Enable ruff-check in pre-commit (#2192)

Signed-off-by: Masaki Kozuki <mkozuki@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

Author: crcrpar

.pre-commit-config.yaml

@@ -38,6 +38,12 @@ repos:
         additional_dependencies: [tomli]
         #args: ["--write-changes"] # uncomment if you want to get automatic fixing
 
+  - repo: https://github.com/astral-sh/ruff-pre-commit
+    rev: v0.11.11
+    hooks:
+      - id: ruff-check
+        args: ["--fix"]
+
   - repo: https://github.com/psf/black
     rev: 25.1.0
     hooks:
@@ -65,12 +71,6 @@ repos:
     hooks:
       - id: yesqa
 
-  #  - repo: https://github.com/charliermarsh/ruff-pre-commit
-  #    rev: v0.0.270
-  #    hooks:
-  #      - id: ruff
-  #        args: ["--fix"]
-
   - repo: https://github.com/pre-commit/mirrors-prettier
     rev: v3.1.0
     hooks:

pyproject.toml

@@ -121,23 +121,12 @@ line_length = 120
 line-length = 120
 # Always generate Python 3.10-compatible code.
 target-version = "py310"
-# Enable Pyflakes `E` and `F` codes by default.
-select = [
-    "E", "W",  # see: https://pypi.org/project/pycodestyle
-    "F",  # see: https://pypi.org/project/pyflakes
-#    "D",  # see: https://pypi.org/project/pydocstyle
-#    "N",  # see: https://pypi.org/project/pep8-naming
-]
 #extend-select = [
 #    "C4",  # see: https://pypi.org/project/flake8-comprehensions
 #    "PT",  # see: https://pypi.org/project/flake8-pytest-style
 #    "RET",  # see: https://pypi.org/project/flake8-return
 #    "SIM",  # see: https://pypi.org/project/flake8-simplify
 #]
-ignore = [
-    "E731",  # Do not assign a lambda expression, use a def
-    "E501",  # todo: Line too long (235 > 120 characters)
-]
 # Exclude a variety of commonly ignored directories.
 exclude = [
     ".eggs",
@@ -148,23 +137,60 @@ exclude = [
     "_build",
     "build",
     "dist",
-    "docs"
+    "docs",
+    "examples",
+    "notebooks",
+]
+
+[tool.ruff.lint]
+# Enable Pyflakes `E` and `F` codes by default.
+select = [
+    "E", "W",  # see: https://pypi.org/project/pycodestyle
+    "F",  # see: https://pypi.org/project/pyflakes
+#    "D",  # see: https://pypi.org/project/pydocstyle
+#    "N",  # see: https://pypi.org/project/pep8-naming
+]
+ignore = [
+    "E731",  # Do not assign a lambda expression, use a def
+    "E501",  # todo: Line too long (235 > 120 characters)
+    # TODO(crcrpar): Resolves the following ignores as these are added while enabling ruff check in pre-commit
+    "F841",  # https://docs.astral.sh/ruff/rules/unused-variable/
+    "F821",  # https://docs.astral.sh/ruff/rules/undefined-name/
+    "E402",  # https://docs.astral.sh/ruff/rules/module-import-not-at-top-of-file/
+    "F401",  # https://docs.astral.sh/ruff/rules/unused-import/
+    "F405",  # https://docs.astral.sh/ruff/rules/undefined-local-with-import-star-usage/
+    "E712",  # https://docs.astral.sh/ruff/rules/true-false-comparison/
+    "E711",  # https://docs.astral.sh/ruff/rules/none-comparison/
+    "E721",  # https://docs.astral.sh/ruff/rules/type-comparison/
+    "E722",  # https://docs.astral.sh/ruff/rules/bare-except/
+    "F403",  # https://docs.astral.sh/ruff/rules/undefined-local-with-import-star/
+    "F601",  # https://docs.astral.sh/ruff/rules/multi-value-repeated-key-literal/
+    "F822",  # https://docs.astral.sh/ruff/rules/undefined-export/
+    "F842",  # https://docs.astral.sh/ruff/rules/unused-annotation/
+    "F722",  # https://docs.astral.sh/ruff/rules/forward-annotation-syntax-error/
+    "E702",  # https://docs.astral.sh/ruff/rules/multiple-statements-on-one-line-semicolon/
+    "E741",  # https://docs.astral.sh/ruff/rules/ambiguous-variable-name/
+    "W293",  # https://docs.astral.sh/ruff/rules/blank-line-with-whitespace/
+    "F811",  # https://docs.astral.sh/ruff/rules/redefined-while-unused/
+    "W291",  # https://docs.astral.sh/ruff/rules/trailing-whitespace/
 ]
-ignore-init-module-imports = true
 
-[tool.ruff.per-file-ignores]
+[tool.ruff.lint.per-file-ignores]
 "setup.py" = ["D100", "SIM115"]
 "__about__.py" = ["D100"]
 "__init__.py" = ["D100"]
+# The default is `true` and this option is deprecated as per
+# https://docs.astral.sh/ruff/settings/#lint_ignore-init-module-imports
+# "ignore-init-module-imports" = true
 
-[tool.ruff.pydocstyle]
+[tool.ruff.lint.pydocstyle]
 # Use Google-style docstrings.
 convention = "google"
 
 #[tool.ruff.pycodestyle]
 #ignore-overlong-task-comments = true
 
-[tool.ruff.mccabe]
+[tool.ruff.lint.mccabe]
 # Unlike Flake8, default to a complexity level of 10.
 max-complexity = 10
 

scripts/bisect_nvfuser.py

@@ -63,6 +63,6 @@ def run(self, fuel: int) -> int:
             high = mid
     assert low == high
 
-    print(f"Bisecting succeeded. Run the following command as a minimal reproducer:")
+    print("Bisecting succeeded. Run the following command as a minimal reproducer:")
     print(f"    {runner.env_var_for_fuel()}={low} {' '.join(args.command_and_args)}")
-    print(f"The last nvFusion likely triggered the failure.")
+    print("The last nvFusion likely triggered the failure.")

thunder/__init__.py

@@ -82,6 +82,7 @@
 
 import thunder.clang as clang
 from thunder.core.pytree import tree_flatten, tree_unflatten, tree_map
+import thunder.transforms as transforms
 
 # Imports executors (to populate default executors and make them accessible)
 import thunder.executors.pythonex
@@ -97,6 +98,7 @@
 
 # TODO RC1 Review exposed names
 __all__ = [
+    "transforms",
     # dtype aliases
     "bool8",
     "uint8",

thunder/benchmarks/__init__.py

@@ -619,19 +619,19 @@ def run_multiprocess_benchmark(
 
     assert (
         torch.distributed.is_available()
-    ), f"Trying to run a distributed benchmark, but torch.distributed is not available"
+    ), "Trying to run a distributed benchmark, but torch.distributed is not available"
 
     # Ensures the benchmark is running on a single CUDA device (which is overridden later)
     assert (
         len(benchmark.devices) == 1
         and Devices.device_from_string(benchmark.devices[0]).devicetype == Devices.DeviceType.CUDA
-    ), f"Distributed benchmarking currently only supports benchmarks that run on a single CUDA device"
+    ), "Distributed benchmarking currently only supports benchmarks that run on a single CUDA device"
 
     # Ensures the benchmark returns a module (because ddp is only supported on modules)
     benchmark_fn = benchmark.fn()
     assert isinstance(
         benchmark_fn, torch.nn.Module
-    ), f"Distributed benchmarking currently only supports module benchmarks"
+    ), "Distributed benchmarking currently only supports module benchmarks"
 
     # Validates world size
     assert (

thunder/benchmarks/benchmark_litgpt.py

@@ -428,7 +428,7 @@ def init_model(self):
         init_device = torch.device("meta") if self.distributed_mode in FSDP_MODES else self.device
         if self.use_hf:
             warnings.warn(
-                f"HuggingFace transformers mode is experimental, many options do not apply. Preliminary testing  with transformers==4.50.3."
+                "HuggingFace transformers mode is experimental, many options do not apply. Preliminary testing  with transformers==4.50.3."
             )
 
             # for the materialization, we need reset_parameters
@@ -478,7 +478,7 @@ def setup_distributed(self, model):
 
         # Distributed Setup
         # TODO: Change compiler call names
-        if "thunder" in self.compile and not "dynamo" in self.compile:
+        if "thunder" in self.compile and "dynamo" not in self.compile:
             if self.distributed_mode == "ddp":
                 from thunder.distributed import ddp
 
@@ -506,7 +506,7 @@ def setup_distributed(self, model):
             else:
                 if self.distributed_mode == "fsdp2":
                     raise ValueError(
-                        f"To use `fsdp2`, use thunder as torch.compile backend by including dynamo in `--compile` option or set `--compile` to either eager or inductor"
+                        "To use `fsdp2`, use thunder as torch.compile backend by including dynamo in `--compile` option or set `--compile` to either eager or inductor"
                     )
         else:
             if self.distributed_mode == "ddp":

thunder/benchmarks/einsum.py

@@ -1,9 +1,7 @@
 from collections.abc import Callable
-from functools import partial, wraps
 from collections.abc import Sequence
 
 import pytest
-import torch
 import thunder
 
 from thunder.benchmarks import EinsumBenchmark

thunder/benchmarks/targets.py

@@ -48,7 +48,6 @@
 from thunder.core.interpreter import interpret
 
 from thunder.tests.litgpt_model import Config as LitGPTConfig
-from thunder.tests.make_tensor import make_tensor
 from thunder.benchmarks.utils import backward_only
 
 LIGER_FUSED_SWIGLU_AVAILABLE: bool = package_available("liger_kernel.ops.swiglu")

thunder/clang/__init__.py

@@ -794,7 +794,7 @@ def _to_tensorproxies(x: list, device: devices.DeviceType):
                 lambda: f"Advanced indexing currently only supports zero or one-dimensional integer tensors, but found a tensor with dtype {x.dtype} and {x.ndim} dimensions",
             )
 
-    utils.check(num_ellipses <= 1, lambda: f"Found two or more ellipses in an advanced indexing key")
+    utils.check(num_ellipses <= 1, lambda: "Found two or more ellipses in an advanced indexing key")
 
     # NOTE When the key has an ellipsis it can be longer than the number of dimensions in a
     #   (in this case the ellipsis matches no dimensions)
@@ -808,7 +808,7 @@ def _to_tensorproxies(x: list, device: devices.DeviceType):
     has_ellipsis: bool = num_ellipses > 0
     utils.check(
         not has_ellipsis or key[0] is Ellipsis,
-        lambda: f"Advanced indexing currently only supports ellipses as the first sequence element",
+        lambda: "Advanced indexing currently only supports ellipses as the first sequence element",
     )
 
     # The following models two advanced indexing cases:
@@ -1020,8 +1020,8 @@ def movedim(a: TensorLike, /, source: int | Sequence[int], destination: int | Se
     # Verifies that dims are uniquely specified
     # NOTE This must be done after canonicalization, since canonicalization resolves different ways of specifying the same dim
     src_set = set(src)
-    utils.check(len(src_set) == len(src), lambda: f"Found at least one source dimension specified multiple times")
-    utils.check(len(set(dst)) == len(dst), lambda: f"Found at least one destination dimension specified multiple times")
+    utils.check(len(src_set) == len(src), lambda: "Found at least one source dimension specified multiple times")
+    utils.check(len(set(dst)) == len(dst), lambda: "Found at least one destination dimension specified multiple times")
 
     # Constructs a permutation that moves the dimensions as requested
     # NOTE Essentially move_dim specifies a partial permutation, where dimensions not explicitly specified as moving
@@ -1318,7 +1318,7 @@ def cat(tensors: list[TensorProxy], dim: int):
 def stack(tensors: list[TensorProxy], dim: int):
     """Concatenates the given sequence of tensors in a new (the given) dimension."""
     shapes = tuple(t.shape for t in tensors)
-    utils.check(shapes, lambda: f"list of tensors cannot be empty")
+    utils.check(shapes, lambda: "list of tensors cannot be empty")
     for i, s in enumerate(shapes[1:], start=1):
         utils.check(
             s == shapes[0], lambda: f"tensors must be of the same shape, tensor at {i} is {s} instead of {shapes[0]}"
@@ -1388,7 +1388,7 @@ def matrix_transpose(a: TensorProxy) -> TensorProxy:
         mT_scalar_warning()
         return a
     elif a.ndim == 1:
-        raise RuntimeError(f"tensor.mT is only supported on matrices or batches of matrices. Got 1-D tensor.")
+        raise RuntimeError("tensor.mT is only supported on matrices or batches of matrices. Got 1-D tensor.")
 
     dim0, dim1 = -2, -1
     dim0, dim1 = utils.canonicalize_dims(a.ndim, (dim0, dim1))
@@ -1805,7 +1805,7 @@ def real(a: TensorProxy | Number):
 def imag(a: TensorProxy | Number, /) -> TensorLike:
     utils.check(
         dtypes.is_complex_dtype(dtypes.to_dtype(a)),
-        lambda: f"imag is not implemented for tensors with non-complex dtypes",
+        lambda: "imag is not implemented for tensors with non-complex dtypes",
     )
 
     return _elementwise_unary_wrapper(
@@ -1874,7 +1874,7 @@ def bitwise_xor(a, b):
 def copysign(a, b):
     utils.check(
         not dtypes.is_complex_dtype(dtypes.to_dtype(a)) and not dtypes.is_complex_dtype(dtypes.to_dtype(b)),
-        lambda: f"copysign is not defined for complex dtypes",
+        lambda: "copysign is not defined for complex dtypes",
     )
 
     computation_dtype, result_dtype = utils.elementwise_type_promotion(
@@ -1968,7 +1968,7 @@ def floor_divide(a: TensorProxy | Number, b: TensorProxy | Number) -> TensorProx
         a, b, type_promotion_kind=utils.ELEMENTWISE_TYPE_PROMOTION_KIND.DEFAULT
     )
 
-    utils.check(not dtypes.is_complex_dtype(computation_dtype), lambda: f"Complex floor division is not supported")
+    utils.check(not dtypes.is_complex_dtype(computation_dtype), lambda: "Complex floor division is not supported")
 
     if dtypes.is_float_dtype(computation_dtype):
         return _floor_divide_float(a, b)

thunder/clang/langctx.py

@@ -1,5 +1,4 @@
-from typing import Any
-from collections.abc import Callable, Sequence
+from collections.abc import Callable
 
 from thunder.core.langctxs import LanguageContext, register_langctx, Languages, resolve_language
 from thunder.core.pytree import tree_flatten