Fix global flake8 issues (pytorch#124771)

Prior to this `lintrunner --all-files --take FLAKE8` failed.

Pull Request resolved: pytorch#124771
Approved by: https://github.com/Skylion007
ghstack dependencies: pytorch#124428

.github/scripts/cherry_pick.py

@@ -29,7 +29,7 @@ def parse_args() -> Any:
         "--onto-branch", type=str, required=True, help="the target release branch"
     )
     parser.add_argument(
-        "--github-actor", type=str, required=True, help="all the world’s a stage"
+        "--github-actor", type=str, required=True, help="all the world's a stage"
     )
     parser.add_argument(
         "--classification",

benchmarks/transformer/better_transformer_vs_mha_functional.py

@@ -152,8 +152,8 @@ def run(
     result_entry["sequence_length"] = sequence_length
     result_entry["n_heads"] = num_heads
     result_entry["embed_dim"] = embed_dim
-    result_entry["time_native_mha_slow(μs)"] = f"{time_native_mha_slow:.3f}"
-    result_entry["time_native_mha_fast (μs)"] = f"{time_native_mha_fast:.3f}"
+    result_entry["time_native_mha_slow(\u00B5s)"] = f"{time_native_mha_slow:.3f}"
+    result_entry["time_native_mha_fast (\u00B5s)"] = f"{time_native_mha_fast:.3f}"
     result_entry["speedup flash_mha v native_mha"] = f"{speedup_fast_internal:.3f}"
     result_entry["padding"] = f"{padding:.3f}"
     return result_entry

benchmarks/transformer/sdp.py

@@ -81,10 +81,10 @@ def get_entries(self) -> List:
     @classmethod
     def get_entry_names(cls) -> List[str]:
         return [
-            "nn_mha_time (μs)",
-            "compiled_nn_mha_time (μs)",
-            "composite_mha_time (μs)",
-            "compiled_composite_mha_time (μs)",
+            "nn_mha_time (\u00B5s)",
+            "compiled_nn_mha_time (\u00B5s)",
+            "composite_mha_time (\u00B5s)",
+            "compiled_composite_mha_time (\u00B5s)",
         ]
 
 

functorch/einops/_parsing.py

@@ -28,7 +28,7 @@
 import warnings
 from typing import Collection, List, Mapping, Optional, Set, Tuple, Union
 
-_ellipsis: str = "…"  # NB, this is a single unicode symbol. String is used as it is not a list, but can be iterated
+_ellipsis: str = "\u2026"  # NB, this is a single unicode symbol. String is used as it is not a list, but can be iterated
 
 
 class AnonymousAxis:

test/distributions/test_distributions.py

@@ -3752,11 +3752,11 @@ def test_dirichlet_log_prob(self):
 
     @unittest.skipIf(not TEST_NUMPY, "NumPy not found")
     def test_dirichlet_log_prob_zero(self):
-        # Specifically test the special case where x=0 and α=1.  The PDF is
-        # proportional to x**(α-1), which in this case works out to 0**0=1.
+        # Specifically test the special case where x=0 and alpha=1.  The PDF is
+        # proportional to x**(alpha-1), which in this case works out to 0**0=1.
         # The log PDF of this term should therefore be 0.  However, it's easy
         # to accidentally introduce NaNs by calculating log(x) without regard
-        # for the value of α-1.
+        # for the value of alpha-1.
         alpha = torch.tensor([1, 2])
         dist = Dirichlet(alpha)
         x = torch.tensor([0, 1])

test/functorch/test_parsing.py

@@ -107,7 +107,7 @@ def test_invalid_expressions(self) -> None:
             ParsedExpression("(a) ((b c) (d ...))")
 
         # invalid identifiers
-        ParsedExpression("camelCase under_scored cApiTaLs ß ...")
+        ParsedExpression("camelCase under_scored cApiTaLs \u00DF ...")
         with self.assertRaises(ValueError):
             ParsedExpression("1a")
         with self.assertRaises(ValueError):

test/inductor/test_templated_attention.py

@@ -346,8 +346,8 @@ def sdpa_hop(q, k, v, score_mod):
         # this means that the base for the LSE computed by ref is e while for the compiled
         # version it is 2. To compare we use the change of base formula
         # log_2(x_compiled) = log_e(x_ref) * log_2(e) where
-        # x_ref      = ∑_i e^(scores[i])
-        # x_compiled = ∑_i 2^(log2(e) * scores[i])
+        # x_ref      = sum(_i e^(scores[i]))
+        # x_compiled = sum(_i 2^(log2(e) * scores[i]))
 
         self.assertTrue(ref_lse.dtype == torch.float32)
         self.assertTrue(compiled_lse.dtype == torch.float32)

test/package/test_directory_reader.py

@@ -111,16 +111,16 @@ def test_resource_reader(self):
         with PackageExporter(filename) as pe:
             # Layout looks like:
             #    package
-            #    ├── one/
-            #    │   ├── a.txt
-            #    │   ├── b.txt
-            #    │   ├── c.txt
-            #    │   └── three/
-            #    │       ├── d.txt
-            #    │       └── e.txt
-            #    └── two/
-            #       ├── f.txt
-            #       └── g.txt
+            #    |-- one/
+            #    |   |-- a.txt
+            #    |   |-- b.txt
+            #    |   |-- c.txt
+            #    |   +-- three/
+            #    |       |-- d.txt
+            #    |       +-- e.txt
+            #    +-- two/
+            #       |-- f.txt
+            #       +-- g.txt
             pe.save_text("one", "a.txt", "hello, a!")
             pe.save_text("one", "b.txt", "hello, b!")
             pe.save_text("one", "c.txt", "hello, c!")

test/package/test_misc.py

@@ -38,46 +38,46 @@ def test_file_structure(self):
 
         export_plain = dedent(
             """\
-                ├── .data
-                │   ├── extern_modules
-                │   ├── python_version
-                │   ├── serialization_id
-                │   └── version
-                ├── main
-                │   └── main
-                ├── obj
-                │   └── obj.pkl
-                ├── package_a
-                │   ├── __init__.py
-                │   └── subpackage.py
-                ├── byteorder
-                └── module_a.py
+                \u251c\u2500\u2500 .data
+                \u2502   \u251c\u2500\u2500 extern_modules
+                \u2502   \u251c\u2500\u2500 python_version
+                \u2502   \u251c\u2500\u2500 serialization_id
+                \u2502   \u2514\u2500\u2500 version
+                \u251c\u2500\u2500 main
+                \u2502   \u2514\u2500\u2500 main
+                \u251c\u2500\u2500 obj
+                \u2502   \u2514\u2500\u2500 obj.pkl
+                \u251c\u2500\u2500 package_a
+                \u2502   \u251c\u2500\u2500 __init__.py
+                \u2502   \u2514\u2500\u2500 subpackage.py
+                \u251c\u2500\u2500 byteorder
+                \u2514\u2500\u2500 module_a.py
             """
         )
         export_include = dedent(
             """\
-                ├── obj
-                │   └── obj.pkl
-                └── package_a
-                    └── subpackage.py
+                \u251c\u2500\u2500 obj
+                \u2502   \u2514\u2500\u2500 obj.pkl
+                \u2514\u2500\u2500 package_a
+                    \u2514\u2500\u2500 subpackage.py
             """
         )
         import_exclude = dedent(
             """\
-                ├── .data
-                │   ├── extern_modules
-                │   ├── python_version
-                │   ├── serialization_id
-                │   └── version
-                ├── main
-                │   └── main
-                ├── obj
-                │   └── obj.pkl
-                ├── package_a
-                │   ├── __init__.py
-                │   └── subpackage.py
-                ├── byteorder
-                └── module_a.py
+                \u251c\u2500\u2500 .data
+                \u2502   \u251c\u2500\u2500 extern_modules
+                \u2502   \u251c\u2500\u2500 python_version
+                \u2502   \u251c\u2500\u2500 serialization_id
+                \u2502   \u2514\u2500\u2500 version
+                \u251c\u2500\u2500 main
+                \u2502   \u2514\u2500\u2500 main
+                \u251c\u2500\u2500 obj
+                \u2502   \u2514\u2500\u2500 obj.pkl
+                \u251c\u2500\u2500 package_a
+                \u2502   \u251c\u2500\u2500 __init__.py
+                \u2502   \u2514\u2500\u2500 subpackage.py
+                \u251c\u2500\u2500 byteorder
+                \u2514\u2500\u2500 module_a.py
             """
         )
 

test/package/test_resources.py

@@ -25,16 +25,16 @@ def test_resource_reader(self):
         with PackageExporter(buffer) as pe:
             # Layout looks like:
             #    package
-            #    ├── one/
-            #    │   ├── a.txt
-            #    │   ├── b.txt
-            #    │   ├── c.txt
-            #    │   └── three/
-            #    │       ├── d.txt
-            #    │       └── e.txt
-            #    └── two/
-            #       ├── f.txt
-            #       └── g.txt
+            #    |-- one/
+            #    |   |-- a.txt
+            #    |   |-- b.txt
+            #    |   |-- c.txt
+            #    |   +-- three/
+            #    |       |-- d.txt
+            #    |       +-- e.txt
+            #    +-- two/
+            #       |-- f.txt
+            #       +-- g.txt
             pe.save_text("one", "a.txt", "hello, a!")
             pe.save_text("one", "b.txt", "hello, b!")
             pe.save_text("one", "c.txt", "hello, c!")

test/test_jit.py

@@ -15679,7 +15679,7 @@ def fn(string):
     def test_unicode_comments(self):
         @torch.jit.script
         def test(self, a):
-            # 🤷🤷🤷🤷
+            # shrug
             return torch.nn.functional.relu(a)
 
     def test_get_set_state_with_tensors(self):

test/test_jit_fuser.py

@@ -70,7 +70,7 @@ def test_abs_cpu(self):
     @unittest.skipIf(IS_SANDCASTLE, "NYI: fuser CPU support for Sandcastle")
     @enable_cpu_fuser
     def test_abs_cpu_unicode_temp_dir(self):
-        with TemporaryDirectoryName(suffix='中文') as dname:
+        with TemporaryDirectoryName(suffix='\u4e2d\u6587') as dname:
             shell_env = os.environ.copy()
             shell_env['TMP'] = dname
             cmd = [sys.executable, os.path.basename(__file__), type(self).__name__ + '.test_abs_cpu']

test/test_linalg.py

@@ -1950,7 +1950,7 @@ def test_eig_errors_and_warnings(self, device, dtype):
 
         # if out tensor with floating dtype is passed for complex output an error is thrown
         if not dtype.is_complex:
-            # The characteristic equation is p(λ) = λ^2 − 2λ + 5 = 0, with roots λ = 1±2i
+            # The characteristic equation is p(lambda) = lambda^2 - 2lambda + 5 = 0, with roots lambda = 1[+-]2i
             a = torch.tensor([[3., -2.], [4., -1.]], dtype=dtype, device=device)
             out0 = torch.empty(0, device=device, dtype=dtype)
             out1 = torch.empty(0, device=device, dtype=dtype)
@@ -2117,7 +2117,7 @@ def test_eigvals_errors_and_warnings(self, device, dtype):
 
         # if out tensor with floating dtype is passed for complex output an error is thrown
         if not dtype.is_complex:
-            # The characteristic equation is p(λ) = λ^2 − 2λ + 5 = 0, with roots λ = 1±2i
+            # The characteristic equation is p(lambda) = lambda^2 - 2lambda + 5 = 0, with roots lambda = 1[+-]2i
             a = torch.tensor([[3., -2.], [4., -1.]], dtype=dtype, device=device)
             out = torch.empty(0, device=device, dtype=dtype)
             with self.assertRaisesRegex(RuntimeError, "Expected eigenvalues to be safely castable"):

test/test_public_bindings.py

@@ -428,7 +428,7 @@ def test_modules_can_be_imported(self):
     def test_correct_module_names(self):
         '''
         An API is considered public, if  its  `__module__` starts with `torch.`
-        and there is no name in `__module__` or the object itself that starts with “_”.
+        and there is no name in `__module__` or the object itself that starts with "_".
         Each public package should either:
         - (preferred) Define `__all__` and all callables and classes in there must have their
          `__module__` start with the current submodule's path. Things not in `__all__` should

test/test_serialization.py

@@ -924,7 +924,7 @@ def test(name_or_buffer):
             test(fname)
 
         if IS_FILESYSTEM_UTF8_ENCODING:
-            with TemporaryDirectoryName(suffix='非ASCIIパス') as dname:
+            with TemporaryDirectoryName(suffix='\u975eASCII\u30d1\u30b9') as dname:
                 with TemporaryFileName(dir=dname) as fname:
                     test(fname)
 

test/test_torch.py

@@ -8046,7 +8046,7 @@ def assert_with_filename(filename):
             assert_with_filename(fname)
 
         if IS_FILESYSTEM_UTF8_ENCODING:
-            with TemporaryDirectoryName(suffix='中文') as dname, TemporaryFileName(dir=dname) as fname:
+            with TemporaryDirectoryName(suffix='\u4e2d\u6587') as dname, TemporaryFileName(dir=dname) as fname:
                 assert_with_filename(fname)
 
     def test_torch_from_file(self):
@@ -8077,7 +8077,7 @@ def assert_with_filename(filename):
             assert_with_filename(fname)
 
         if IS_FILESYSTEM_UTF8_ENCODING:
-            with TemporaryDirectoryName(suffix='中文') as dname, TemporaryFileName(dir=dname) as fname:
+            with TemporaryDirectoryName(suffix='\u4e2d\u6587') as dname, TemporaryFileName(dir=dname) as fname:
                 assert_with_filename(fname)
 
     def test_print(self):

torch/_decomp/decompositions.py

@@ -744,7 +744,7 @@ def slice_forward(
         raise RuntimeError("slice step must be positive")
 
     start_val = start if start is not None else 0
-    end_val = end if end is not None else sys.maxsize  # 2^63 – 1
+    end_val = end if end is not None else sys.maxsize  # 2^63 - 1
 
     if start_val < 0:
         start_val += sizes[dim]

torch/_dynamo/polyfill.py

@@ -57,7 +57,7 @@ def list_cmp(op: Callable[[Any, Any], bool], left: Sequence[Any], right: Sequenc
 
 
 def dropwhile(predicate, iterable):
-    # dropwhile(lambda x: x<5, [1,4,6,4,1]) → 6 4 1
+    # dropwhile(lambda x: x<5, [1,4,6,4,1]) -> 6 4 1
     iterable = iter(iterable)
     for x in iterable:
         if not predicate(x):

torch/_export/error.py

@@ -5,13 +5,13 @@ class ExportErrorType(Enum):
     # User providing invalid inputs to either tracer, or other public facing APIs
     INVALID_INPUT_TYPE = 1
 
-    # User returning values from their models that we don’t support.
+    # User returning values from their models that we don't support.
     INVALID_OUTPUT_TYPE = 2
 
     # Generated IR does not conform to Export IR Specification.
     VIOLATION_OF_SPEC = 3
 
-    # User’s code contains types and functionalities we don’t support.
+    # User's code contains types and functionalities we don't support.
     NOT_SUPPORTED = 4
 
     # User's code didn't provide necessary details for us to successfully trace and export.

torch/_functorch/autograd_function.py

@@ -498,7 +498,7 @@ def get_tangents_in_dims(input_dims, tangents):
 # in_dims = 0
 # vmap(Sum.apply, in_dims)(x)
 #
-# Let’s assume for a moment that we didn’t vmap setup_context in VmappedSum:
+# Let's assume for a moment that we didn't vmap setup_context in VmappedSum:
 #
 # class VmappedSum(torch.autograd.Function):
 #    @staticmethod
@@ -519,7 +519,7 @@ def get_tangents_in_dims(input_dims, tangents):
 #        return gx
 #
 # We end up saving [B, 4] as x_shape. In the backward, gy has shape [B],
-# and we’re doing:
+# and we're doing:
 #
 # def backward_no_context(gy):
 #     return gy.expand([B, 4])

torch/_inductor/codegen/memory_planning.py

@@ -62,8 +62,8 @@ class LiveRange:
     Invariant: begin <= end
     """
 
-    begin: float  # int | ±inf
-    end: float  # int | ±inf
+    begin: float  # int | +/-inf
+    end: float  # int | +/-inf
 
     def contains(self, other: LiveRange):
         """Is other entirely within self"""

torch/_meta_registrations.py

@@ -5373,7 +5373,7 @@ def meta__scaled_dot_product_flash_attention_for_cpu_backward(
     scale: Optional[float] = None,
 ):
     # cpus's grad layout is different from cuda's,
-    # i.e. (batch_size, seq_len，num_heads, head_dim）
+    # i.e. (batch_size, seq_len,num_heads, head_dim)
     batch_size = query.size(0)
     num_heads = query.size(1)
     head_dim = query.size(3)

torch/_numpy/_funcs_impl.py

@@ -2008,7 +2008,7 @@ def min_scalar_type(a: ArrayLike, /):
     from ._dtypes import DType
 
     if a.numel() > 1:
-        # numpy docs: "For non-scalar array a, returns the vector’s dtype unmodified."
+        # numpy docs: "For non-scalar array a, returns the vector's dtype unmodified."
         return DType(a.dtype)
 
     if a.dtype == torch.bool:

torch/_refs/__init__.py

@@ -485,7 +485,7 @@ def _make_alias(fn, name):
     """
     This function defines an alias of another function and sets its __name__ argument.
     It also sets its __module__ argument to the module of the caller.
-    Note that when naïvely doing `alias = fn`, we have that `alias.__name__ == "fn"`, and
+    Note that when naively doing `alias = fn`, we have that `alias.__name__ == "fn"`, and
     `alias.__module__ == fn.__module__`.
     """
 

torch/_refs/nn/functional/__init__.py

@@ -600,7 +600,7 @@ def margin_ranking_loss(
     margin: float = 0.0,
     reduction: str = "mean",
 ) -> TensorLikeType:
-    # loss_without_reduction = max(0, −target * (input1 − input2) + margin)
+    # loss_without_reduction = max(0, -target * (input1 - input2) + margin)
     if input1.ndim != input2.ndim or input1.ndim != target.ndim:
         raise RuntimeError(
             "margin_ranking_loss : All input tensors should have same dimension but got sizes: "