naming issue as described in"
-                " https://github.com/huggingface/diffusers/issues/2011#issuecomment-1547958131. Passing"
-                " `num_attention_heads` will only be supported in diffusers v0.19."
+                "At the moment it is not possible to define the number of attention heads via `num_attention_heads` because of a naming issue as described in https://github.com/huggingface/diffusers/issues/2011#issuecomment-1547958131. Passing `num_attention_heads` will only be supported in diffusers v0.19."
             )
 
         # If `num_attention_heads` is not defined (which is the case for most models)
@@ -442,44 +439,37 @@ class UNetFlatConditionModel(ModelMixin, ConfigMixin):
         # Check inputs
         if len(down_block_types) != len(up_block_types):
             raise ValueError(
-                "Must provide the same number of `down_block_types` as `up_block_types`. `down_block_types`:"
-                f" {down_block_types}. `up_block_types`: {up_block_types}."
+                f"Must provide the same number of `down_block_types` as `up_block_types`. `down_block_types`: {down_block_types}. `up_block_types`: {up_block_types}."
             )
 
         if len(block_out_channels) != len(down_block_types):
             raise ValueError(
-                "Must provide the same number of `block_out_channels` as `down_block_types`. `block_out_channels`:"
-                f" {block_out_channels}. `down_block_types`: {down_block_types}."
+                f"Must provide the same number of `block_out_channels` as `down_block_types`. `block_out_channels`: {block_out_channels}. `down_block_types`: {down_block_types}."
             )
 
         if not isinstance(only_cross_attention, bool) and len(only_cross_attention) != len(down_block_types):
             raise ValueError(
-                "Must provide the same number of `only_cross_attention` as `down_block_types`."
-                f" `only_cross_attention`: {only_cross_attention}. `down_block_types`: {down_block_types}."
+                f"Must provide the same number of `only_cross_attention` as `down_block_types`. `only_cross_attention`: {only_cross_attention}. `down_block_types`: {down_block_types}."
             )
 
         if not isinstance(num_attention_heads, int) and len(num_attention_heads) != len(down_block_types):
             raise ValueError(
-                "Must provide the same number of `num_attention_heads` as `down_block_types`. `num_attention_heads`:"
-                f" {num_attention_heads}. `down_block_types`: {down_block_types}."
+                f"Must provide the same number of `num_attention_heads` as `down_block_types`. `num_attention_heads`: {num_attention_heads}. `down_block_types`: {down_block_types}."
             )
 
         if not isinstance(attention_head_dim, int) and len(attention_head_dim) != len(down_block_types):
             raise ValueError(
-                "Must provide the same number of `attention_head_dim` as `down_block_types`. `attention_head_dim`:"
-                f" {attention_head_dim}. `down_block_types`: {down_block_types}."
+                f"Must provide the same number of `attention_head_dim` as `down_block_types`. `attention_head_dim`: {attention_head_dim}. `down_block_types`: {down_block_types}."
             )
 
         if isinstance(cross_attention_dim, list) and len(cross_attention_dim) != len(down_block_types):
             raise ValueError(
-                "Must provide the same number of `cross_attention_dim` as `down_block_types`. `cross_attention_dim`:"
-                f" {cross_attention_dim}. `down_block_types`: {down_block_types}."
+                f"Must provide the same number of `cross_attention_dim` as `down_block_types`. `cross_attention_dim`: {cross_attention_dim}. `down_block_types`: {down_block_types}."
             )
 
         if not isinstance(layers_per_block, int) and len(layers_per_block) != len(down_block_types):
             raise ValueError(
-                "Must provide the same number of `layers_per_block` as `down_block_types`. `layers_per_block`:"
-                f" {layers_per_block}. `down_block_types`: {down_block_types}."
+                f"Must provide the same number of `layers_per_block` as `down_block_types`. `layers_per_block`: {layers_per_block}. `down_block_types`: {down_block_types}."
             )
         if isinstance(transformer_layers_per_block, list) and reverse_transformer_layers_per_block is None:
             for layer_number_per_block in transformer_layers_per_block:
@@ -897,8 +887,7 @@ class UNetFlatConditionModel(ModelMixin, ConfigMixin):
             processor = AttnProcessor()
         else:
             raise ValueError(
-                "Cannot call `set_default_attn_processor` when attention processors are of type"
-                f" {next(iter(self.attn_processors.values()))}"
+                f"Cannot call `set_default_attn_processor` when attention processors are of type {next(iter(self.attn_processors.values()))}"
             )
 
         self.set_attn_processor(processor, _remove_lora=True)
@@ -1166,8 +1155,7 @@ class UNetFlatConditionModel(ModelMixin, ConfigMixin):
             # Kandinsky 2.1 - style
             if "image_embeds" not in added_cond_kwargs:
                 raise ValueError(
-                    f"{self.__class__} has the config param `addition_embed_type` set to 'text_image' which requires"
-                    " the keyword argument `image_embeds` to be passed in `added_cond_kwargs`"
+                    f"{self.__class__} has the config param `addition_embed_type` set to 'text_image' which requires the keyword argument `image_embeds` to be passed in `added_cond_kwargs`"
                 )
 
             image_embs = added_cond_kwargs.get("image_embeds")
@@ -1177,14 +1165,12 @@ class UNetFlatConditionModel(ModelMixin, ConfigMixin):
             # SDXL - style
             if "text_embeds" not in added_cond_kwargs:
                 raise ValueError(
-                    f"{self.__class__} has the config param `addition_embed_type` set to 'text_time' which requires"
-                    " the keyword argument `text_embeds` to be passed in `added_cond_kwargs`"
+                    f"{self.__class__} has the config param `addition_embed_type` set to 'text_time' which requires the keyword argument `text_embeds` to be passed in `added_cond_kwargs`"
                 )
             text_embeds = added_cond_kwargs.get("text_embeds")
             if "time_ids" not in added_cond_kwargs:
                 raise ValueError(
-                    f"{self.__class__} has the config param `addition_embed_type` set to 'text_time' which requires"
-                    " the keyword argument `time_ids` to be passed in `added_cond_kwargs`"
+                    f"{self.__class__} has the config param `addition_embed_type` set to 'text_time' which requires the keyword argument `time_ids` to be passed in `added_cond_kwargs`"
                 )
             time_ids = added_cond_kwargs.get("time_ids")
             time_embeds = self.add_time_proj(time_ids.flatten())
@@ -1196,8 +1182,7 @@ class UNetFlatConditionModel(ModelMixin, ConfigMixin):
             # Kandinsky 2.2 - style
             if "image_embeds" not in added_cond_kwargs:
                 raise ValueError(
-                    f"{self.__class__} has the config param `addition_embed_type` set to 'image' which requires the"
-                    " keyword argument `image_embeds` to be passed in `added_cond_kwargs`"
+                    f"{self.__class__} has the config param `addition_embed_type` set to 'image' which requires the keyword argument `image_embeds` to be passed in `added_cond_kwargs`"
                 )
             image_embs = added_cond_kwargs.get("image_embeds")
             aug_emb = self.add_embedding(image_embs)
@@ -1205,8 +1190,7 @@ class UNetFlatConditionModel(ModelMixin, ConfigMixin):
             # Kandinsky 2.2 - style
             if "image_embeds" not in added_cond_kwargs or "hint" not in added_cond_kwargs:
                 raise ValueError(
-                    f"{self.__class__} has the config param `addition_embed_type` set to 'image_hint' which requires"
-                    " the keyword arguments `image_embeds` and `hint` to be passed in `added_cond_kwargs`"
+                    f"{self.__class__} has the config param `addition_embed_type` set to 'image_hint' which requires the keyword arguments `image_embeds` and `hint` to be passed in `added_cond_kwargs`"
                 )
             image_embs = added_cond_kwargs.get("image_embeds")
             hint = added_cond_kwargs.get("hint")
@@ -1224,8 +1208,7 @@ class UNetFlatConditionModel(ModelMixin, ConfigMixin):
             # Kadinsky 2.1 - style
             if "image_embeds" not in added_cond_kwargs:
                 raise ValueError(
-                    f"{self.__class__} has the config param `encoder_hid_dim_type` set to 'text_image_proj' which"
-                    " requires the keyword argument `image_embeds` to be passed in  `added_conditions`"
+                    f"{self.__class__} has the config param `encoder_hid_dim_type` set to 'text_image_proj' which requires the keyword argument `image_embeds` to be passed in  `added_conditions`"
                 )
 
             image_embeds = added_cond_kwargs.get("image_embeds")
@@ -1234,11 +1217,19 @@ class UNetFlatConditionModel(ModelMixin, ConfigMixin):
             # Kandinsky 2.2 - style
             if "image_embeds" not in added_cond_kwargs:
                 raise ValueError(
-                    f"{self.__class__} has the config param `encoder_hid_dim_type` set to 'image_proj' which requires"
-                    " the keyword argument `image_embeds` to be passed in  `added_conditions`"
+                    f"{self.__class__} has the config param `encoder_hid_dim_type` set to 'image_proj' which requires the keyword argument `image_embeds` to be passed in  `added_conditions`"
                 )
             image_embeds = added_cond_kwargs.get("image_embeds")
             encoder_hidden_states = self.encoder_hid_proj(image_embeds)
+        elif self.encoder_hid_proj is not None and self.config.encoder_hid_dim_type == "ip_image_proj":
+            if "image_embeds" not in added_cond_kwargs:
+                raise ValueError(
+                    f"{self.__class__} has the config param `encoder_hid_dim_type` set to 'ip_image_proj' which requires the keyword argument `image_embeds` to be passed in  `added_conditions`"
+                )
+            image_embeds = added_cond_kwargs.get("image_embeds")
+            image_embeds = self.encoder_hid_proj(image_embeds).to(encoder_hidden_states.dtype)
+            encoder_hidden_states = torch.cat([encoder_hidden_states, image_embeds], dim=1)
+
         # 2. pre-process
         sample = self.conv_in(sample)
 
@@ -1264,10 +1255,9 @@ class UNetFlatConditionModel(ModelMixin, ConfigMixin):
             deprecate(
                 "T2I should not use down_block_additional_residuals",
                 "1.3.0",
-                "Passing intrablock residual connections with `down_block_additional_residuals` is deprecated         "
-                "               and will be removed in diffusers 1.3.0.  `down_block_additional_residuals` should only"
-                " be used                        for ControlNet. Please make sure use"
-                " `down_intrablock_additional_residuals` instead. ",
+                "Passing intrablock residual connections with `down_block_additional_residuals` is deprecated \
+                       and will be removed in diffusers 1.3.0.  `down_block_additional_residuals` should only be used \
+                       for ControlNet. Please make sure use `down_intrablock_additional_residuals` instead. ",
                 standard_warn=False,
             )
             down_intrablock_additional_residuals = down_block_additional_residuals
@@ -2102,8 +2092,7 @@ class UNetMidBlockFlat(nn.Module):
 
         if attention_head_dim is None:
             logger.warn(
-                "It is not recommend to pass `attention_head_dim=None`. Defaulting `attention_head_dim` to"
-                f" `in_channels`: {in_channels}."
+                f"It is not recommend to pass `attention_head_dim=None`. Defaulting `attention_head_dim` to `in_channels`: {in_channels}."
             )
             attention_head_dim = in_channels
 

diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_dual_guided.py

@@ -58,6 +58,7 @@ class VersatileDiffusionDualGuidedPipeline(DiffusionPipeline):
             A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
             [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
     """
+
     model_cpu_offload_seq = "bert->unet->vqvae"
 
     tokenizer: CLIPTokenizer

diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_image_variation.py

@@ -52,6 +52,7 @@ class VersatileDiffusionImageVariationPipeline(DiffusionPipeline):
             A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
             [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
     """
+
     model_cpu_offload_seq = "bert->unet->vqvae"
 
     image_feature_extractor: CLIPImageProcessor

diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_text_to_image.py

@@ -51,6 +51,7 @@ class VersatileDiffusionTextToImagePipeline(DiffusionPipeline):
             A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
             [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
     """
+
     model_cpu_offload_seq = "bert->unet->vqvae"
 
     tokenizer: CLIPTokenizer

diffusers/pipelines/wuerstchen/modeling_wuerstchen_common.py

@@ -17,6 +17,8 @@ import torch
 import torch.nn as nn
 
 from ...models.attention_processor import Attention
+from ...models.lora import LoRACompatibleConv, LoRACompatibleLinear
+from ...utils import USE_PEFT_BACKEND
 
 
 class WuerstchenLayerNorm(nn.LayerNorm):
@@ -32,7 +34,8 @@ class WuerstchenLayerNorm(nn.LayerNorm):
 class TimestepBlock(nn.Module):
     def __init__(self, c, c_timestep):
         super().__init__()
-        self.mapper = nn.Linear(c_timestep, c * 2)
+        linear_cls = nn.Linear if USE_PEFT_BACKEND else LoRACompatibleLinear
+        self.mapper = linear_cls(c_timestep, c * 2)
 
     def forward(self, x, t):
         a, b = self.mapper(t)[:, :, None, None].chunk(2, dim=1)
@@ -42,10 +45,14 @@ class TimestepBlock(nn.Module):
 class ResBlock(nn.Module):
     def __init__(self, c, c_skip=0, kernel_size=3, dropout=0.0):
         super().__init__()
-        self.depthwise = nn.Conv2d(c + c_skip, c, kernel_size=kernel_size, padding=kernel_size // 2, groups=c)
+
+        conv_cls = nn.Conv2d if USE_PEFT_BACKEND else LoRACompatibleConv
+        linear_cls = nn.Linear if USE_PEFT_BACKEND else LoRACompatibleLinear
+
+        self.depthwise = conv_cls(c + c_skip, c, kernel_size=kernel_size, padding=kernel_size // 2, groups=c)
         self.norm = WuerstchenLayerNorm(c, elementwise_affine=False, eps=1e-6)
         self.channelwise = nn.Sequential(
-            nn.Linear(c, c * 4), nn.GELU(), GlobalResponseNorm(c * 4), nn.Dropout(dropout), nn.Linear(c * 4, c)
+            linear_cls(c, c * 4), nn.GELU(), GlobalResponseNorm(c * 4), nn.Dropout(dropout), linear_cls(c * 4, c)
         )
 
     def forward(self, x, x_skip=None):
@@ -73,10 +80,13 @@ class GlobalResponseNorm(nn.Module):
 class AttnBlock(nn.Module):
     def __init__(self, c, c_cond, nhead, self_attn=True, dropout=0.0):
         super().__init__()
+
+        linear_cls = nn.Linear if USE_PEFT_BACKEND else LoRACompatibleLinear
+
         self.self_attn = self_attn
         self.norm = WuerstchenLayerNorm(c, elementwise_affine=False, eps=1e-6)
         self.attention = Attention(query_dim=c, heads=nhead, dim_head=c // nhead, dropout=dropout, bias=True)
-        self.kv_mapper = nn.Sequential(nn.SiLU(), nn.Linear(c_cond, c))
+        self.kv_mapper = nn.Sequential(nn.SiLU(), linear_cls(c_cond, c))
 
     def forward(self, x, kv):
         kv = self.kv_mapper(kv)

diffusers/pipelines/wuerstchen/modeling_wuerstchen_prior.py

@@ -28,8 +28,9 @@ from ...models.attention_processor import (
     AttnAddedKVProcessor,
     AttnProcessor,
 )
+from ...models.lora import LoRACompatibleConv, LoRACompatibleLinear
 from ...models.modeling_utils import ModelMixin
-from ...utils import is_torch_version
+from ...utils import USE_PEFT_BACKEND, is_torch_version
 from .modeling_wuerstchen_common import AttnBlock, ResBlock, TimestepBlock, WuerstchenLayerNorm
 
 
@@ -40,12 +41,15 @@ class WuerstchenPrior(ModelMixin, ConfigMixin, UNet2DConditionLoadersMixin):
     @register_to_config
     def __init__(self, c_in=16, c=1280, c_cond=1024, c_r=64, depth=16, nhead=16, dropout=0.1):
         super().__init__()
+        conv_cls = nn.Conv2d if USE_PEFT_BACKEND else LoRACompatibleConv
+        linear_cls = nn.Linear if USE_PEFT_BACKEND else LoRACompatibleLinear
+
         self.c_r = c_r
-        self.projection = nn.Conv2d(c_in, c, kernel_size=1)
+        self.projection = conv_cls(c_in, c, kernel_size=1)
         self.cond_mapper = nn.Sequential(
-            nn.Linear(c_cond, c),
+            linear_cls(c_cond, c),
             nn.LeakyReLU(0.2),
-            nn.Linear(c, c),
+            linear_cls(c, c),
         )
 
         self.blocks = nn.ModuleList()
@@ -55,7 +59,7 @@ class WuerstchenPrior(ModelMixin, ConfigMixin, UNet2DConditionLoadersMixin):
             self.blocks.append(AttnBlock(c, c, nhead, self_attn=True, dropout=dropout))
         self.out = nn.Sequential(
             WuerstchenLayerNorm(c, elementwise_affine=False, eps=1e-6),
-            nn.Conv2d(c, c_in * 2, kernel_size=1),
+            conv_cls(c, c_in * 2, kernel_size=1),
         )
 
         self.gradient_checkpointing = False

diffusers/pipelines/wuerstchen/pipeline_wuerstchen.py

@@ -269,7 +269,7 @@ class WuerstchenDecoderPipeline(DiffusionPipeline):
             callback_on_step_end_tensor_inputs (`List`, *optional*):
                 The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
                 will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
-                `._callback_tensor_inputs` attribute of your pipeine class.
+                `._callback_tensor_inputs` attribute of your pipeline class.
 
         Examples:
 

diffusers/pipelines/wuerstchen/pipeline_wuerstchen_combined.py

@@ -234,7 +234,7 @@ class WuerstchenCombinedPipeline(DiffusionPipeline):
             prior_callback_on_step_end_tensor_inputs (`List`, *optional*):
                 The list of tensor inputs for the `prior_callback_on_step_end` function. The tensors specified in the
                 list will be passed as `callback_kwargs` argument. You will only be able to include variables listed in
-                the `._callback_tensor_inputs` attribute of your pipeine class.
+                the `._callback_tensor_inputs` attribute of your pipeline class.
             callback_on_step_end (`Callable`, *optional*):
                 A function that calls at the end of each denoising steps during the inference. The function is called
                 with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
@@ -243,7 +243,7 @@ class WuerstchenCombinedPipeline(DiffusionPipeline):
             callback_on_step_end_tensor_inputs (`List`, *optional*):
                 The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
                 will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
-                `._callback_tensor_inputs` attribute of your pipeine class.
+                `._callback_tensor_inputs` attribute of your pipeline class.
 
         Examples:
 

diffusers/pipelines/wuerstchen/pipeline_wuerstchen_prior.py

@@ -349,7 +349,7 @@ class WuerstchenPriorPipeline(DiffusionPipeline, LoraLoaderMixin):
             callback_on_step_end_tensor_inputs (`List`, *optional*):
                 The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
                 will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
-                `._callback_tensor_inputs` attribute of your pipeine class.
+                `._callback_tensor_inputs` attribute of your pipeline class.
 
         Examples:
 

diffusers/schedulers/__init__.py

@@ -38,6 +38,7 @@ except OptionalDependencyNotAvailable:
     _dummy_modules.update(get_objects_from_module(dummy_pt_objects))
 
 else:
+    _import_structure["deprecated"] = ["KarrasVeScheduler", "ScoreSdeVpScheduler"]
     _import_structure["scheduling_consistency_decoder"] = ["ConsistencyDecoderScheduler"]
     _import_structure["scheduling_consistency_models"] = ["CMStochasticIterativeScheduler"]
     _import_structure["scheduling_ddim"] = ["DDIMScheduler"]
@@ -56,12 +57,10 @@ else:
     _import_structure["scheduling_ipndm"] = ["IPNDMScheduler"]
     _import_structure["scheduling_k_dpm_2_ancestral_discrete"] = ["KDPM2AncestralDiscreteScheduler"]
     _import_structure["scheduling_k_dpm_2_discrete"] = ["KDPM2DiscreteScheduler"]
-    _import_structure["scheduling_karras_ve"] = ["KarrasVeScheduler"]
     _import_structure["scheduling_lcm"] = ["LCMScheduler"]
     _import_structure["scheduling_pndm"] = ["PNDMScheduler"]
     _import_structure["scheduling_repaint"] = ["RePaintScheduler"]
     _import_structure["scheduling_sde_ve"] = ["ScoreSdeVeScheduler"]
-    _import_structure["scheduling_sde_vp"] = ["ScoreSdeVpScheduler"]
     _import_structure["scheduling_unclip"] = ["UnCLIPScheduler"]
     _import_structure["scheduling_unipc_multistep"] = ["UniPCMultistepScheduler"]
     _import_structure["scheduling_utils"] = ["KarrasDiffusionSchedulers", "SchedulerMixin"]
@@ -129,6 +128,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
     except OptionalDependencyNotAvailable:
         from ..utils.dummy_pt_objects import *  # noqa F403
     else:
+        from .deprecated import KarrasVeScheduler, ScoreSdeVpScheduler
         from .scheduling_consistency_decoder import ConsistencyDecoderScheduler
         from .scheduling_consistency_models import CMStochasticIterativeScheduler
         from .scheduling_ddim import DDIMScheduler
@@ -147,12 +147,10 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
         from .scheduling_ipndm import IPNDMScheduler
         from .scheduling_k_dpm_2_ancestral_discrete import KDPM2AncestralDiscreteScheduler
         from .scheduling_k_dpm_2_discrete import KDPM2DiscreteScheduler
-        from .scheduling_karras_ve import KarrasVeScheduler
         from .scheduling_lcm import LCMScheduler
         from .scheduling_pndm import PNDMScheduler
         from .scheduling_repaint import RePaintScheduler
         from .scheduling_sde_ve import ScoreSdeVeScheduler
-        from .scheduling_sde_vp import ScoreSdeVpScheduler
         from .scheduling_unclip import UnCLIPScheduler
         from .scheduling_unipc_multistep import UniPCMultistepScheduler
         from .scheduling_utils import KarrasDiffusionSchedulers, SchedulerMixin

diffusers/schedulers/deprecated/__init__.py

@@ -0,0 +1,50 @@
+from typing import TYPE_CHECKING
+
+from ...utils import (
+    DIFFUSERS_SLOW_IMPORT,
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    get_objects_from_module,
+    is_torch_available,
+    is_transformers_available,
+)
+
+
+_dummy_objects = {}
+_import_structure = {}
+
+try:
+    if not (is_transformers_available() and is_torch_available()):
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from ...utils import dummy_pt_objects  # noqa F403
+
+    _dummy_objects.update(get_objects_from_module(dummy_pt_objects))
+else:
+    _import_structure["scheduling_karras_ve"] = ["KarrasVeScheduler"]
+    _import_structure["scheduling_sde_vp"] = ["ScoreSdeVpScheduler"]
+
+if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+
+    except OptionalDependencyNotAvailable:
+        from ..utils.dummy_pt_objects import *  # noqa F403
+    else:
+        from .scheduling_karras_ve import KarrasVeScheduler
+        from .scheduling_sde_vp import ScoreSdeVpScheduler
+
+
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(
+        __name__,
+        globals()["__file__"],
+        _import_structure,
+        module_spec=__spec__,
+    )
+
+    for name, value in _dummy_objects.items():
+        setattr(sys.modules[__name__], name, value)

diffusers/schedulers/{scheduling_karras_ve.py → deprecated/scheduling_karras_ve.py}

@@ -19,10 +19,10 @@ from typing import Optional, Tuple, Union
 import numpy as np
 import torch
 
-from ..configuration_utils import ConfigMixin, register_to_config
-from ..utils import BaseOutput
-from ..utils.torch_utils import randn_tensor
-from .scheduling_utils import SchedulerMixin
+from ...configuration_utils import ConfigMixin, register_to_config
+from ...utils import BaseOutput
+from ...utils.torch_utils import randn_tensor
+from ..scheduling_utils import SchedulerMixin
 
 
 @dataclass

diffusers/schedulers/{scheduling_sde_vp.py → deprecated/scheduling_sde_vp.py}

@@ -19,9 +19,9 @@ from typing import Union
 
 import torch
 
-from ..configuration_utils import ConfigMixin, register_to_config
-from ..utils.torch_utils import randn_tensor
-from .scheduling_utils import SchedulerMixin
+from ...configuration_utils import ConfigMixin, register_to_config
+from ...utils.torch_utils import randn_tensor
+from ..scheduling_utils import SchedulerMixin
 
 
 class ScoreSdeVpScheduler(SchedulerMixin, ConfigMixin):
@@ -79,9 +79,7 @@ class ScoreSdeVpScheduler(SchedulerMixin, ConfigMixin):
 
         # TODO(Patrick) better comments + non-PyTorch
         # postprocess model score
-        log_mean_coeff = (
-            -0.25 * t**2 * (self.config.beta_max - self.config.beta_min) - 0.5 * t * self.config.beta_min
-        )
+        log_mean_coeff = -0.25 * t**2 * (self.config.beta_max - self.config.beta_min) - 0.5 * t * self.config.beta_min
         std = torch.sqrt(1.0 - torch.exp(2.0 * log_mean_coeff))
         std = std.flatten()
         while len(std.shape) < len(score.shape):

diffusers/schedulers/scheduling_ddim.py

@@ -208,9 +208,7 @@ class DDIMScheduler(SchedulerMixin, ConfigMixin):
             self.betas = torch.linspace(beta_start, beta_end, num_train_timesteps, dtype=torch.float32)
         elif beta_schedule == "scaled_linear":
             # this schedule is very specific to the latent diffusion model.
-            self.betas = (
-                torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
-            )
+            self.betas = torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
         elif beta_schedule == "squaredcos_cap_v2":
             # Glide cosine schedule
             self.betas = betas_for_alpha_bar(num_train_timesteps)

diffusers/schedulers/scheduling_ddim_inverse.py

@@ -204,9 +204,7 @@ class DDIMInverseScheduler(SchedulerMixin, ConfigMixin):
             self.betas = torch.linspace(beta_start, beta_end, num_train_timesteps, dtype=torch.float32)
         elif beta_schedule == "scaled_linear":
             # this schedule is very specific to the latent diffusion model.
-            self.betas = (
-                torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
-            )
+            self.betas = torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
         elif beta_schedule == "squaredcos_cap_v2":
             # Glide cosine schedule
             self.betas = betas_for_alpha_bar(num_train_timesteps)

diffusers/schedulers/scheduling_ddim_parallel.py

@@ -215,9 +215,7 @@ class DDIMParallelScheduler(SchedulerMixin, ConfigMixin):
             self.betas = torch.linspace(beta_start, beta_end, num_train_timesteps, dtype=torch.float32)
         elif beta_schedule == "scaled_linear":
             # this schedule is very specific to the latent diffusion model.
-            self.betas = (
-                torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
-            )
+            self.betas = torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
         elif beta_schedule == "squaredcos_cap_v2":
             # Glide cosine schedule
             self.betas = betas_for_alpha_bar(num_train_timesteps)

diffusers/schedulers/scheduling_ddpm.py

@@ -160,9 +160,7 @@ class DDPMScheduler(SchedulerMixin, ConfigMixin):
             self.betas = torch.linspace(beta_start, beta_end, num_train_timesteps, dtype=torch.float32)
         elif beta_schedule == "scaled_linear":
             # this schedule is very specific to the latent diffusion model.
-            self.betas = (
-                torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
-            )
+            self.betas = torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
         elif beta_schedule == "squaredcos_cap_v2":
             # Glide cosine schedule
             self.betas = betas_for_alpha_bar(num_train_timesteps)

diffusers/schedulers/scheduling_ddpm_parallel.py

@@ -170,9 +170,7 @@ class DDPMParallelScheduler(SchedulerMixin, ConfigMixin):
             self.betas = torch.linspace(beta_start, beta_end, num_train_timesteps, dtype=torch.float32)
         elif beta_schedule == "scaled_linear":
             # this schedule is very specific to the latent diffusion model.
-            self.betas = (
-                torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
-            )
+            self.betas = torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
         elif beta_schedule == "squaredcos_cap_v2":
             # Glide cosine schedule
             self.betas = betas_for_alpha_bar(num_train_timesteps)

diffusers/schedulers/scheduling_deis_multistep.py

@@ -149,9 +149,7 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
             self.betas = torch.linspace(beta_start, beta_end, num_train_timesteps, dtype=torch.float32)
         elif beta_schedule == "scaled_linear":
             # this schedule is very specific to the latent diffusion model.
-            self.betas = (
-                torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
-            )
+            self.betas = torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
         elif beta_schedule == "squaredcos_cap_v2":
             # Glide cosine schedule
             self.betas = betas_for_alpha_bar(num_train_timesteps)
@@ -325,8 +323,20 @@ class DEISMultistepScheduler(SchedulerMixin, ConfigMixin):
     def _convert_to_karras(self, in_sigmas: torch.FloatTensor, num_inference_steps) -> torch.FloatTensor:
         """Constructs the noise schedule of Karras et al. (2022)."""
 
-        sigma_min: float = in_sigmas[-1].item()
-        sigma_max: float = in_sigmas[0].item()
+        # Hack to make sure that other schedulers which copy this function don't break
+        # TODO: Add this logic to the other schedulers
+        if hasattr(self.config, "sigma_min"):
+            sigma_min = self.config.sigma_min
+        else:
+            sigma_min = None
+
+        if hasattr(self.config, "sigma_max"):
+            sigma_max = self.config.sigma_max
+        else:
+            sigma_max = None
+
+        sigma_min = sigma_min if sigma_min is not None else in_sigmas[-1].item()
+        sigma_max = sigma_max if sigma_max is not None else in_sigmas[0].item()
 
         rho = 7.0  # 7.0 is the value used in the paper
         ramp = np.linspace(0, 1, num_inference_steps)

diffusers/schedulers/scheduling_dpmsolver_multistep.py

@@ -176,9 +176,7 @@ class DPMSolverMultistepScheduler(SchedulerMixin, ConfigMixin):
             self.betas = torch.linspace(beta_start, beta_end, num_train_timesteps, dtype=torch.float32)
         elif beta_schedule == "scaled_linear":
             # this schedule is very specific to the latent diffusion model.
-            self.betas = (
-                torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
-            )
+            self.betas = torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
         elif beta_schedule == "squaredcos_cap_v2":
             # Glide cosine schedule
             self.betas = betas_for_alpha_bar(num_train_timesteps)
@@ -360,8 +358,20 @@ class DPMSolverMultistepScheduler(SchedulerMixin, ConfigMixin):
     def _convert_to_karras(self, in_sigmas: torch.FloatTensor, num_inference_steps) -> torch.FloatTensor:
         """Constructs the noise schedule of Karras et al. (2022)."""
 
-        sigma_min: float = in_sigmas[-1].item()
-        sigma_max: float = in_sigmas[0].item()
+        # Hack to make sure that other schedulers which copy this function don't break
+        # TODO: Add this logic to the other schedulers
+        if hasattr(self.config, "sigma_min"):
+            sigma_min = self.config.sigma_min
+        else:
+            sigma_min = None
+
+        if hasattr(self.config, "sigma_max"):
+            sigma_max = self.config.sigma_max
+        else:
+            sigma_max = None
+
+        sigma_min = sigma_min if sigma_min is not None else in_sigmas[-1].item()
+        sigma_max = sigma_max if sigma_max is not None else in_sigmas[0].item()
 
         rho = 7.0  # 7.0 is the value used in the paper
         ramp = np.linspace(0, 1, num_inference_steps)

diffusers/schedulers/scheduling_dpmsolver_multistep_inverse.py

@@ -171,9 +171,7 @@ class DPMSolverMultistepInverseScheduler(SchedulerMixin, ConfigMixin):
             self.betas = torch.linspace(beta_start, beta_end, num_train_timesteps, dtype=torch.float32)
         elif beta_schedule == "scaled_linear":
             # this schedule is very specific to the latent diffusion model.
-            self.betas = (
-                torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
-            )
+            self.betas = torch.linspace(beta_start**0.5, beta_end**0.5, num_train_timesteps, dtype=torch.float32) ** 2
         elif beta_schedule == "squaredcos_cap_v2":
             # Glide cosine schedule
             self.betas = betas_for_alpha_bar(num_train_timesteps)
@@ -360,8 +358,20 @@ class DPMSolverMultistepInverseScheduler(SchedulerMixin, ConfigMixin):
     def _convert_to_karras(self, in_sigmas: torch.FloatTensor, num_inference_steps) -> torch.FloatTensor:
         """Constructs the noise schedule of Karras et al. (2022)."""
 
-        sigma_min: float = in_sigmas[-1].item()
-        sigma_max: float = in_sigmas[0].item()
+        # Hack to make sure that other schedulers which copy this function don't break
+        # TODO: Add this logic to the other schedulers
+        if hasattr(self.config, "sigma_min"):
+            sigma_min = self.config.sigma_min
+        else:
+            sigma_min = None
+
+        if hasattr(self.config, "sigma_max"):
+            sigma_max = self.config.sigma_max
+        else:
+            sigma_max = None
+
+        sigma_min = sigma_min if sigma_min is not None else in_sigmas[-1].item()
+        sigma_max = sigma_max if sigma_max is not None else in_sigmas[0].item()
 
         rho = 7.0  # 7.0 is the value used in the paper
         ramp = np.linspace(0, 1, num_inference_steps)