run lint rules using latest isort settings

scripts/conversion/convert_diffusers_autoencoder_kl.py

@@ -1,10 +1,12 @@
 import argparse
 from pathlib import Path
+
 import torch
-from torch import nn
 from diffusers import AutoencoderKL  # type: ignore
-from refiners.foundationals.latent_diffusion.auto_encoder import LatentDiffusionAutoencoder
+from torch import nn
+
 from refiners.fluxion.model_converter import ModelConverter
+from refiners.foundationals.latent_diffusion.auto_encoder import LatentDiffusionAutoencoder
 
 
 class Args(argparse.Namespace):

scripts/conversion/convert_diffusers_controlnet.py

@@ -1,15 +1,17 @@
 # pyright: reportPrivateUsage=false
 import argparse
 from pathlib import Path
+
 import torch
-from torch import nn
 from diffusers import ControlNetModel  # type: ignore
-from refiners.fluxion.utils import save_to_safetensors
+from torch import nn
+
 from refiners.fluxion.model_converter import ModelConverter
+from refiners.fluxion.utils import save_to_safetensors
 from refiners.foundationals.latent_diffusion import (
-    SD1UNet,
-    SD1ControlnetAdapter,
     DPMSolver,
+    SD1ControlnetAdapter,
+    SD1UNet,
 )
 
 

scripts/conversion/convert_diffusers_ip_adapter.py

@@ -1,11 +1,11 @@
+import argparse
 from pathlib import Path
 from typing import Any
-import argparse
 
 import torch
 
-from refiners.foundationals.latent_diffusion import SD1UNet, SD1IPAdapter, SDXLUNet, SDXLIPAdapter
 from refiners.fluxion.utils import save_to_safetensors
+from refiners.foundationals.latent_diffusion import SD1IPAdapter, SD1UNet, SDXLIPAdapter, SDXLUNet
 
 # Running:
 #

scripts/conversion/convert_diffusers_lora.py

@@ -3,16 +3,15 @@
 from typing import cast
 
 import torch
+from diffusers import DiffusionPipeline  # type: ignore
 from torch import Tensor
-from torch.nn.init import zeros_
 from torch.nn import Parameter as TorchParameter
-
-from diffusers import DiffusionPipeline  # type: ignore
+from torch.nn.init import zeros_
 
 import refiners.fluxion.layers as fl
+from refiners.fluxion.adapters.lora import Lora, LoraAdapter
 from refiners.fluxion.model_converter import ModelConverter
 from refiners.fluxion.utils import save_to_safetensors
-from refiners.fluxion.adapters.lora import Lora, LoraAdapter
 from refiners.foundationals.latent_diffusion import SD1UNet
 from refiners.foundationals.latent_diffusion.lora import LoraTarget, lora_targets
 

scripts/conversion/convert_diffusers_t2i_adapter.py

@@ -1,10 +1,12 @@
 import argparse
 from pathlib import Path
+
 import torch
-from torch import nn
 from diffusers import T2IAdapter  # type: ignore
-from refiners.foundationals.latent_diffusion.t2i_adapter import ConditionEncoder, ConditionEncoderXL
+from torch import nn
+
 from refiners.fluxion.model_converter import ModelConverter
+from refiners.foundationals.latent_diffusion.t2i_adapter import ConditionEncoder, ConditionEncoderXL
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="Convert a pretrained diffusers T2I-Adapter model to refiners")

scripts/conversion/convert_diffusers_unet.py

@@ -1,9 +1,11 @@
 import argparse
 from pathlib import Path
+
 import torch
+from diffusers import UNet2DConditionModel  # type: ignore
 from torch import nn
+
 from refiners.fluxion.model_converter import ModelConverter
-from diffusers import UNet2DConditionModel  # type: ignore
 from refiners.foundationals.latent_diffusion import SD1UNet, SDXLUNet
 
 

scripts/conversion/convert_informative_drawings.py

@@ -1,7 +1,9 @@
 import argparse
 from typing import TYPE_CHECKING, cast
+
 import torch
 from torch import nn
+
 from refiners.fluxion.model_converter import ModelConverter
 from refiners.foundationals.latent_diffusion.preprocessors.informative_drawings import InformativeDrawings
 

scripts/conversion/convert_refiners_lora_to_sdwebui.py

@@ -1,20 +1,22 @@
 import argparse
 from functools import partial
+
+from convert_diffusers_unet import Args as UnetConversionArgs, setup_converter as convert_unet
+from convert_transformers_clip_text_model import (
+    Args as TextEncoderConversionArgs,
+    setup_converter as convert_text_encoder,
+)
 from torch import Tensor
+
+import refiners.fluxion.layers as fl
 from refiners.fluxion.utils import (
     load_from_safetensors,
     load_metadata_from_safetensors,
     save_to_safetensors,
 )
-from convert_diffusers_unet import setup_converter as convert_unet, Args as UnetConversionArgs
-from convert_transformers_clip_text_model import (
-    setup_converter as convert_text_encoder,
-    Args as TextEncoderConversionArgs,
-)
 from refiners.foundationals.clip.text_encoder import CLIPTextEncoderL
 from refiners.foundationals.latent_diffusion import SD1UNet
 from refiners.foundationals.latent_diffusion.lora import LoraTarget
-import refiners.fluxion.layers as fl
 
 
 def get_unet_mapping(source_path: str) -> dict[str, str]:

scripts/conversion/convert_segment_anything.py

@@ -1,20 +1,19 @@
 import argparse
 import types
 from typing import Any, Callable, cast
+
 import torch
 import torch.nn as nn
+from segment_anything import build_sam_vit_h  # type: ignore
+from segment_anything.modeling.common import LayerNorm2d  # type: ignore
 from torch import Tensor
 
 import refiners.fluxion.layers as fl
 from refiners.fluxion.model_converter import ModelConverter
 from refiners.fluxion.utils import manual_seed, save_to_safetensors
 from refiners.foundationals.segment_anything.image_encoder import SAMViTH
-from refiners.foundationals.segment_anything.prompt_encoder import PointEncoder, MaskEncoder
-
-from segment_anything import build_sam_vit_h  # type: ignore
-from segment_anything.modeling.common import LayerNorm2d  # type: ignore
-
 from refiners.foundationals.segment_anything.mask_decoder import MaskDecoder
+from refiners.foundationals.segment_anything.prompt_encoder import MaskEncoder, PointEncoder
 
 
 class FacebookSAM(nn.Module):
@@ -134,9 +133,10 @@ def convert_mask_decoder(mask_decoder: nn.Module) -> dict[str, Tensor]:
     point_embedding = torch.randn(1, 3, 256)
     mask_embedding = torch.randn(1, 256, 64, 64)
 
-    import refiners.fluxion.layers as fl
     from segment_anything.modeling.common import LayerNorm2d  # type: ignore
 
+    import refiners.fluxion.layers as fl
+
     assert issubclass(LayerNorm2d, nn.Module)
     custom_layers = {LayerNorm2d: fl.LayerNorm2d}
 

scripts/conversion/convert_transformers_clip_image_model.py

@@ -1,12 +1,14 @@
 import argparse
 from pathlib import Path
+
+import torch
 from torch import nn
-from refiners.fluxion.model_converter import ModelConverter
 from transformers import CLIPVisionModelWithProjection  # type: ignore
-from refiners.foundationals.clip.image_encoder import CLIPImageEncoder
-from refiners.fluxion.utils import save_to_safetensors
-import torch
+
 import refiners.fluxion.layers as fl
+from refiners.fluxion.model_converter import ModelConverter
+from refiners.fluxion.utils import save_to_safetensors
+from refiners.foundationals.clip.image_encoder import CLIPImageEncoder
 
 
 class Args(argparse.Namespace):

scripts/conversion/convert_transformers_clip_text_model.py

@@ -1,14 +1,16 @@
 import argparse
 from pathlib import Path
 from typing import cast
+
 from torch import nn
-from refiners.fluxion.model_converter import ModelConverter
 from transformers import CLIPTextModelWithProjection  # type: ignore
-from refiners.foundationals.clip.text_encoder import CLIPTextEncoder, CLIPTextEncoderL, CLIPTextEncoderG
+
+import refiners.fluxion.layers as fl
+from refiners.fluxion.model_converter import ModelConverter
+from refiners.fluxion.utils import save_to_safetensors
+from refiners.foundationals.clip.text_encoder import CLIPTextEncoder, CLIPTextEncoderG, CLIPTextEncoderL
 from refiners.foundationals.clip.tokenizer import CLIPTokenizer
 from refiners.foundationals.latent_diffusion.stable_diffusion_xl.text_encoder import DoubleTextEncoder
-from refiners.fluxion.utils import save_to_safetensors
-import refiners.fluxion.layers as fl
 
 
 class Args(argparse.Namespace):

scripts/training/finetune-ldm-lora.py

@@ -1,20 +1,21 @@
 import random
 from typing import Any
-from pydantic import BaseModel
+
 from loguru import logger
-from refiners.fluxion.utils import save_to_safetensors
-from refiners.foundationals.latent_diffusion.lora import LoraTarget, LoraAdapter, MODELS, lora_targets
-import refiners.fluxion.layers as fl
+from pydantic import BaseModel
 from torch import Tensor
 from torch.utils.data import Dataset
 
+import refiners.fluxion.layers as fl
+from refiners.fluxion.utils import save_to_safetensors
+from refiners.foundationals.latent_diffusion.lora import MODELS, LoraAdapter, LoraTarget, lora_targets
 from refiners.training_utils.callback import Callback
 from refiners.training_utils.latent_diffusion import (
     FinetuneLatentDiffusionConfig,
+    LatentDiffusionConfig,
+    LatentDiffusionTrainer,
     TextEmbeddingLatentsBatch,
     TextEmbeddingLatentsDataset,
-    LatentDiffusionTrainer,
-    LatentDiffusionConfig,
 )
 
 

scripts/training/finetune-ldm-textual-inversion.py

@@ -1,24 +1,24 @@
+import random
 from typing import Any
-from pydantic import BaseModel
+
 from loguru import logger
+from pydantic import BaseModel
+from torch import Tensor, randn
 from torch.utils.data import Dataset
-from torch import randn, Tensor
-import random
 
+from refiners.fluxion.utils import save_to_safetensors
 from refiners.foundationals.clip.concepts import ConceptExtender, EmbeddingExtender
 from refiners.foundationals.clip.text_encoder import CLIPTextEncoder, TokenEncoder
 from refiners.foundationals.clip.tokenizer import CLIPTokenizer
-from refiners.fluxion.utils import save_to_safetensors
 from refiners.training_utils.callback import Callback
 from refiners.training_utils.latent_diffusion import (
     FinetuneLatentDiffusionConfig,
-    TextEmbeddingLatentsBatch,
-    LatentDiffusionTrainer,
     LatentDiffusionConfig,
+    LatentDiffusionTrainer,
+    TextEmbeddingLatentsBatch,
     TextEmbeddingLatentsDataset,
 )
 
-
 IMAGENET_TEMPLATES_SMALL = [
     "a photo of a {}",
     "a rendering of a {}",

src/refiners/fluxion/__init__.py

@@ -1,3 +1,3 @@
-from refiners.fluxion.utils import save_to_safetensors, load_from_safetensors, norm, manual_seed, pad
+from refiners.fluxion.utils import load_from_safetensors, manual_seed, norm, pad, save_to_safetensors
 
 __all__ = ["norm", "manual_seed", "save_to_safetensors", "load_from_safetensors", "pad"]

src/refiners/fluxion/adapters/adapter.py

@@ -1,7 +1,7 @@
 import contextlib
-import refiners.fluxion.layers as fl
-from typing import Any, Generic, TypeVar, Iterator
+from typing import Any, Generic, Iterator, TypeVar
 
+import refiners.fluxion.layers as fl
 
 T = TypeVar("T", bound=fl.Module)
 TAdapter = TypeVar("TAdapter", bound="Adapter[Any]")  # Self (see PEP 673)

src/refiners/fluxion/adapters/lora.py

@@ -1,11 +1,11 @@
-from typing import Iterable, Generic, TypeVar, Any
-
-import refiners.fluxion.layers as fl
-from refiners.fluxion.adapters.adapter import Adapter
+from typing import Any, Generic, Iterable, TypeVar
 
 from torch import Tensor, device as Device, dtype as DType
 from torch.nn import Parameter as TorchParameter
-from torch.nn.init import zeros_, normal_
+from torch.nn.init import normal_, zeros_
+
+import refiners.fluxion.layers as fl
+from refiners.fluxion.adapters.adapter import Adapter
 
 T = TypeVar("T", bound=fl.Chain)
 TLoraAdapter = TypeVar("TLoraAdapter", bound="LoraAdapter[Any]")  # Self (see PEP 673)

src/refiners/fluxion/context.py

@@ -1,4 +1,5 @@
 from typing import Any
+
 from torch import Tensor
 
 Context = dict[str, Any]

src/refiners/fluxion/layers/__init__.py

@@ -1,50 +1,50 @@
-from refiners.fluxion.layers.activations import GLU, SiLU, ReLU, ApproximateGeLU, GeLU, Sigmoid
-from refiners.fluxion.layers.norm import LayerNorm, GroupNorm, LayerNorm2d, InstanceNorm2d
+from refiners.fluxion.layers.activations import GLU, ApproximateGeLU, GeLU, ReLU, Sigmoid, SiLU
 from refiners.fluxion.layers.attentions import Attention, SelfAttention, SelfAttention2d
 from refiners.fluxion.layers.basics import (
-    Identity,
-    View,
+    Buffer,
+    Chunk,
+    Cos,
     Flatten,
-    Unflatten,
-    Transpose,
     GetArg,
+    Identity,
+    Multiply,
+    Parameter,
     Permute,
     Reshape,
-    Squeeze,
-    Unsqueeze,
-    Slicing,
     Sin,
-    Cos,
-    Chunk,
-    Multiply,
+    Slicing,
+    Squeeze,
+    Transpose,
     Unbind,
-    Parameter,
-    Buffer,
+    Unflatten,
+    Unsqueeze,
+    View,
 )
 from refiners.fluxion.layers.chain import (
+    Breakpoint,
+    Chain,
+    Concatenate,
+    Distribute,
     Lambda,
-    Sum,
+    Matmul,
+    Parallel,
+    Passthrough,
     Residual,
     Return,
-    Chain,
-    UseContext,
     SetContext,
-    Parallel,
-    Distribute,
-    Passthrough,
-    Breakpoint,
-    Concatenate,
-    Matmul,
+    Sum,
+    UseContext,
 )
 from refiners.fluxion.layers.conv import Conv2d, ConvTranspose2d
+from refiners.fluxion.layers.converter import Converter
+from refiners.fluxion.layers.embedding import Embedding
 from refiners.fluxion.layers.linear import Linear, MultiLinear
-from refiners.fluxion.layers.module import Module, WeightedModule, ContextModule
+from refiners.fluxion.layers.maxpool import MaxPool1d, MaxPool2d
+from refiners.fluxion.layers.module import ContextModule, Module, WeightedModule
+from refiners.fluxion.layers.norm import GroupNorm, InstanceNorm2d, LayerNorm, LayerNorm2d
 from refiners.fluxion.layers.padding import ReflectionPad2d
 from refiners.fluxion.layers.pixelshuffle import PixelUnshuffle
-from refiners.fluxion.layers.sampling import Downsample, Upsample, Interpolate
-from refiners.fluxion.layers.embedding import Embedding
-from refiners.fluxion.layers.converter import Converter
-from refiners.fluxion.layers.maxpool import MaxPool1d, MaxPool2d
+from refiners.fluxion.layers.sampling import Downsample, Interpolate, Upsample
 
 __all__ = [
     "Embedding",