Switched linters, black -> ruff

.github/workflows/python-quality.yml

@@ -16,5 +16,5 @@ jobs:
         run: |
           pip install --upgrade pip
           pip install .[dev]
-      - run: black --check bench examples optimum test
-      - run: ruff check bench examples optimum test
+      - run: ruff format bench examples optimum test --diff
+      - run: ruff check --show-fixes bench examples optimum test

Makefile

@@ -3,12 +3,12 @@
 check_dirs := optimum test bench examples
 
 check:
-	black --check ${check_dirs}
-	ruff check ${check_dirs}
+	ruff check --show-fixes ${check_dirs}
+	ruff format ${check_dirs} --diff
 
 style:
-	black ${check_dirs}
 	ruff check ${check_dirs} --fix
+	ruff format ${check_dirs}
 
 test:
 	python -m pytest -sv test

examples/nlp/text-generation/quantize_causal_lm_model.py

@@ -120,7 +120,9 @@ def main():
     torch_dtype = (
         torch.float16
         if args.load_dtype == "float16"
-        else torch.bfloat16 if args.load_dtype == "bfloat16" else torch.float32
+        else torch.bfloat16
+        if args.load_dtype == "bfloat16"
+        else torch.float32
     )
     model = AutoModelForCausalLM.from_pretrained(args.model, torch_dtype=torch_dtype, low_cpu_mem_usage=True).to(
         device

examples/vision/object-detection/quantize_owl_model.py

@@ -12,7 +12,6 @@
 
 
 def detect(model, processor, image, texts):
-
     inputs = processor(text=texts, images=image, return_tensors="pt").to(model.device)
 
     # forward pass

external/awq/test_awq_kernels.py

@@ -13,8 +13,8 @@
 # limitations under the License.
 import pytest
 import torch
-
 from pack import pack_awq
+
 from optimum.quanto import AffineQuantizer, MaxOptimizer, qint4, ungroup
 
 

external/awq/test_awq_packing.py

@@ -14,9 +14,9 @@
 import numpy as np
 import pytest
 import torch
-
 from pack_intweight import pack_intweight
 from packing_utils import pack_awq, reverse_awq_order, unpack_awq
+
 from optimum.quanto import AWQPackedTensor, AWQPacking
 
 

external/smoothquant/smoothquant.py

@@ -8,9 +8,9 @@
 from tqdm import tqdm
 from transformers import AutoModelForCausalLM, AutoTokenizer
 from transformers.models.bloom.modeling_bloom import BloomBlock
-from transformers.models.opt.modeling_opt import OPTDecoderLayer
 from transformers.models.llama.modeling_llama import LlamaDecoderLayer, LlamaRMSNorm
 from transformers.models.mistral.modeling_mistral import MistralDecoderLayer, MistralRMSNorm
+from transformers.models.opt.modeling_opt import OPTDecoderLayer
 
 
 def get_act_scales(model, tokenizer, dataset, num_samples=512, seq_len=512):

optimum/quanto/library/extensions/extension.py

@@ -11,7 +11,6 @@
 
 
 class Extension(object):
-
     def __init__(
         self,
         name: str,

optimum/quanto/models/diffusers_models.py

@@ -42,7 +42,6 @@
 
 
 class QuantizedDiffusersModel(ModelHubMixin):
-
     BASE_NAME = "quanto"
     base_class = None
 
@@ -188,5 +187,4 @@ def _save_pretrained(self, save_directory: Path) -> None:
 
 
 class QuantizedPixArtTransformer2DModel(QuantizedDiffusersModel):
-
     base_class = PixArtTransformer2DModel

optimum/quanto/models/transformers_models.py

@@ -36,7 +36,6 @@
 
 
 class QuantizedTransformersModel(ModelHubMixin):
-
     BASE_NAME = "quanto"
     auto_class = None
 
@@ -178,5 +177,4 @@ def _save_pretrained(self, save_directory: Path) -> None:
 
 
 class QuantizedModelForCausalLM(QuantizedTransformersModel):
-
     auto_class = AutoModelForCausalLM

optimum/quanto/subpackage/commands/base.py

@@ -23,7 +23,6 @@
 
 @optimum_cli_subcommand()
 class QuantoCommand(BaseOptimumCLICommand):
-
     COMMAND = CommandInfo(name="quanto", help="Hugging Face models quantization tools")
     SUBCOMMANDS = (
         CommandInfo(

optimum/quanto/tensor/activations/qbytes.py

@@ -26,7 +26,6 @@
 
 
 class ActivationQBytesQuantizer(Function):
-
     @staticmethod
     def forward(ctx, base: torch.Tensor, qtype: qtype, scale: torch.Tensor) -> torch.Tensor:
         if qtype.bits != 8:

optimum/quanto/tensor/optimizers/absmax_optimizer.py

@@ -24,7 +24,6 @@
 
 
 class AbsmaxOptimizer(SymmetricOptimizer):
-
     def optimize(
         self, base: torch.Tensor, qtype: qtype, axis: Optional[int] = None
     ) -> Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]:

optimum/quanto/tensor/optimizers/affine_optimizer.py

@@ -25,7 +25,6 @@
 
 
 class AffineOptimizer(Optimizer):
-
     def __call__(
         self,
         base: torch.Tensor,

optimum/quanto/tensor/optimizers/max_optimizer.py

@@ -24,7 +24,6 @@
 
 
 class MaxOptimizer(AffineOptimizer):
-
     def optimize(
         self, base: torch.Tensor, qtype: qtype, axis: int
     ) -> Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]:

optimum/quanto/tensor/optimizers/optimizer.py

@@ -22,7 +22,6 @@
 
 
 class Optimizer(ABC):
-
     def __call__(
         self, base: torch.Tensor, bits: int, axis: int, group_size: Optional[int] = None
     ) -> Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]:

optimum/quanto/tensor/optimizers/symmetric_optimizer.py

@@ -24,7 +24,6 @@
 
 
 class SymmetricOptimizer(Optimizer):
-
     def __call__(self, base: torch.Tensor, qtype: qtype, axis: Optional[int] = None) -> torch.Tensor:
         if axis not in [None, 0, -1]:
             raise ValueError("axis parameter must be None, 0 (first axis) or -1 (last axis)")

optimum/quanto/tensor/qbits.py

@@ -54,7 +54,6 @@ def backward(ctx, gO):
 
 
 class QBitsTensor(QTensor):
-
     def __init__(self, qtype, axis, group_size, size, stride, data, scale, shift, requires_grad=False):
         super().__init__(qtype, axis)
         self._data = data

optimum/quanto/tensor/qbytes.py

@@ -37,7 +37,6 @@ def backward(ctx, gO):
 
 
 class QBytesTensor(QTensor):
-
     def __init__(self, qtype, axis, size, stride, data, scale, requires_grad=False):
         super().__init__(qtype, axis)
         self._data = data

optimum/quanto/tensor/qtensor.py

@@ -30,7 +30,6 @@ def qfallback(callable, *args, **kwargs):
 
 
 class QTensor(torch.Tensor):
-
     def __init__(self, qtype, axis):
         self._qtype = qtype
         self._axis = axis

optimum/quanto/tensor/weights/marlin/fp8/qbits.py

@@ -61,7 +61,6 @@ def __new__(cls, qtype, axis, size, stride, data, scale, requires_grad=False):
         )
 
     def __init__(self, qtype, axis, size, stride, data, scale, requires_grad=False):
-
         assert axis == 0
         assert data.ndim == 2
 

optimum/quanto/tensor/weights/qbits.py

@@ -31,7 +31,6 @@
 
 
 class WeightsQBitsQuantizer(Function):
-
     @staticmethod
     def forward(
         ctx,

optimum/quanto/tensor/weights/qbytes.py

@@ -28,7 +28,6 @@
 
 
 class WeightQBytesQuantizer(Function):
-
     @staticmethod
     def forward(
         ctx, base: torch.Tensor, qtype: qtype, axis: int, scale: torch.Tensor, activation_qtype: qtype, optimized: bool

pyproject.toml

@@ -28,7 +28,7 @@ dynamic = ['version']
 homepage = 'https://github.com/huggingface/optimum-quanto'
 
 [project.optional-dependencies]
-dev = ['pytest', 'ruff', 'black']
+dev = ['pytest', 'ruff']
 examples = [
     'torchvision',
     'transformers',
@@ -50,19 +50,20 @@ version = {attr = 'optimum.quanto.__version__'}
 requires = ['setuptools>65.5.1', 'setuptools_scm']
 build-backend = 'setuptools.build_meta'
 
-[tool.black]
-line-length = 119
-
 [tool.ruff]
-# Never enforce `E501` (line length violations).
+# Configuration for Ruff
+line-length = 119  # Same line-length as Black had
+
+# Linting rules:
+# Never enforce `E501` (line length violations) and other specific rules.
 lint.ignore = ['C901', 'E501', 'E741']
 lint.select = ['C', 'E', 'F', 'I', 'W']
-line-length = 119
 
 # Ignore import violations in all `__init__.py` files.
 [tool.ruff.lint.per-file-ignores]
 '__init__.py' = ['E402', 'F401', 'F403', 'F811']
 
+# isort configuration (to sort imports)
 [tool.ruff.lint.isort]
 lines-after-imports = 2
-known-first-party = ['optimum.quanto']
+known-first-party = ['optimum.quanto']

setup.sh

@@ -12,6 +12,6 @@ else
     pip install --upgrade --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu118
 fi
 # Build tools
-pip install black ruff pytest build
+pip install ruff pytest build
 # For examples
 pip install accelerate transformers datasets

test/library/test_quantize.py

@@ -96,9 +96,7 @@ def test_affine_quantize(input_shape, dtype, qtype, axis, group_size, shift_mode
             "zeropoint": 6e-2,
             "float": 5e-2,
         },
-    }[
-        qtype
-    ][shift_mode]
+    }[qtype][shift_mode]
     if group_size is not None:
         qa = ungroup(qa, axis=axis, orig_shape=a.shape)
     assert_similar(a, qa, atol=atol)

test/quantize/test_quantize_patterns.py

@@ -36,7 +36,6 @@ def forward(self, inputs):
 
 
 class ClassificationModel(torch.nn.Module):
-
     def __init__(self, input_size, output_size, hidden_size, classes):
         super().__init__()
         self.model = MLP(input_size, output_size, hidden_size)

test/tensor/weights/test_weight_qbits_tensor_quantize.py

@@ -49,9 +49,7 @@ def test_weight_qbits_tensor_quantize(input_shape, dtype, qtype, axis, group_siz
             "zeropoint": 6e-2,
             "float": 5e-2,
         },
-    }[
-        qtype
-    ][shift_mode]
+    }[qtype][shift_mode]
     assert_similar(a, qa, atol=atol)