Ruff fixes (#984)

* Adjust Ruff configuration

* do not autofix always
* be less strict around tests and benchmarks
* adjust ignores for now

* Ruff: autofix I and F401

* Apply ruff autofixes

* Fix RUF013 complaint

* Fix mutable default in replace_linear

* Don't use bare except

* Wrap bitsandbytes.__main__ entrypoint in function; fix "sensible" typo

* Fix ruff B008 (function call in arguments)

* Add ruff noqas as suitable

* Fix RUF005 (splat instead of concatenating)

* Fix B018 (useless expression)

* Add pre-commit configuration + GitHub Actions lint workflow

* Fix unused `e` in bitsandbytes/__main__.py

* fix merge conflict resolution error

* run pre-commit hook

---------
Co-authored-by: Titus <9048635+Titus-von-Koeller@users.noreply.github.com>

bitsandbytes/research/nn/__init__.py

-from .modules import LinearFP8Mixed, LinearFP8Global
+from .modules import LinearFP8Global, LinearFP8Mixed

bitsandbytes/research/nn/modules.py

-from typing import Optional, TypeVar, Union, overload
+from typing import TypeVar
 
 import torch
-import torch.nn.functional as F
-from torch import Tensor, device, dtype, nn
+from torch import nn
 
 import bitsandbytes as bnb
-from bitsandbytes.optim import GlobalOptimManager
-from bitsandbytes.utils import OutlierTracer, find_outlier_dims
 
 T = TypeVar("T", bound="torch.nn.Module")
 

bitsandbytes/triton/dequantize_rowwise.py

 import math
+
 import torch
-import time
+
 from bitsandbytes.triton.triton_utils import is_triton_available
 
 if not is_triton_available():
@@ -9,7 +10,6 @@ else:
 
     import triton
     import triton.language as tl
-    from triton.ops.matmul_perf_model import early_config_prune, estimate_matmul_time
 
     # rowwise quantize
 

bitsandbytes/triton/int8_matmul_mixed_dequantize.py

 import torch
+
 from bitsandbytes.triton.triton_utils import is_triton_available
 
 if not is_triton_available():
@@ -57,7 +58,8 @@ else:
             triton.Config({'BLOCK_M': 64, 'BLOCK_N': 128, 'BLOCK_K': 64, 'SPLIT_K': 1}, num_stages=4, num_warps=4),
             triton.Config({'BLOCK_M': 128, 'BLOCK_N': 32, 'BLOCK_K': 64, 'SPLIT_K': 1}, num_stages=4, num_warps=4),
             triton.Config({'BLOCK_M': 64, 'BLOCK_N': 32, 'BLOCK_K': 64, 'SPLIT_K': 1}, num_stages=5, num_warps=2),
-        ] + get_configs_io_bound(),
+            *get_configs_io_bound(),
+        ],
         key=['M', 'N', 'K'],
         prune_configs_by={
             'early_config_prune': early_config_prune,

bitsandbytes/triton/int8_matmul_rowwise_dequantize.py

@@ -57,7 +57,8 @@ else:
             triton.Config({'BLOCK_M': 64, 'BLOCK_N': 128, 'BLOCK_K': 64, 'SPLIT_K': 1}, num_stages=4, num_warps=4),
             triton.Config({'BLOCK_M': 128, 'BLOCK_N': 32, 'BLOCK_K': 64, 'SPLIT_K': 1}, num_stages=4, num_warps=4),
             triton.Config({'BLOCK_M': 64, 'BLOCK_N': 32, 'BLOCK_K': 64, 'SPLIT_K': 1}, num_stages=5, num_warps=2),
-        ] + get_configs_io_bound(),
+            *get_configs_io_bound(),
+        ],
         key=['M', 'N', 'K'],
         prune_configs_by={
             'early_config_prune': early_config_prune,

bitsandbytes/triton/quantize_columnwise_and_transpose.py

 import math
+
 import torch
-import time
+
 from bitsandbytes.triton.triton_utils import is_triton_available
 
 if not is_triton_available():
@@ -9,7 +10,6 @@ else:
 
     import triton
     import triton.language as tl
-    from triton.ops.matmul_perf_model import early_config_prune, estimate_matmul_time
 
     # This kernel does fused columnwise quantization and transpose.
 

bitsandbytes/triton/quantize_global.py

-import math
+
 import torch
-import time
+
 from bitsandbytes.triton.triton_utils import is_triton_available
 
 if not is_triton_available():
@@ -10,7 +10,6 @@ else:
 
     import triton
     import triton.language as tl
-    from triton.ops.matmul_perf_model import early_config_prune, estimate_matmul_time
 
     # global quantize
     @triton.autotune(

bitsandbytes/triton/quantize_rowwise.py

 import math
+
 import torch
-import time
 
 from bitsandbytes.triton.triton_utils import is_triton_available
 
@@ -10,7 +10,6 @@ else:
 
     import triton
     import triton.language as tl
-    from triton.ops.matmul_perf_model import early_config_prune, estimate_matmul_time
 
     # rowwise quantize
 

bitsandbytes/triton/triton_utils.py

 import importlib
 
+
 def is_triton_available():
     return importlib.util.find_spec("triton") is not None

bitsandbytes/utils.py

 import json
 import shlex
 import subprocess
-import torch
 from typing import Tuple
 
+import torch
+
+
 def outlier_hook(module, input):
     assert isinstance(module, torch.nn.Linear)
     tracer = OutlierTracer.get_instance()
@@ -37,7 +39,7 @@ def outlier_hook(module, input):
             hook.remove()
 
 
-class OutlierTracer(object):
+class OutlierTracer:
     _instance = None
 
     def __init__(self):
@@ -122,7 +124,13 @@ def execute_and_return(command_string: str) -> Tuple[str, str]:
 
 
 
-def replace_linear(model, linear_replacement, skip_modules=["lm_head"], copy_weights=False, post_processing_function=None):
+def replace_linear(
+    model,
+    linear_replacement,
+    skip_modules=("lm_head",),
+    copy_weights=False,
+    post_processing_function=None,
+):
     """
     Replace linear modules with a new Linear module.
     Parameters:

check_bnb_install.py

-import bitsandbytes as bnb
 import torch
 
+import bitsandbytes as bnb
+
 p = torch.nn.Parameter(torch.rand(10,10).cuda())
 a = torch.rand(10,10).cuda()
 

install_cuda.py

 import os
-import sys
 import subprocess
+import sys
 from urllib.request import urlretrieve
 
 cuda_versions = {

pyproject.toml

@@ -11,9 +11,7 @@ src = [
     "tests",
     "benchmarking"
 ]
-fix = true
 select = [
-    "A",    # prevent using keywords that clobber python builtins
     "B",    # bugbear: security warnings
     "E",    # pycodestyle
     "F",    # pyflakes
@@ -24,12 +22,29 @@ select = [
 ]
 target-version = "py38"
 ignore = [
-    "E712",  # Allow using if x == False, as it's not always equivalent to if x.
+    "B007",  # Loop control variable not used within the loop body (TODO: enable)
+    "B028",  # Warning without stacklevel (TODO: enable)
     "E501",  # Supress line-too-long warnings: trust yapf's judgement on this one.
-    "F401", 
+    "E701",  # Multiple statements on one line (TODO: enable)
+    "E712",  # Allow using if x == False, as it's not always equivalent to if x.
+    "E731",  # Do not use lambda
+    "F841",  # Local assigned but not used (TODO: enable, these are likely bugs)
+    "RUF012",  # Mutable class attribute annotations
 ]
 ignore-init-module-imports = true  # allow to expose in __init__.py via imports
 
+[tool.ruff.extend-per-file-ignores]
+"**/__init__.py" = ["F401"]  # allow unused imports in __init__.py
+"{benchmarking,tests}/**/*.py" = [
+    "B007",
+    "B011",
+    "B023",
+    "E701",
+    "E731",
+    "F841",
+    "UP030",
+]
+
 [tool.ruff.isort]
 combine-as-imports = true
 detect-same-package = true

scripts/stale.py

@@ -15,13 +15,11 @@
 Script to close stale issue. Taken in part from the AllenNLP repository.
 https://github.com/allenai/allennlp.
 """
+from datetime import datetime as dt, timezone
 import os
-from datetime import datetime as dt
-from datetime import timezone
 
 from github import Github
 
-
 # All labels that we don't want to touch
 LABELS_TO_EXEMPT = [
     "feature-request",

setup.py

@@ -7,7 +7,6 @@ import os
 
 from setuptools import find_packages, setup
 
-
 libs = list(glob.glob("./bitsandbytes/libbitsandbytes*.so"))
 libs += list(glob.glob("./bitsandbytes/libbitsandbytes*.dll"))
 libs = [os.path.basename(p) for p in libs]
@@ -19,7 +18,7 @@ def read(fname):
 
 
 setup(
-    name=f"bitsandbytes",
+    name="bitsandbytes",
     version="0.42.0",
     author="Tim Dettmers",
     author_email="dettmers@cs.washington.edu",

tests/test_autograd.py

-from itertools import permutations, product
+from itertools import product
 
 import pytest
 import torch

tests/test_cuda_setup_evaluator.py

 import os
-import pytest
-import torch
 from pathlib import Path
 
+import torch
+
+
 # hardcoded test. Not good, but a sanity check for now
 # TODO: improve this
 def test_manual_override(requires_cuda):

tests/test_functional.py

+from itertools import product
 import math
 import random
 import time
-from itertools import product
 
 import einops
+import numpy as np
 import pytest
+from scipy.stats import norm
 import torch
-import numpy as np
 
 import bitsandbytes as bnb
 from bitsandbytes import functional as F
-from scipy.stats import norm
 
 torch.set_printoptions(
     precision=5, sci_mode=False, linewidth=120, edgeitems=20, threshold=10000

tests/test_generation.py

-import pytest
-import torch
-import math
-
 from itertools import product
+import math
 
+import pytest
+import torch
 import transformers
 from transformers import (
-  AutoConfig,
   AutoModelForCausalLM,
-  AutoTokenizer,
   BitsAndBytesConfig,
-  GenerationConfig,
-  set_seed,
-
 )
 
 
-
 def get_4bit_config():
   return BitsAndBytesConfig(
     load_in_4bit=True,

tests/test_linear4bit.py

-import os
-from contextlib import nullcontext
 from itertools import product
+import os
 from tempfile import TemporaryDirectory
 
 import pytest