docs: fix typos (#164879)

Correct typos in the comments

Pull Request resolved: https://github.com/pytorch/pytorch/pull/164879
Approved by: https://github.com/Lucaskabela, https://github.com/mlazos, https://github.com/cyyever

test/jit/test_cuda.py

@@ -522,7 +522,7 @@ class TestCUDA(JitTestCase):
         self.assertGreater(test_event(), 0)
 
         # Check for stream synchronization , when a large tensor multiplication is
-        # computed on the stream. The stream.query should be true once the synchroniztion is done
+        # computed on the stream. The stream.query should be true once the synchronization is done
         @torch.jit.script
         def test_stream_synchronize() -> float:
             device_index = torch.cuda.current_device()

test/test_dynamic_shapes.py

@@ -3662,7 +3662,7 @@ def forward(self, arg0_1: "i64[2][1]cpu", arg1_1: "Sym(u2)", arg2_1: "Sym(u3)",
         self.assertEqual(result_compiled, result_eager)
         self.assertEqual(cnt.frame_count, 1)
 
-        # Pass a contiguous tensor. A recompilation will happen due to 0/1 speciialization on stride.
+        # Pass a contiguous tensor. A recompilation will happen due to 0/1 specialization on stride.
         log_stream, ctx = logs_to_string(
             "torch._functorch._aot_autograd.graph_capture", "aot_graphs"
         )

test/test_linalg.py

@@ -5661,7 +5661,7 @@ class TestLinalg(TestCase):
 
             # TunableOp is running in a subprocess
             # online tuning needs filename set through API
-            # offline tuning needs filename set through environment variableq
+            # offline tuning needs filename set through environment variable
             result_filename = torch.cuda.tunable.get_filename()
             untuned_filename = get_tunableop_untuned_filename()
 

test/torch_np/numpy_tests/core/test_dtype.py

@@ -307,7 +307,7 @@ class TestPromotion(TestCase):
     )
     def test_permutations_do_not_influence_result(self, dtypes, expected):
         # Tests that most permutations do not influence the result.  In the
-        # above some uint and int combintations promote to a larger integer
+        # above some uint and int combinations promote to a larger integer
         # type, which would then promote to a larger than necessary float.
         for perm in permutations(dtypes):
             assert np.result_type(*perm) == expected

torch/_dynamo/guards.py

@@ -3664,7 +3664,7 @@ class CheckFunctionManager:
             # increase in compile time. We first do a cache flush to measure the
             # guard latency more accurately. This cache flush is expensive.
             # Note  - If you are working on a guard optimization, it might be a
-            # good idea to increase this number for more stabiilty during
+            # good idea to increase this number for more stability during
             # development.
             latency = profile_guard_manager(
                 self.guard_manager.root, output_graph.local_scope, 1

torch/_dynamo/variables/torch.py

@@ -1619,7 +1619,7 @@ For now, dynamo will explicitly graph break when it encounters user code with th
             # take the conservative approach to graph break on size changes, and
             # assume other cases can fall through soundly.
             #
-            # Note that although these tensor variablels would hold different
+            # Note that although these tensor variables would hold different
             # proxies, the in-place mutation semantics is preserved in the FX
             # graph, so we won't have correctness issues.
             if isinstance(saved_out_shapes, list):

torch/_export/serde/schema.py

@@ -156,7 +156,7 @@ class TokenArgument:
 
 # This is use for storing the contents of a list which contain optional tensors
 # (Tensor?[], ex. [Tensor, None, ...]), where the list will be serialized to the
-# type List[OptionalTensorArgument], with tensor values seiralized to the
+# type List[OptionalTensorArgument], with tensor values serialized to the
 # "as_tensor" field, and None values serialized to the "as_none" field.
 @_union_dataclass
 class OptionalTensorArgument(_Union):

torch/_higher_order_ops/while_loop.py

@@ -692,7 +692,7 @@ class WhileLoopStackOutputOp(HigherOrderOperator):
 #
 # idx = 0
 # init_grad_carries = grads
-# init_grad_additional_inputs = torch.zeros_like(g_additioanl_inputs)
+# init_grad_additional_inputs = torch.zeros_like(g_additional_inputs)
 # fw_inps = torch.cat([ctx.fw_carried_inputs, fw_outputs[:-1]])
 # while_loop(cond_fn, body_fn, (idx, init_grad_carries, init_grad_additional_inputs,), (fw_additional_inputs, fw_inps))
 

torch/nn/utils/_expanded_weights/expanded_weights_impl.py

@@ -100,7 +100,7 @@ def implements_per_sample_grads(torch_function):
 # This is a __torch_function__ object but it could have also been a Tensor Extension
 # with a dispatch key.
 #
-# Needs to be a tensor subclass to allow reparamaterization
+# Needs to be a tensor subclass to allow reparameterization
 class ExpandedWeight(torch.Tensor):
     def __init__(self, orig_weight, batch_size, loss_reduction):
         self.batch_size = batch_size

torch/testing/_internal/distributed/_tensor/common_dtensor.py

@@ -427,7 +427,7 @@ class DTensorTestBase(MultiProcessTestCase):
 
         # For nccl backend, bind the device to the process if device_id is not None
         # so the nccl communicator is immediately formed and we can use `ncclCommSplit`
-        # for form subgroup to avoid unnecesssary overhead.
+        # for form subgroup to avoid unnecessary overhead.
         dist.init_process_group(
             backend=backend,
             world_size=self.world_size,