diff --git a/bitsandbytes/__init__.py b/bitsandbytes/__init__.py
index 6d1177f..041df4b 100644
--- a/bitsandbytes/__init__.py
+++ b/bitsandbytes/__init__.py
@@ -3,6 +3,7 @@
 # This source code is licensed under the MIT license found in the
 # LICENSE file in the root directory of this source tree.
 
+from . import cuda_setup, utils
 from .autograd._functions import (
     MatmulLtState,
     bmm_cublas,
@@ -12,7 +13,6 @@ from .autograd._functions import (
 )
 from .cextension import COMPILED_WITH_CUDA
 from .nn import modules
-from . import cuda_setup, utils
 
 if COMPILED_WITH_CUDA:
     from .optim import adam
diff --git a/bitsandbytes/__main__.py b/bitsandbytes/__main__.py
index f1b7670..ac7948b 100644
--- a/bitsandbytes/__main__.py
+++ b/bitsandbytes/__main__.py
@@ -28,8 +28,8 @@ print()
 
 
 from . import COMPILED_WITH_CUDA, PACKAGE_GITHUB_URL
-from .cuda_setup.main import get_compute_capabilities, get_cuda_lib_handle
 from .cuda_setup.env_vars import to_be_ignored
+from .cuda_setup.main import get_compute_capabilities, get_cuda_lib_handle
 
 print_header("POTENTIALLY LIBRARY-PATH-LIKE ENV VARS")
 for k, v in os.environ.items():
diff --git a/bitsandbytes/autograd/_functions.py b/bitsandbytes/autograd/_functions.py
index cc57061..c3b0ac6 100644
--- a/bitsandbytes/autograd/_functions.py
+++ b/bitsandbytes/autograd/_functions.py
@@ -1,12 +1,13 @@
 import operator
 import warnings
-
-import torch
-import bitsandbytes.functional as F
-
 from dataclasses import dataclass
 from functools import reduce  # Required in Python 3
 
+import torch
+
+import bitsandbytes.functional as F
+
+
 # math.prod not compatible with python < 3.8
 def prod(iterable):
     return reduce(operator.mul, iterable, 1)
diff --git a/bitsandbytes/cextension.py b/bitsandbytes/cextension.py
index 7b10aaa..9066f91 100644
--- a/bitsandbytes/cextension.py
+++ b/bitsandbytes/cextension.py
@@ -1,9 +1,9 @@
 import ctypes as ct
-import torch
-
 from pathlib import Path
 from warnings import warn
 
+import torch
+
 
 class CUDASetup:
     _instance = None
diff --git a/bitsandbytes/cuda_setup/__init__.py b/bitsandbytes/cuda_setup/__init__.py
index d8ebba8..e781b9d 100644
--- a/bitsandbytes/cuda_setup/__init__.py
+++ b/bitsandbytes/cuda_setup/__init__.py
@@ -1,2 +1,6 @@
-from .paths import CUDA_RUNTIME_LIB, extract_candidate_paths, determine_cuda_runtime_lib_path
 from .main import evaluate_cuda_setup
+from .paths import (
+    CUDA_RUNTIME_LIB,
+    determine_cuda_runtime_lib_path,
+    extract_candidate_paths,
+)
diff --git a/bitsandbytes/cuda_setup/main.py b/bitsandbytes/cuda_setup/main.py
index 0bdce1a..8ea7d28 100644
--- a/bitsandbytes/cuda_setup/main.py
+++ b/bitsandbytes/cuda_setup/main.py
@@ -18,9 +18,10 @@ evaluation:
 
 import ctypes
 
-from .paths import determine_cuda_runtime_lib_path
 from bitsandbytes.cextension import CUDASetup
 
+from .paths import determine_cuda_runtime_lib_path
+
 
 def check_cuda_result(cuda, result_val):
     # 3. Check for CUDA errors
diff --git a/bitsandbytes/cuda_setup/paths.py b/bitsandbytes/cuda_setup/paths.py
index 3223359..f1adf4c 100644
--- a/bitsandbytes/cuda_setup/paths.py
+++ b/bitsandbytes/cuda_setup/paths.py
@@ -1,6 +1,7 @@
 import errno
 from pathlib import Path
 from typing import Set, Union
+
 from bitsandbytes.cextension import CUDASetup
 
 from .env_vars import get_potentially_lib_path_containing_env_vars
diff --git a/bitsandbytes/functional.py b/bitsandbytes/functional.py
index b3ca007..f6bd2be 100644
--- a/bitsandbytes/functional.py
+++ b/bitsandbytes/functional.py
@@ -5,13 +5,14 @@
 import ctypes as ct
 import operator
 import random
-import torch
-
+from functools import reduce  # Required in Python 3
 from typing import Tuple
+
+import torch
 from torch import Tensor
 
 from .cextension import COMPILED_WITH_CUDA, lib
-from functools import reduce  # Required in Python 3
+
 
 # math.prod not compatible with python < 3.8
 def prod(iterable):
diff --git a/bitsandbytes/optim/__init__.py b/bitsandbytes/optim/__init__.py
index d18f1d1..8c8a8f4 100644
--- a/bitsandbytes/optim/__init__.py
+++ b/bitsandbytes/optim/__init__.py
@@ -5,12 +5,11 @@
 
 from bitsandbytes.cextension import COMPILED_WITH_CUDA
 
+from .adagrad import Adagrad, Adagrad8bit, Adagrad32bit
 from .adam import Adam, Adam8bit, Adam32bit
 from .adamw import AdamW, AdamW8bit, AdamW32bit
-from .sgd import SGD, SGD8bit, SGD32bit
-from .lars import LARS, LARS8bit, LARS32bit, PytorchLARS
 from .lamb import LAMB, LAMB8bit, LAMB32bit
-from .rmsprop import RMSprop, RMSprop8bit, RMSprop32bit
-from .adagrad import Adagrad, Adagrad8bit, Adagrad32bit
-
+from .lars import LARS, LARS8bit, LARS32bit, PytorchLARS
 from .optimizer import GlobalOptimManager
+from .rmsprop import RMSprop, RMSprop8bit, RMSprop32bit
+from .sgd import SGD, SGD8bit, SGD32bit
diff --git a/tests/test_autograd.py b/tests/test_autograd.py
index f1cfa8d..c67126d 100644
--- a/tests/test_autograd.py
+++ b/tests/test_autograd.py
@@ -1,4 +1,4 @@
-from itertools import product, permutations
+from itertools import permutations, product
 
 import pytest
 import torch
diff --git a/tests/test_cuda_setup_evaluator.py b/tests/test_cuda_setup_evaluator.py
index bf9a003..7edc01f 100644
--- a/tests/test_cuda_setup_evaluator.py
+++ b/tests/test_cuda_setup_evaluator.py
@@ -1,13 +1,13 @@
 import os
-import pytest
-import bitsandbytes as bnb
-
 from typing import List, NamedTuple
 
+import pytest
+
+import bitsandbytes as bnb
 from bitsandbytes.cuda_setup import (
     CUDA_RUNTIME_LIB,
-    evaluate_cuda_setup,
     determine_cuda_runtime_lib_path,
+    evaluate_cuda_setup,
     extract_candidate_paths,
 )