start folderizing the loaders.

sayakpaul · sayakpaul · commit 8267677a241f · 2025-04-16T12:02:06.000+05:30
diff --git a/src/diffusers/loaders/__init__.py b/src/diffusers/loaders/__init__.py
@@ -54,7 +54,7 @@ def text_encoder_attn_modules(text_encoder):
 _import_structure = {}
 
 if is_torch_available():
-    _import_structure["single_file_model"] = ["FromOriginalModelMixin"]
+    _import_structure["single_file.single_file_model"] = ["FromOriginalModelMixin"]
     _import_structure["transformer_flux"] = ["FluxTransformer2DLoadersMixin"]
     _import_structure["transformer_sd3"] = ["SD3Transformer2DLoadersMixin"]
     _import_structure["unet"] = ["UNet2DConditionLoadersMixin"]
@@ -77,6 +77,7 @@ def text_encoder_attn_modules(text_encoder):
             "SanaLoraLoaderMixin",
             "Lumina2LoraLoaderMixin",
             "WanLoraLoaderMixin",
+            "LoraBaseMixin",
         ]
         _import_structure["textual_inversion"] = ["TextualInversionLoaderMixin"]
         _import_structure["ip_adapter"] = [
@@ -90,25 +91,21 @@ def text_encoder_attn_modules(text_encoder):
 
 if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
     if is_torch_available():
-        from .single_file_model import FromOriginalModelMixin
-        from .transformer_flux import FluxTransformer2DLoadersMixin
-        from .transformer_sd3 import SD3Transformer2DLoadersMixin
+        from .ip_adapter import FluxTransformer2DLoadersMixin, SD3Transformer2DLoadersMixin
+        from .single_file import FromOriginalModelMixin
         from .unet import UNet2DConditionLoadersMixin
         from .utils import AttnProcsLayers
 
         if is_transformers_available():
-            from .ip_adapter import (
-                FluxIPAdapterMixin,
-                IPAdapterMixin,
-                SD3IPAdapterMixin,
-            )
-            from .lora_pipeline import (
+            from .ip_adapter import FluxIPAdapterMixin, IPAdapterMixin, SD3IPAdapterMixin
+            from .lora import (
                 AmusedLoraLoaderMixin,
                 AuraFlowLoraLoaderMixin,
                 CogVideoXLoraLoaderMixin,
                 CogView4LoraLoaderMixin,
                 FluxLoraLoaderMixin,
                 HunyuanVideoLoraLoaderMixin,
+                LoraBaseMixin,
                 LoraLoaderMixin,
                 LTXVideoLoraLoaderMixin,
                 Lumina2LoraLoaderMixin,
diff --git a/src/diffusers/loaders/ip_adapter/__init__.py b/src/diffusers/loaders/ip_adapter/__init__.py
@@ -0,0 +1,9 @@
+from ...utils.import_utils import is_torch_available, is_transformers_available
+
+
+if is_torch_available():
+    from .transformer_flux import FluxTransformer2DLoadersMixin
+    from .transformer_sd3 import SD3Transformer2DLoadersMixin
+
+    if is_transformers_available():
+        from .ip_adapter import FluxIPAdapterMixin, IPAdapterMixin, SD3IPAdapterMixin
diff --git a/src/diffusers/loaders/ip_adapter/ip_adapter.py b/src/diffusers/loaders/ip_adapter/ip_adapter.py
diff --git a/src/diffusers/loaders/ip_adapter/transformer_flux.py b/src/diffusers/loaders/ip_adapter/transformer_flux.py
@@ -13,21 +13,11 @@
 # limitations under the License.
 from contextlib import nullcontext
 
-from ..models.embeddings import (
-    ImageProjection,
-    MultiIPAdapterImageProjection,
-)
-from ..models.modeling_utils import _LOW_CPU_MEM_USAGE_DEFAULT, load_model_dict_into_meta
-from ..utils import (
-    is_accelerate_available,
-    is_torch_version,
-    logging,
-)
+from ...models.embeddings import ImageProjection, MultiIPAdapterImageProjection
+from ...models.modeling_utils import _LOW_CPU_MEM_USAGE_DEFAULT, load_model_dict_into_meta
+from ...utils import is_accelerate_available, is_torch_version, logging
 
 
-if is_accelerate_available():
-    pass
-
 logger = logging.get_logger(__name__)
 
 
@@ -88,9 +78,7 @@ def _convert_ip_adapter_image_proj_to_diffusers(self, state_dict, low_cpu_mem_us
         return image_projection
 
     def _convert_ip_adapter_attn_to_diffusers(self, state_dicts, low_cpu_mem_usage=_LOW_CPU_MEM_USAGE_DEFAULT):
-        from ..models.attention_processor import (
-            FluxIPAdapterJointAttnProcessor2_0,
-        )
+        from ...models.attention_processor import FluxIPAdapterJointAttnProcessor2_0
 
         if low_cpu_mem_usage:
             if is_accelerate_available():
diff --git a/src/diffusers/loaders/ip_adapter/transformer_sd3.py b/src/diffusers/loaders/ip_adapter/transformer_sd3.py
@@ -14,10 +14,10 @@
 from contextlib import nullcontext
 from typing import Dict
 
-from ..models.attention_processor import SD3IPAdapterJointAttnProcessor2_0
-from ..models.embeddings import IPAdapterTimeImageProjection
-from ..models.modeling_utils import _LOW_CPU_MEM_USAGE_DEFAULT, load_model_dict_into_meta
-from ..utils import is_accelerate_available, is_torch_version, logging
+from ...models.attention_processor import SD3IPAdapterJointAttnProcessor2_0
+from ...models.embeddings import IPAdapterTimeImageProjection
+from ...models.modeling_utils import _LOW_CPU_MEM_USAGE_DEFAULT, load_model_dict_into_meta
+from ...utils import is_accelerate_available, is_torch_version, logging
 
 
 logger = logging.get_logger(__name__)
diff --git a/src/diffusers/loaders/lora/__init__.py b/src/diffusers/loaders/lora/__init__.py
@@ -0,0 +1,24 @@
+from ...utils import is_peft_available, is_torch_available, is_transformers_available
+
+
+if is_torch_available():
+    from .lora_base import LoraBaseMixin
+
+    if is_transformers_available():
+        from .lora_pipeline import (
+            AmusedLoraLoaderMixin,
+            AuraFlowLoraLoaderMixin,
+            CogVideoXLoraLoaderMixin,
+            CogView4LoraLoaderMixin,
+            FluxLoraLoaderMixin,
+            HunyuanVideoLoraLoaderMixin,
+            LoraLoaderMixin,
+            LTXVideoLoraLoaderMixin,
+            Lumina2LoraLoaderMixin,
+            Mochi1LoraLoaderMixin,
+            SanaLoraLoaderMixin,
+            SD3LoraLoaderMixin,
+            StableDiffusionLoraLoaderMixin,
+            StableDiffusionXLLoraLoaderMixin,
+            WanLoraLoaderMixin,
+        )
diff --git a/src/diffusers/loaders/lora/lora_base.py b/src/diffusers/loaders/lora/lora_base.py
diff --git a/src/diffusers/loaders/lora/lora_conversion_utils.py b/src/diffusers/loaders/lora/lora_conversion_utils.py
diff --git a/src/diffusers/loaders/lora/lora_pipeline.py b/src/diffusers/loaders/lora/lora_pipeline.py
diff --git a/src/diffusers/loaders/single_file/__init__.py b/src/diffusers/loaders/single_file/__init__.py
@@ -0,0 +1,8 @@
+from ...utils import is_torch_available, is_transformers_available
+
+
+if is_torch_available():
+    from .single_file_model import FromOriginalModelMixin
+
+    if is_transformers_available():
+        from .single_file import FromSingleFileMixin
diff --git a/src/diffusers/loaders/single_file/single_file.py b/src/diffusers/loaders/single_file/single_file.py
@@ -21,7 +21,7 @@
 from packaging import version
 from typing_extensions import Self
 
-from ..utils import deprecate, is_transformers_available, logging
+from ...utils import deprecate, is_transformers_available, logging
 from .single_file_utils import (
     SingleFileComponentError,
     _is_legacy_scheduler_kwargs,
diff --git a/src/diffusers/loaders/single_file/single_file_model.py b/src/diffusers/loaders/single_file/single_file_model.py
@@ -21,9 +21,9 @@
 from huggingface_hub.utils import validate_hf_hub_args
 from typing_extensions import Self
 
-from .. import __version__
-from ..quantizers import DiffusersAutoQuantizer
-from ..utils import deprecate, is_accelerate_available, logging
+from ... import __version__
+from ...quantizers import DiffusersAutoQuantizer
+from ...utils import deprecate, is_accelerate_available, logging
 from .single_file_utils import (
     SingleFileComponentError,
     convert_animatediff_checkpoint_to_diffusers,
@@ -58,7 +58,7 @@
 if is_accelerate_available():
     from accelerate import dispatch_model, init_empty_weights
 
-    from ..models.modeling_utils import load_model_dict_into_meta
+    from ...models.modeling_utils import load_model_dict_into_meta
 
 
 SINGLE_FILE_LOADABLE_CLASSES = {
diff --git a/src/diffusers/loaders/single_file/single_file_utils.py b/src/diffusers/loaders/single_file/single_file_utils.py
@@ -25,8 +25,8 @@
 import torch
 import yaml
 
-from ..models.modeling_utils import load_state_dict
-from ..schedulers import (
+from ...models.modeling_utils import load_state_dict
+from ...schedulers import (
     DDIMScheduler,
     DPMSolverMultistepScheduler,
     EDMDPMSolverMultistepScheduler,
@@ -54,7 +54,7 @@
 if is_accelerate_available():
     from accelerate import init_empty_weights
 
-    from ..models.modeling_utils import load_model_dict_into_meta
+    from ...models.modeling_utils import load_model_dict_into_meta
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
diff --git a/src/diffusers/loaders/unet/__init__.py b/src/diffusers/loaders/unet/__init__.py
@@ -0,0 +1,5 @@
+from ...utils import is_torch_available
+
+
+if is_torch_available():
+    from .unet import UNet2DConditionLoadersMixin
diff --git a/src/diffusers/loaders/unet/unet.py b/src/diffusers/loaders/unet/unet.py
@@ -22,16 +22,16 @@
 import torch.nn.functional as F
 from huggingface_hub.utils import validate_hf_hub_args
 
-from ..models.embeddings import (
+from ...models.embeddings import (
     ImageProjection,
     IPAdapterFaceIDImageProjection,
     IPAdapterFaceIDPlusImageProjection,
     IPAdapterFullImageProjection,
     IPAdapterPlusImageProjection,
     MultiIPAdapterImageProjection,
 )
-from ..models.modeling_utils import _LOW_CPU_MEM_USAGE_DEFAULT, load_model_dict_into_meta, load_state_dict
-from ..utils import (
+from ...models.modeling_utils import _LOW_CPU_MEM_USAGE_DEFAULT, load_model_dict_into_meta, load_state_dict
+from ...utils import (
     USE_PEFT_BACKEND,
     _get_model_file,
     convert_unet_state_dict_to_peft,
@@ -43,9 +43,9 @@
     is_torch_version,
     logging,
 )
-from .lora_base import _func_optionally_disable_offloading
-from .lora_pipeline import LORA_WEIGHT_NAME, LORA_WEIGHT_NAME_SAFE, TEXT_ENCODER_NAME, UNET_NAME
-from .utils import AttnProcsLayers
+from ..lora import _func_optionally_disable_offloading
+from ..lora.lora_pipeline import LORA_WEIGHT_NAME, LORA_WEIGHT_NAME_SAFE, TEXT_ENCODER_NAME, UNET_NAME
+from ..utils import AttnProcsLayers
 
 
 logger = logging.get_logger(__name__)
@@ -247,7 +247,7 @@ def load_attn_procs(self, pretrained_model_name_or_path_or_dict: Union[str, Dict
         # Unsafe code />
 
     def _process_custom_diffusion(self, state_dict):
-        from ..models.attention_processor import CustomDiffusionAttnProcessor
+        from ...models.attention_processor import CustomDiffusionAttnProcessor
 
         attn_processors = {}
         custom_diffusion_grouped_dict = defaultdict(dict)
@@ -451,7 +451,7 @@ def save_attn_procs(
         pipeline.unet.save_attn_procs("path-to-save-model", weight_name="pytorch_custom_diffusion_weights.bin")
         ```
         """
-        from ..models.attention_processor import (
+        from ...models.attention_processor import (
             CustomDiffusionAttnProcessor,
             CustomDiffusionAttnProcessor2_0,
             CustomDiffusionXFormersAttnProcessor,
@@ -513,7 +513,7 @@ def save_function(weights, filename):
         logger.info(f"Model weights saved in {save_path}")
 
     def _get_custom_diffusion_state_dict(self):
-        from ..models.attention_processor import (
+        from ...models.attention_processor import (
             CustomDiffusionAttnProcessor,
             CustomDiffusionAttnProcessor2_0,
             CustomDiffusionXFormersAttnProcessor,
@@ -759,7 +759,7 @@ def _convert_ip_adapter_image_proj_to_diffusers(self, state_dict, low_cpu_mem_us
         return image_projection
 
     def _convert_ip_adapter_attn_to_diffusers(self, state_dicts, low_cpu_mem_usage=_LOW_CPU_MEM_USAGE_DEFAULT):
-        from ..models.attention_processor import (
+        from ...models.attention_processor import (
             IPAdapterAttnProcessor,
             IPAdapterAttnProcessor2_0,
             IPAdapterXFormersAttnProcessor,
diff --git a/src/diffusers/loaders/unet/unet_loader_utils.py b/src/diffusers/loaders/unet/unet_loader_utils.py
