[BE][2/16] fix typos in torch/ (torch/_*/) (pytorch#156312)

Pull Request resolved: pytorch#156312
Approved by: https://github.com/albanD

Author: XuehaiPan

.lintrunner.toml

@@ -1169,7 +1169,6 @@ exclude_patterns = [
     'aten/src/ATen/[a-mA-M]*/**',
     'test/**',
     'test/[a-hA-h]*/**',
-    'torch/_*/**',
     'torch/distributed/tensor/**',
 ]
 init_command = [

test/cpp/jit/test_custom_class_registrations.cpp

@@ -371,12 +371,12 @@ struct ElementwiseInterpreter : torch::CustomClassHolder {
     return environment.at(*output_name_);
   }
 
-  // Ser/De infrastructure. See
+  // SerDe infrastructure. See
   // https://pytorch.org/tutorials/advanced/torch_script_custom_classes.html#defining-serialization-deserialization-methods-for-custom-c-classes
   // for more info.
 
   // This is the type we will use to marshall information on disk during
-  // ser/de. It is a simple tuple composed of primitive types and simple
+  // SerDe. It is a simple tuple composed of primitive types and simple
   // collection types like vector, optional, and dict.
   using SerializationType = std::tuple<
       std::vector<std::string> /*input_names_*/,

test/export/test_export.py

@@ -487,7 +487,7 @@ def _is_tensor_leaf(x):
             eps = [ep]
             if test_serdes:
                 # test dynamic shapes serialization
-                # test that behavior remains the same when exporting with ser/des specs:
+                # test that behavior remains the same when exporting with SerDes specs:
                 # serialize + deserialize original specs, and export.
                 ep_serdes = export(
                     model,
@@ -5011,7 +5011,7 @@ def forward(self, x):
         # There should be nonzero view nodes in the graph
         self.assertTrue(view_count > 0)
 
-    @testing.expectedFailureCppSerDes  # cpp ser/der not handling complicated symbols
+    @testing.expectedFailureCppSerDes  # cpp SerDer not handling complicated symbols
     def test_solver_unsupported_sympy_function(self):
         # repro of https://github.com/pytorch/pytorch/issues/131897
 

test/test_fx.py

@@ -954,7 +954,7 @@ def __init__(self, interpreter):
         script_out = scripted_lowered(x)
         torch.testing.assert_close(script_out, ref_out)
 
-        # Test TorchScript ser/de
+        # Test TorchScript SerDe
         import_copy = self.getExportImportCopy(scripted_lowered)
         imported_out = import_copy(x)
         torch.testing.assert_close(imported_out, ref_out)

test/test_unary_ufuncs.py

@@ -1104,7 +1104,7 @@ def test_silu_complex(self, device, dtype):
             self.assertEqual(res.real, out.real, atol=atol, rtol=rtol)
             self.assertEqual(res.imag, out.imag, atol=atol, rtol=rtol)
 
-    # It is not obvious how to merge this into OpInfo becuase these inputs
+    # It is not obvious how to merge this into OpInfo because these inputs
     # succeed for gradcheck but are expected to fail for gradgradcheck
     @dtypes(torch.double)
     def test_sinc(self, device, dtype):

torch/_decomp/decompositions.py

@@ -52,7 +52,7 @@ class Reduction(Enum):
 
 
 # This wraps a decomposition and performs various type promotion logic within it, depending on the strategy provided
-# We're currently re-using ELEMENTWISE_TYPE_PROMOTION_KIND, although some of the usages are on non-elementwise ops
+# We're currently reusing ELEMENTWISE_TYPE_PROMOTION_KIND, although some of the usages are on non-elementwise ops
 # Will need to validate the non-elementwise uses
 def type_casts(
     f: Callable,
@@ -947,7 +947,7 @@ def check_positive(param, param_name, strict=True):
     )
     torch._check(
         all(c > 0 for c in output_size),
-        lambda: f"Given an input with spacial size {tuple(shape[-2:])}, "
+        lambda: f"Given an input with spatial size {tuple(shape[-2:])}, "
         f"kernel_size={kernel_size}, dilation={dilation}, "
         f"padding={padding}, stride={stride}, "
         "the calculated shape of the array of sliding blocks "
@@ -4046,7 +4046,7 @@ def nll_loss2d_forward(
     return _nll_loss_forward(self, target, weight, reduction, ignore_index)
 
 
-# These are adapted from aten/src/ATen/native/UpSample.h, wich is based on
+# These are adapted from aten/src/ATen/native/UpSample.h, which is based on
 # https://en.wikipedia.org/wiki/Bicubic_interpolation#Bicubic_convolution_algorithm
 def _upsample_cubic_convolution1(x: Tensor, A: float) -> Tensor:
     return ((A + 2) * x - (A + 3)) * x * x + 1

torch/_dynamo/convert_frame.py

@@ -1063,7 +1063,7 @@ def format_func_info(code: CodeType) -> str:
                 return f"'{code.co_name}' ({code.co_filename}:{code.co_firstlineno})"
 
             # NS: Don't add period at the end of string, as it'll be added to URL
-            # renderring it incorrect
+            # rendering it incorrect
             log.warning(
                 "torch._dynamo hit config.%s (%s)\n"
                 "   function: %s\n"

torch/_dynamo/output_graph.py

@@ -347,7 +347,7 @@ class StackLocalsMetadata:
 
 def get_builtins_dict(global_scope):
     # f_globals["__builtins__"] can be a dict or a module. This is an
-    # implemenation detail -
+    # implementation detail -
     # https://docs.python.org/3/library/builtins.html.
 
     # This makes guarding on any builtin messy because the guard check_fn

torch/_dynamo/variables/builder.py

@@ -1662,13 +1662,13 @@ def wrap_listlike(self, value: Union[tuple, list, odict_values, NamedTuple]):
             # <==> variable tracker" 1-to-1 mapping, which is mainly handled via
             # `side_effects`. Note that constructing `tensor_variable` above
             # already adds it to graph arg, but we never registered it with
-            # `side_effects`. The pre-emptive `realize` calls here basically
+            # `side_effects`. The preemptive `realize` calls here basically
             # does that registration (at the end of `self.__call__`).
             #
             # A slightly cleaner alternative is to register the
             # `tensor_variable`s above with `side_effects` directly, and just
             # return the `list_variable`, but that breaks some tensor-subclass
-            # releated tests like `test_inputs_aliasing_bytecode_stack_restore`,
+            # related tests like `test_inputs_aliasing_bytecode_stack_restore`,
             # because `tensor_variable` is constructed via
             # `handle_traced_output`, which doesn't really expect/handle tensor
             # subclass.

torch/_export/converter.py

@@ -134,7 +134,7 @@ def execute_subgraph_from_prim_loop(
 ):
     """
     subgraph: GraphModule from sub-block.
-    iter_idx: The index of interation.
+    iter_idx: The index of interaction.
     len_loop_local_arguments: The number of loop local arguments in args.
     """
 
@@ -810,7 +810,7 @@ def convert_call_function_op(self, node: torch._C.Node):
 
         fx_node = self.fx_graph.call_function(target, args, kwargs)
 
-        # TODO: covnert sourceRange() into stack_trace
+        # TODO: convert sourceRange() into stack_trace
         # fx_node.meta["stack_trace"] = node.sourceRange()
 
         if node.outputsSize() == 1:
@@ -883,7 +883,7 @@ def convert_aten_Int(self, node: torch._C.Node):
             torch.ops.aten._local_scalar_dense.default, (to_copy_node,)
         )
 
-        # TODO: covnert sourceRange() into stack_trace
+        # TODO: convert sourceRange() into stack_trace
         # fx_node.meta["stack_trace"] = node.sourceRange()
 
         output_name = node.output().debugName()
@@ -942,7 +942,7 @@ def convert_aten_div(self, node: torch._C.Node):
                         kwargs,
                     )
 
-                    # TODO: covnert sourceRange() into stack_trace
+                    # TODO: convert sourceRange() into stack_trace
                     # fx_node.meta["stack_trace"] = node.sourceRange()
 
                     output_name = node.output().debugName()
@@ -1006,7 +1006,7 @@ def convert_aten_add(self, node: torch._C.Node):
             ):
                 target = torch.ops.aten.add.t
             else:
-                raise RuntimeError(f"unable to determind the target for {node}")
+                raise RuntimeError(f"unable to determined the target for {node}")
         else:
             target = get_op_overload(node)
 
@@ -1565,7 +1565,7 @@ def lift_get_attr(self):
         #
         # This function should happen in TS2EPConverter instead of
         # TS2FXGraphConverter since it gets attributes from self.ts_model
-        # which is not accessable in TS2FXGraphConverter. It is similar to where
+        # which is not accessible in TS2FXGraphConverter. It is similar to where
         # we collect self.name_to_param and self.name_to_buffer.
         name_to_attribute_fqn: dict[str, str] = {}