Fix typos in messages under torch (#89049)

This PR fixes typos of messages in `.py` files under torch directory.
Only in `torch/onnx/symbolic_opset16.py`, fix a typo in comment to make the operator name correct.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/89049
Approved by: https://github.com/lezcano

torch/_refs/nn/functional/__init__.py

@@ -595,7 +595,7 @@ def _nll_loss_nd(
 ) -> TensorLikeType:
     utils.check(
         input.ndim > 0 and input.ndim <= 3,
-        lambda: f"Expected input dimension to be either [1, 2, 3] but recieved {input.ndim}.",
+        lambda: f"Expected input dimension to be either [1, 2, 3] but received {input.ndim}.",
     )
 
     utils.check(

torch/ao/nn/intrinsic/qat/modules/linear_fused.py

@@ -35,7 +35,7 @@ class LinearBn1d(nn.modules.linear.Linear, nni._FusedModule):
                  freeze_bn=False,
                  qconfig=None):
         nn.modules.linear.Linear.__init__(self, in_features, out_features, bias)
-        assert qconfig, 'qconfig must be provded for QAT module'
+        assert qconfig, 'qconfig must be provided for QAT module'
         self.qconfig = qconfig
         self.freeze_bn = freeze_bn if self.training else True
         self.bn = nn.BatchNorm1d(out_features, eps, momentum, True, True)

torch/ao/quantization/fx/_model_report/model_report.py

@@ -385,7 +385,7 @@ class ModelReport:
                         module_fqns_to_features[module_fqn] = {**new_info, **present_info}
                     else:
                         error_str = "You have the same key with different values across detectors. "
-                        error_str += "Someone incorrectly implemented a detector with conflicting keys to exisiting detectors."
+                        error_str += "Someone incorrectly implemented a detector with conflicting keys to existing detectors."
                         raise ValueError(error_str)
                 else:
                     # we just set it

torch/ao/quantization/observer.py

@@ -1019,7 +1019,7 @@ class HistogramObserver(UniformQuantizationObserverBase):
         This follows the implementation of NormMinimization::NonlinearQuantizationParamsSearch in
         caffe2/quantization/server/norm_minimization.cc
         """
-        assert self.histogram.size()[0] == self.bins, "bins mistmatch"
+        assert self.histogram.size()[0] == self.bins, "bins mismatch"
         bin_width = (self.max_val - self.min_val) / self.bins
 
         # cumulative sum

torch/backends/xeon/run_cpu.py

@@ -598,7 +598,7 @@ def create_args(parser=None):
     _add_multi_instance_params(parser)
     # positional
     parser.add_argument("program", type=str,
-                        help="The full path to the proram/script to be launched. "
+                        help="The full path to the program/script to be launched. "
                              "followed by all the arguments for the script")
 
     # rest from the training program

torch/cuda/memory.py

@@ -61,7 +61,7 @@ def caching_allocator_alloc(size, device: Union[Device, int] = None, stream=None
     if not isinstance(stream, int):
         raise TypeError('Invalid type for stream argument, must be '
                         '`torch.cuda.Stream` or `int` representing a pointer '
-                        'to a exisiting stream')
+                        'to a existing stream')
     with torch.cuda.device(device):
         return torch._C._cuda_cudaCachingAllocator_raw_alloc(size, stream)
 

torch/distributed/benchmarks/benchmark_ddp_rpc.py

@@ -335,7 +335,7 @@ if __name__ == "__main__":
         "--embedding-dim",
         type=int,
         default=EMBEDDING_DIM,
-        help="Number of embedding dimentions.",
+        help="Number of embedding dimensions.",
     )
     parser.add_argument(
         "--warmup-cycles",

torch/distributed/elastic/multiprocessing/api.py

@@ -537,7 +537,7 @@ class MultiprocessContext(PContext):
         for proc in self._pc.processes:
             if proc.is_alive():
                 log.warning(
-                    f"Unable to shutdown process {proc.pid} via {death_sig}, forcefully exitting via {_get_kill_signal()}"
+                    f"Unable to shutdown process {proc.pid} via {death_sig}, forcefully exiting via {_get_kill_signal()}"
                 )
                 try:
                     os.kill(proc.pid, _get_kill_signal())
@@ -714,7 +714,7 @@ class SubprocessContext(PContext):
         for handler in self.subprocess_handlers.values():
             if handler.proc.poll() is None:
                 log.warning(
-                    f"Unable to shutdown process {handler.proc.pid} via {death_sig}, forcefully exitting via {_get_kill_signal()}"
+                    f"Unable to shutdown process {handler.proc.pid} via {death_sig}, forcefully exiting via {_get_kill_signal()}"
                 )
                 handler.close(death_sig=_get_kill_signal())
                 handler.proc.wait()

torch/distributed/elastic/rendezvous/etcd_rendezvous.py

@@ -293,7 +293,7 @@ class EtcdRendezvous(object):
                 time.sleep(1)
 
             except RendezvousTimeoutError:
-                log.info("Rendezvous timeout occured in EtcdRendezvousHandler")
+                log.info("Rendezvous timeout occurred in EtcdRendezvousHandler")
                 raise
 
             except RendezvousClosedError:

torch/distributions/mixture_same_family.py

@@ -60,7 +60,7 @@ class MixtureSameFamily(Distribution):
 
         if not isinstance(self._mixture_distribution, Categorical):
             raise ValueError(" The Mixture distribution needs to be an "
-                             " instance of torch.distribtutions.Categorical")
+                             " instance of torch.distributions.Categorical")
 
         if not isinstance(self._component_distribution, Distribution):
             raise ValueError("The Component distribution need to be an "

torch/fx/experimental/accelerator_partitioner.py

@@ -696,7 +696,7 @@ class Partitioner:
             return find_combination, partitions
 
         def reset_partition_in_sparse_nn(partition, new_partition=True):
-            """If crossing the boudary between non-embedding nodes and
+            """If crossing the boundary between non-embedding nodes and
             embedding nodes, create a new partition
             """
             if in_embedding_region:

torch/fx/experimental/graph_gradual_typechecker.py

@@ -184,7 +184,7 @@ def get_attr_inference_rule(n: Node, traced):
     if attr_name == "shape":
         n.type = Dyn
     else:
-        raise TypeError("Not yet implelemted")
+        raise TypeError("Not yet implemented")
 
     # TODO. We leave it like this till we add a type to represent tensor sizes
     return n.type
@@ -507,7 +507,7 @@ def flatten_check(tensor_type, start_dim, end_dim):
         new_type_list = lhs + mid + rhs
         return TensorType(tuple(new_type_list))
     else:
-        raise TypeError(f'Incompatable dimentions {start_dim}, {end_dim - 1} in type {tensor_type}')
+        raise TypeError(f'Incompatable dimensions {start_dim}, {end_dim - 1} in type {tensor_type}')
 
 @register_inference_rule(torch.flatten)
 def flatten_inference_rule(n: Node):

torch/fx/passes/split_module.py

@@ -28,8 +28,8 @@ class Partition:
             f" nodes: {self.node_names},\n"
             f" inputs: {self.inputs},\n"
             f" outputs: {self.outputs},\n"
-            f" partitions depenent on: {self.partitions_dependent_on},\n"
-            f" parition dependents: {self.partition_dependents}"
+            f" partitions dependent on: {self.partitions_dependent_on},\n"
+            f" partition dependents: {self.partition_dependents}"
         )
 
 

torch/jit/frontend.py

@@ -614,7 +614,7 @@ class StmtBuilder(Builder):
         else:
             raise NotSupportedError(
                 find_before(ctx, rhs.range().start, '=', offsets=(-1, 0)),
-                "unsupported kind of augumented assignment: " + op.__name__)
+                "unsupported kind of augmented assignment: " + op.__name__)
         return AugAssign(lhs, op_token, rhs)
 
     @staticmethod

torch/nn/utils/parametrize.py

@@ -242,7 +242,7 @@ class ParametrizationList(ModuleList):
                 if len(value) != self.ntensors:
                     raise ValueError(
                         "'right_inverse' must return a sequence of tensors of length "
-                        f"{self.ntensors}. Got a sequence of lenght {len(value)}."
+                        f"{self.ntensors}. Got a sequence of length {len(value)}."
                     )
                 for i, tensor in enumerate(value):
                     original_i = getattr(self, f"original{i}")

torch/onnx/symbolic_helper.py

@@ -1308,7 +1308,7 @@ def _index_fill_reshape_helper(g: jit_utils.GraphContext, self, dim, index):
         from torch.onnx.symbolic_opset11 import scatter  # type: ignore[no-redef]
 
     if self.type().dim() is None:
-        return _unimplemented("index_fill", "input rank not accesible")
+        return _unimplemented("index_fill", "input rank not accessible")
     self_dim = self.type().dim()
     dim_value = _parse_arg(dim, "i")
     unsqueezed_index = _unsqueeze_helper(

torch/onnx/symbolic_opset16.py

@@ -15,7 +15,7 @@ Updated operators:
     PRelu
     RoiAlign
     Scan
-    ScatterElemenets
+    ScatterElements
     ScatterND
     Where
     GreaterOrEqual

torch/profiler/_pattern_matcher.py

@@ -161,7 +161,7 @@ class ExtraCUDACopyPattern(Pattern):
     def __init__(self, prof: profile, should_benchmark: bool = False):
         super().__init__(prof, should_benchmark)
         self.name = "Extra CUDA Copy Pattern"
-        self.description = "Filled a CPU tensor and immediately moved it to GPU. Please initalize it on GPU."
+        self.description = "Filled a CPU tensor and immediately moved it to GPU. Please initialize it on GPU."
         self.url = "https://pytorch.org/tutorials/recipes/recipes/tuning_guide.html#create-tensors-directly-on-the-target-device"
         self.init_ops = {
             "aten::fill_", "aten::zero_", "aten::normal_", "aten::uniform_"

torch/serialization.py

@@ -773,7 +773,7 @@ def load(
 
     if weights_only:
         if pickle_module is not None:
-            raise RuntimeError("Can not safely load weights when expiclit picke_module is specified")
+            raise RuntimeError("Can not safely load weights when explicit picke_module is specified")
     else:
         if pickle_module is None:
             pickle_module = pickle

torch/testing/_internal/common_distributed.py

@@ -333,7 +333,7 @@ def skip_if_rocm(func):
 def skip_if_win32():
     return sandcastle_skip_if(
         sys.platform == "win32",
-        "This unit test case is not supportted on Windows platform",
+        "This unit test case is not supported on Windows platform",
     )
 
 

torch/testing/_internal/composite_compliance.py

@@ -311,7 +311,7 @@ def generate_subclass_choices_args_kwargs(args, kwargs, CCT, cct_mode):
 
 def raise_composite_compliance_error(err, additional_info=''):
     raise RuntimeError(
-        "Composite compilance check failed with "
+        "Composite compliance check failed with "
         "the above error.\n"
         f"{additional_info}"
         "If you are adding an OpInfo of an "

torch/utils/benchmark/examples/fuzzer.py

@@ -65,7 +65,7 @@ def main():
     print()
 
     # More string munging to make pretty output.
-    print(f"Average attemts per valid config: {1. / (1. - add_fuzzer.rejection_rate):.1f}")
+    print(f"Average attempts per valid config: {1. / (1. - add_fuzzer.rejection_rate):.1f}")
 
     def time_fn(m):
         return m.median / m.metadata["numel"]

torch/utils/benchmark/examples/sparse/fuzzer.py

@@ -80,7 +80,7 @@ def main():
     print()
 
     # More string munging to make pretty output.
-    print(f"Average attemts per valid config: {1. / (1. - add_fuzzer.rejection_rate):.1f}")
+    print(f"Average attempts per valid config: {1. / (1. - add_fuzzer.rejection_rate):.1f}")
 
     def time_fn(m):
         return m.mean / m.metadata["nnz"]

torch/utils/data/datapipes/dataframe/dataframes.py

@@ -408,7 +408,7 @@ class CaptureDataFrameWithDataPipeOps(CaptureDataFrame):
 
     def __getattr__(self, attrname):  # ?
         if attrname in UNIMPLEMENTED_ATTR:
-            raise AttributeError('Attemping to get ', attrname)
+            raise AttributeError('Attempting to get ', attrname)
         if attrname in DATAPIPES_OPS:
             return (self.as_datapipe()).__getattr__(attrname)
         return super().__getattr__(attrname)

torch/utils/data/datapipes/iter/callable.py

@@ -155,7 +155,7 @@ def _collate_helper(conversion, item):
                 import torcharrow.pytorch as tap  # type: ignore[import]
                 collation_fn = tap.rec.Default()
             except Exception:
-                raise Exception("unable to import default collation function from the TorchArrrow")
+                raise Exception("unable to import default collation function from the TorchArrow")
 
         tuple_names.append(str(name))
         value = collation_fn(df[name])