[BE][9/16] fix typos in torch/ (torch/csrc/) (#156319)

Pull Request resolved: https://github.com/pytorch/pytorch/pull/156319
Approved by: https://github.com/albanD
ghstack dependencies: #156313, #156314, #156315, #156316, #156317

.lintrunner.toml

@@ -1179,7 +1179,6 @@ exclude_patterns = [
     'torch/utils/**',
     'torch/csrc/jit/**',
     'torch/csrc/jit/[a-o]*/**',
-    'torch/csrc/[a-i]*/**',
     'torch/csrc/distributed/**',
 ]
 init_command = [

torch/csrc/api/include/torch/data/dataloader/stateless.h

@@ -15,7 +15,7 @@ namespace torch::data {
 /// A dataloader for stateless datasets.
 ///
 /// This dataloader follows the traditional PyTorch dataloader design, whereby a
-/// (posssibly) stateful sampler produces *batch requests* for a stateless
+/// (possibly) stateful sampler produces *batch requests* for a stateless
 /// dataset, which acts as a simple batch request to batch mapping. The batch
 /// request will often be an array of indices, and if the dataset is a simple
 /// image dataset, the dataset would produce the images at those indices.

torch/csrc/api/include/torch/data/datasets/chunk.h

@@ -234,7 +234,7 @@ class BatchDataBuffer {
   // NOLINTNEXTLINE(cppcoreguidelines-avoid-const-or-ref-data-members)
   ExampleSampler& example_sampler_;
 
-  // configurable maximun number of elements the queue can hold at one time.
+  // configurable maximum number of elements the queue can hold at one time.
   size_t queue_capacity_;
 
   // When set to true, it wakes the writer threads from the wait and exit
@@ -286,7 +286,7 @@ struct ChunkDatasetOptions {
   /// The capacity of the queue for batch caching.
   TORCH_ARG(size_t, cache_size) = 2048;
 
-  // The number of chunks to perfrom cross-chunk shuffling. Default to 1 meaning
+  // The number of chunks to perform cross-chunk shuffling. Default to 1 meaning
   // no cross-chunk shuffling. When it is equal to n (n > 1), n random
   // chunks will be loaded at once and example shuffling will be performed
   // across all those n chunks.
@@ -303,9 +303,10 @@ struct ChunkDatasetOptions {
 ///
 /// Unlike regular dataset, chunk dataset require two samplers to operate and
 /// keeps an internal state. `ChunkSampler` selects, which chunk to load next,
-/// while the `ExampleSampler` determins the order of Examples that are returned
-/// in each `get_batch` call. The hierarchical sampling approach used here is
-/// inspired by this paper http://martin.zinkevich.org/publications/nips2010.pdf
+/// while the `ExampleSampler` determines the order of Examples that are
+/// returned in each `get_batch` call. The hierarchical sampling approach used
+/// here is inspired by this paper
+/// http://martin.zinkevich.org/publications/nips2010.pdf
 template <
     typename ChunkReader,
     typename ChunkSampler = samplers::RandomSampler,
@@ -346,7 +347,7 @@ class ChunkDataset final
   }
 
   /// Default get_batch method of BatchDataset. This method returns
-  /// Example batches created from the preloaded chunks. The implemenation
+  /// Example batches created from the preloaded chunks. The implementation
   /// is dataset agnostic and does not need overriding in different chunk
   /// datasets.
   BatchType get_batch(size_t batch_size) override {

torch/csrc/api/include/torch/data/samplers/base.h

@@ -24,7 +24,7 @@ class Sampler {
 
   /// Resets the `Sampler`'s internal state.
   /// Typically called before a new epoch.
-  /// Optionally, accepts a new size when reseting the sampler.
+  /// Optionally, accepts a new size when resetting the sampler.
   virtual void reset(std::optional<size_t> new_size) = 0;
 
   /// Returns the next index if possible, or an empty optional if the

torch/csrc/api/include/torch/nn/functional/activation.h

@@ -344,7 +344,7 @@ namespace detail {
 inline Tensor glu(const Tensor& input, int64_t dim) {
   TORCH_CHECK(
       input.dim() != 0,
-      "glu does not suppport scalars because halving size must be even");
+      "glu does not support scalars because halving size must be even");
   return torch::glu(input, dim);
 }
 } // namespace detail

torch/csrc/api/include/torch/nn/modules/container/moduledict.h

@@ -130,7 +130,7 @@ class ModuleDictImpl : public Cloneable<ModuleDictImpl> {
     return modules_.is_empty();
   }
 
-  /// Check if the centain parameter with the key in the `ModuleDict`.
+  /// Check if the certain parameter with the key in the `ModuleDict`.
   bool contains(const std::string& key) const noexcept {
     return modules_.contains(key);
   }

torch/csrc/api/include/torch/nn/modules/container/parameterdict.h

@@ -107,7 +107,7 @@ class ParameterDictImpl : public Cloneable<ParameterDictImpl> {
     parameters_.clear();
   }
 
-  /// Check if the centain parameter with the key in the ParameterDict
+  /// Check if the certain parameter with the key in the ParameterDict
   bool contains(const std::string& key) const noexcept {
     return parameters_.contains(key);
   }

torch/csrc/api/include/torch/serialize/input-archive.h

@@ -101,7 +101,7 @@ class TORCH_API InputArchive final {
   std::vector<std::string> keys();
 
   /// Forwards all arguments to `read()`.
-  /// Useful for generic code that can be re-used for both `InputArchive` and
+  /// Useful for generic code that can be reused for both `InputArchive` and
   /// `OutputArchive` (where `operator()` forwards to `write()`).
   template <typename... Ts>
   void operator()(Ts&&... ts) {

torch/csrc/api/include/torch/serialize/output-archive.h

@@ -66,7 +66,7 @@ class TORCH_API OutputArchive final {
   void save_to(const std::function<size_t(const void*, size_t)>& func);
 
   /// Forwards all arguments to `write()`.
-  /// Useful for generic code that can be re-used for both `OutputArchive` and
+  /// Useful for generic code that can be reused for both `OutputArchive` and
   /// `InputArchive` (where `operator()` forwards to `read()`).
   template <typename... Ts>
   void operator()(Ts&&... ts) {

torch/csrc/api/src/nn/modules/transformer.cpp

@@ -19,7 +19,7 @@ TransformerEncoderLayerImpl::TransformerEncoderLayerImpl(
 
 void TransformerEncoderLayerImpl::reset() {
   // NOTE: reset() is for initializing the model only, calling reset() after the
-  // model is created will throw exceptionss. Call reset_parameter() if the
+  // model is created will throw exceptions. Call reset_parameter() if the
   // created model needs a reset
 
   self_attn = this->register_module(

torch/csrc/autograd/FunctionsManual.cpp

@@ -2904,7 +2904,7 @@ Tensor softplus_double_backward(
 //   4. Return the as_strided view of the storage tensor using input geometry.
 //
 // See NOTE [ Detecting Memory Overlap Within A Strided Tensor ] on how to
-// roughly detech overlapping memory.
+// roughly detect overlapping memory.
 
 // NOTE [ Detecting Memory Overlap Within A Strided Tensor ]
 //
@@ -2994,7 +2994,7 @@ Tensor softplus_double_backward(
 //              Now that we established the above claim (***), we consider the
 //              view operation as first sorting the dimensions (i.e., blocks),
 //              apply the original view (since it only cares dimensions being
-//              consecutive and contiguous withtin each block), and then undo
+//              consecutive and contiguous within each block), and then undo
 //              the sort.
 //
 //              Consider a single block B in the output,
@@ -3046,7 +3046,7 @@ Tensor softplus_double_backward(
 //                  size'[i] <= floor(size[i] / k)
 //
 //              If size'[i] = 1, invariant is obviously satisfied as we are
-//              just removing a dimension (afte step (1)).
+//              just removing a dimension (after step (1)).
 //
 //              Assume size'[i] > 1.
 //
@@ -5244,7 +5244,7 @@ bool any_variable_defined(const variable_list& variables) {
 // Derivations for the householder_product.backward method.
 //
 // Given a sequence of vectors v_1, ..., v_n and a sequence of scalars tau_1,
-// ..., tau_k, the torch.linalg.householder_product computes the firt n columns
+// ..., tau_k, the torch.linalg.householder_product computes the first n columns
 // of the following product: Q = (I - tau_1 v_1 v_1^H) ... (I - tau_k v_k
 // v_k^H). Let
 //     H_i(sigma) := I - sigma v_i v_i^H, so Q = (H_1(sigma_1) ...
@@ -5648,7 +5648,7 @@ std::tuple<Tensor, Tensor, Tensor> ormqr_backward(
       // left = false and transpose = true is very much similar with just
       // transposed arguments passed into householder_product_backward.
       // Ormqr computes B = H_1 * ... * H_k * A.
-      // The sensivity wrt H_i is given by (see notes in
+      // The sensitivity wrt H_i is given by (see notes in
       // householder_product_backward) Tr(H_i_plus B B_grad^H H_i_minus dH_i),
       // so, since householder_product_backward respects `for i in range(k)`, we
       // could reuse householder_product_backward with

torch/csrc/autograd/TraceTypeManual.cpp

@@ -278,7 +278,7 @@ static void general_trace_function(
           tracer::addOutput(node, iter->toTensorList());
         } else {
           throw std::runtime_error(
-              "unsupported ouptut list type: " + elem_type->str());
+              "unsupported output list type: " + elem_type->str());
         }
       } else if (type->kind() == TypeKind::ClassType) {
         AT_ASSERT(iter->isObject());

torch/csrc/autograd/anomaly_mode.h

@@ -30,7 +30,7 @@ struct TORCH_API AnomalyMode {
 ///
 /// Anomaly detection mode is useful for debugging problems happening
 /// in the backward, such as unexpectedly modified tensors or NaNs
-/// occuring in the backward.
+/// occurring in the backward.
 ///
 /// The enabling of anomaly mode is global - as soon as there is one
 /// such guard, it is enabled for all computation and threads. It also

torch/csrc/autograd/autograd_meta.cpp

@@ -53,7 +53,7 @@ using at::Tensor;
 //
 // This layout constraint is ensured in the `set_fw_grad` function below
 
-// More complex cases arrise when non-dual Tensor interact with dual Tensors.
+// More complex cases arise when non-dual Tensor interact with dual Tensors.
 // The two most important cases are:
 //
 //     # Have:
@@ -222,7 +222,7 @@ void AutogradMeta::set_fw_grad(
           if (utils::has_same_meta(new_grad, base) &&
               utils::has_same_meta(new_grad, self)) {
             // TODO extend this special case to when the underlying storage of
-            // new_grad can be re-used.
+            // new_grad can be reused.
             new_base_fw_grad = new_grad;
           } else {
             new_base_fw_grad =

torch/csrc/autograd/engine.cpp

@@ -611,7 +611,7 @@ auto Engine::thread_main(const std::shared_ptr<GraphTask>& graph_task) -> void {
   }
 }
 
-// Reentrant call will re-use the graph_task's owner thread ready_queue for
+// Reentrant call will reuse the graph_task's owner thread ready_queue for
 // queueing tasks (NOTE: this is not true in the async_mode of the engine).
 // While we can create separate ready queue for each new reentrant
 // thread, but sharing the same cpu_ready_queue with parent thread is a
@@ -1228,7 +1228,7 @@ void Engine::evaluate_function(
 }
 
 static uint64_t compute_min_topological_nr(const edge_list& outputs) {
-  // Computes the mininum topological number among all the outputs
+  // Computes the minimum topological number among all the outputs
   if (outputs.empty()) {
     return 0;
   }

torch/csrc/autograd/forward_grad.h

@@ -27,7 +27,7 @@ struct ForwardGrad;
 //   - Ensure that we can keep the level that we expose to the user API simple
 //   (an integer
 //     that represents the nesting depth) while avoiding confusions when the
-//     level index is re-used.
+//     level index is reused.
 
 // The important external APIs from this file are:
 //   - ForwardADLevel::get_next_idx() that can be used to enter a new level and

torch/csrc/autograd/function.h

@@ -67,7 +67,7 @@ TORCH_API std::shared_ptr<Node> get_current_node();
 // or more input `Variable`s and producing zero or more output `Variable`s. All
 // functions in PyTorch's autograd machinery derive from this class and
 // override its `apply` method. Instances of such subclasses will then be
-// invokable via the call operator.
+// invocable via the call operator.
 //
 //                    Nodes in the Autograd Graph
 //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -592,7 +592,7 @@ struct TORCH_API Node : std::enable_shared_from_this<Node> {
   //   1) Extract tensors/symint args
   //   2) Collect node information for specialization and caching
   // Implementations in subclasses should call args.collect() with all node
-  // attrs. These functions are only called durring backward.
+  // attrs. These functions are only called during backward.
   virtual void compiled_args(CompiledNodeArgs& args) const {
     TORCH_CHECK_NOT_IMPLEMENTED(
         false, std::string("compiled_args not implemented: ") + name());

torch/csrc/autograd/functions/basic_ops.cpp

@@ -21,7 +21,7 @@ variable_list Error::apply(variable_list&& inputs) const {
 }
 
 void Error::compiled_args(CompiledNodeArgs& args) const {
-  // throw the error durring collect, the graph won't get compiled
+  // throw the error during collect, the graph won't get compiled
   apply(variable_list());
 }
 

torch/csrc/autograd/init.cpp

@@ -605,7 +605,7 @@ static PyObject* set_autocast_enabled(
   HANDLE_TH_ERRORS
   static PythonArgParser parser(
       {"set_autocast_enabled(std::string_view device_type, bool enabled)",
-       "set_autocast_enabled(bool enabled)"}); // this signature is depracated.
+       "set_autocast_enabled(bool enabled)"}); // this signature is deprecated.
   ParsedArgs<2> parsed_args;
   auto r = parser.parse(args, kwargs, parsed_args);
   // Set at::kCUDA as default value to prevent BC-breaking changes.
@@ -628,7 +628,7 @@ static PyObject* is_autocast_enabled(
   HANDLE_TH_ERRORS
   static PythonArgParser parser(
       {"is_autocast_enabled(std::string_view device_type)",
-       "is_autocast_enabled()"}); // this signature is depracated.
+       "is_autocast_enabled()"}); // this signature is deprecated.
   ParsedArgs<1> parsed_args;
   auto r = parser.parse(args, kwargs, parsed_args);
   // Set at::kCUDA as default value to prevent BC-breaking changes.

torch/csrc/autograd/profiler_kineto.cpp

@@ -622,7 +622,7 @@ void prepareProfiler(
     /*
      * Sending a warning and passing the non-standard event to the backend
      * Backend can abort if the event is not supported.
-     * TODO Should we gracefully drop the invalid event if we have atleast one
+     * TODO Should we gracefully drop the invalid event if we have at least one
      * valid?
      */
     auto is_standard_event = [](const std::string& event) -> bool {

torch/csrc/cuda/CUDAPluggableAllocator.h

@@ -186,7 +186,7 @@ struct TORCH_CUDA_CPP_API CUDAPluggableAllocator
   std::function<void(int, c10::cuda::MempoolId_t)> end_allocate_to_pool_fn_;
   std::function<void(int, c10::cuda::MempoolId_t)> relase_pool_fn_;
   std::mutex allocator_mutex_;
-  // We do the bookeeping here in order to simplify custom allocators
+  // We do the bookkeeping here in order to simplify custom allocators
   std::unordered_map<void*, _AllocationMetadata> allocation_metadata_;
 
   bool initialized_ = false;

torch/csrc/cuda/GdsFile.cpp

@@ -21,7 +21,7 @@ std::string cuGDSFileGetErrorString(T status) {
                                : std::string(c10::utils::str_error(errno));
 }
 
-// To get error message for Buf/Handle registeration APIs that return
+// To get error message for Buf/Handle registration APIs that return
 // CUfileError_t
 template <
     class T,

torch/csrc/deploy/README.md

@@ -1,2 +1,2 @@
-# torch::deploy has been moved to pytorch/multipy
-Please check out [https://github.com/pytorch/multipy](https://github.com/pytorch/multipy) to find the new home for torch::deploy.
+# torch::deploy has been moved to pytorch/multipy <!-- codespell:ignore -->
+Please check out [https://github.com/pytorch/multipy](https://github.com/pytorch/multipy) to find the new home for torch::deploy. <!-- codespell:ignore -->

torch/csrc/dynamo/compiled_autograd.h

@@ -98,7 +98,7 @@ struct TORCH_API PyCompilerGuard {
 // including torch/csrc/autograd/engine.h breaks BC by somehow introducing
 // symbol resolution issues. Instead requiring downstream users to include
 // engine.h to access collect_input_metadata, we provide it here (with a
-// different name to avoid ambigous symbols...)
+// different name to avoid ambiguous symbols...)
 TORCH_API std::vector<std::optional<InputMetadata>> get_input_metadata(
     const edge_list& edges);
 
@@ -1068,7 +1068,7 @@ class SwapSavedVariables {
 // (e.g. MulBackward0_apply_functional). Compiled Autograd's initial graph
 // capture wants to take a variant of this function and proxy it into the graph.
 // Every autograd node defines an apply_with_saved function, that when invoked,
-// proxys a call to a function into the Compiled Autograd graph.
+// proxies a call to a function into the Compiled Autograd graph.
 //
 // Some requirements that we have are:
 // - The proxy'ed function must have inputs that are FX-graphable types.

torch/csrc/dynamo/eval_frame_cpp.cpp

@@ -274,7 +274,7 @@ PyObject* dynamo__custom_eval_frame(
     // NB: We could use extract_cache_entry to get the cache_entry, but
     // extract_cache_entry returns a borrowed reference. Modifying a borrowed
     // reference seems wrong. Therefore, we directly access the
-    // extra->cache_entry. extra wont be NULL here.
+    // extra->cache_entry. extra won't be NULL here.
     CacheEntry* new_cache_entry =
         create_cache_entry(extra, guarded_code, backend);
 

torch/csrc/dynamo/extra_state.h

@@ -132,7 +132,7 @@ void destroy_extra_state(void* obj);
 // Clears the existing object sitting on the extra scratch spance and sets it
 // up with the new state. Note that _PyCode_SetExtra calls the
 // destroy_extra_state deleter internally, and therefore we don't call it
-// explicity here.
+// explicitly here.
 
 // Ownership contract
 // args
@@ -148,7 +148,7 @@ void destroy_extra_state(void* obj);
 // scratch space.
 void set_extra_state(PyCodeObject* code, ExtraState* extra_state);
 
-// Creates a new extra state and put it on the extra scrach space of the code
+// Creates a new extra state and put it on the extra scratch space of the code
 // object.
 
 // Ownership contract

torch/csrc/dynamo/guards.cpp

@@ -60,7 +60,7 @@ typedef struct {
   PyTupleObject* it_seq; /* Set to NULL when iterator is exhausted */
 } _PyTupleIterObject;
 
-// Copied from CPython, and given a unified name for different Python verions.
+// Copied from CPython, and given a unified name for different Python versions.
 // https://github.com/python/cpython/blob/7f71003b222ad398713514c2b55d34dc05dba6bc/Objects/rangeobject.c#L765-L771
 typedef struct {
   PyObject_HEAD
@@ -124,7 +124,7 @@ TensorCheck::TensorCheck(
 // See note in guards.py [Note - On Export Tensor Guards]
 // Logic parallel to here must be maintained in python
 bool TensorCheck::check(const LocalState& state, const at::Tensor& v) {
-  // In terms of a sparse_csr tensor, it does not support strides informatio
+  // In terms of a sparse_csr tensor, it does not support strides information
   c10::SymIntArrayRef sym_strides(std::vector<SymInt>(v.ndimension(), -1));
   bool does_not_support_stride = v.layout() == c10::kSparseCsr ||
       v.layout() == c10::kSparseCsc || v.layout() == c10::kSparseBsc ||
@@ -2407,7 +2407,7 @@ class GuardAccessor {
  * value passed to the check function to call the check function of the child
  * guard manager.
  *
- * Performace optimization for fail fast - An optimization for runtime here is
+ * Performance optimization for fail fast - An optimization for runtime here is
  * to sort the execution of child guards depending on the failure count.  This
  * ensures that we run the guards that are more prone to fail statistically
  * first. This can improve the cache lookup time when we have multiple cache
@@ -2831,7 +2831,7 @@ class RootGuardManager : public GuardManager {
   template <typename T>
   bool check_nopybind_template(T* value) { // borrowed ref
     // Check [Note on GIL interaction with mutex lock] for details on why we
-    // need mutex and its interactions wth GIL.
+    // need mutex and its interactions with GIL.
     PyThreadState* _save = nullptr;
     Py_UNBLOCK_THREADS; // ; is added to avoid clang-formatting
     std::lock_guard<std::mutex> lock_guard(_lock);
@@ -2889,7 +2889,7 @@ class RootGuardManager : public GuardManager {
   GuardDebugInfo check_verbose_nopybind(
       PyObject* value) override { // borrowed ref
     // Check [Note on GIL interaction with mutex lock] for details on why we
-    // need mutex and its interactions wth GIL.
+    // need mutex and its interactions with GIL.
     PyThreadState* _save = nullptr;
     Py_UNBLOCK_THREADS; // ; is added to avoid clang-formatting
     std::lock_guard<std::mutex> lock_guard(_lock);
@@ -2992,7 +2992,7 @@ class RootGuardManager : public GuardManager {
   LocalState _local_state;
 
  private:
-  // All the relational guards under this guard mananger. We only use these
+  // All the relational guards under this guard manager. We only use these
   // when the guard evaluates to False. This ensures that guard state is reset
   // on guard failure so that next invocation is clean.
   std::vector<std::shared_ptr<RelationalGuard>> _relational_guard_resetters;
@@ -3575,7 +3575,7 @@ class TENSOR_MATCH : public LeafGuard {
 };
 
 /**
- * Represents __getattr__ acccessor.
+ * Represents __getattr__ accessor.
  */
 class GetAttrGuardAccessor : public GuardAccessor {
  public:
@@ -3623,7 +3623,7 @@ class GetAttrGuardAccessor : public GuardAccessor {
   }
 
   std::string repr() const override {
-    // Helpful when priting GuardManager tree structure.
+    // Helpful when printing GuardManager tree structure.
     return "GetAttrGuardAccessor(" + py::str(_attr_name).cast<std::string>() +
         ")";
   }
@@ -3651,7 +3651,7 @@ class GetAttrGuardAccessor : public GuardAccessor {
 };
 
 /**
- * Represents object.__getattribute__(obj, attr_name) acccessor.
+ * Represents object.__getattribute__(obj, attr_name) accessor.
  */
 class GenericGetAttrGuardAccessor : public GuardAccessor {
  public:
@@ -3699,7 +3699,7 @@ class GenericGetAttrGuardAccessor : public GuardAccessor {
   }
 
   std::string repr() const override {
-    // Helpful when priting GuardManager tree structure.
+    // Helpful when printing GuardManager tree structure.
     return "GenericGetAttrGuardAccessor(" +
         py::str(_attr_name).cast<std::string>() + ")";
   }
@@ -3730,7 +3730,7 @@ class GenericGetAttrGuardAccessor : public GuardAccessor {
 };
 
 /**
- * Represents x.__dict__ acccessor.
+ * Represents x.__dict__ accessor.
  */
 class GetGenericDictGuardAccessor : public GuardAccessor {
  public:
@@ -3777,7 +3777,7 @@ class GetGenericDictGuardAccessor : public GuardAccessor {
   }
 
   std::string repr() const override {
-    // Helpful when priting GuardManager tree structure.
+    // Helpful when printing GuardManager tree structure.
     return "GetGenericDictGuardAccessor";
   }
 
@@ -3798,7 +3798,7 @@ class GetGenericDictGuardAccessor : public GuardAccessor {
 };
 
 /**
- * Represents __getitem__ acccessor.
+ * Represents __getitem__ accessor.
  */
 class GetItemGuardAccessor : public GuardAccessor {
  public:
@@ -3995,7 +3995,7 @@ class FrameLocalsGuardAccessor : public GuardAccessor {
 };
 
 /**
- * Represents dict[name] acccessor. Needed since DictGuardManager does not
+ * Represents dict[name] accessor. Needed since DictGuardManager does not
  * support sorting. We differentiate it from GetItemGuardAccessor because
  * PyDict_GetItem should be faster than PyObject_GetItem.
  */
@@ -4023,7 +4023,7 @@ class DictGetItemGuardAccessor : public GuardAccessor {
         _guard_manager->has_no_accessors()) {
       // immutable object and dict tag matches, we can skip the guard subtree.
       // NB: We only skip the subtree if there are no accessors in the subtree.
-      // This is specificallly for tensors which are used in symbolic shape C++
+      // This is specifically for tensors which are used in symbolic shape C++
       // guards, and therefore have accessors on the tensor GuardManager itself.
       return true;
     }
@@ -4244,7 +4244,7 @@ std::string to_string(TensorProperty prop) {
 }
 
 /**
- * Represents tensor.size/shape/storage_offset acccessor.
+ * Represents tensor.size/shape/storage_offset accessor.
  */
 template <TensorProperty _prop>
 class TensorPropertyGuardAccessor : public GuardAccessor {
@@ -4342,7 +4342,7 @@ class TensorPropertyGuardAccessor : public GuardAccessor {
   }
 
   std::string repr() const override {
-    // Helpful when priting GuardManager tree structure.
+    // Helpful when printing GuardManager tree structure.
     return "TensorPropertyGuardAccessor<" + to_string(_prop) + +">(" +
         std::to_string(_index) + ")";
   }
@@ -4434,7 +4434,7 @@ class IndexedGuardAccessor : public GuardAccessor {
 };
 
 /**
- * Represents tensor.grad acccessor.
+ * Represents tensor.grad accessor.
  */
 class GradGuardAccessor : public GuardAccessor {
  public:
@@ -4485,7 +4485,7 @@ class GradGuardAccessor : public GuardAccessor {
   }
 
   std::string repr() const override {
-    // Helpful when priting GuardManager tree structure.
+    // Helpful when printing GuardManager tree structure.
     return "GradGuardAccessor(grad)";
   }
 
@@ -4654,7 +4654,7 @@ class FuncKwDefaultsGuardAccessor : public GuardAccessor {
 };
 
 /**
- * Represents f_globals acccessor. This sits as a child accessor of the
+ * Represents f_globals accessor. This sits as a child accessor of the
  * RootGuardManager.
  */
 class GlobalsGuardAccessor : public GuardAccessor {
@@ -4847,7 +4847,7 @@ class TupleIteratorGetItemAccessor : public GuardAccessor {
  * GlobalWeakRef accessor. Dynamo can insert a weakref object into the frame
  * globals. This accessor reads the globals and then calls the weakref object
  * to get the underlying object. This is a child of GlobalsGuardAccessor.
- * Therefore, we will get the globals dict while caling check_nopybind.
+ * Therefore, we will get the globals dict while calling check_nopybind.
  */
 class GlobalWeakRefGuardAccessor : public GuardAccessor {
  public:
@@ -5207,7 +5207,7 @@ void install_object_aliasing_guard(
   std::shared_ptr<RelationalGuard> guard =
       std::make_shared<OBJECT_ALIASING>(std::move(verbose_code_parts));
 
-  // Register the resetter on the root guard mananger, so that it can reset
+  // Register the resetter on the root guard manager, so that it can reset
   // the newly added relational guard when the guard eval fails.
   x->get_root()->add_relational_guard_resetter(guard);
 
@@ -5227,7 +5227,7 @@ void install_no_tensor_aliasing_guard(
   std::shared_ptr<RelationalGuard> guard = std::make_shared<NO_TENSOR_ALIASING>(
       tensor_names, std::move(verbose_code_parts));
 
-  // Register the resetter on the root guard mananger, so that it can reset
+  // Register the resetter on the root guard manager, so that it can reset
   // the newly added relational guard when the guard eval fails.
   py::cast<GuardManager*>(guard_managers[0])
       ->get_root()
@@ -5255,7 +5255,7 @@ void install_symbolic_shape_guard(
           std::move(py_addr_keep_alive),
           std::move(verbose_code_parts));
 
-  // Register the resetter on the root guard mananger, so that it can reset
+  // Register the resetter on the root guard manager, so that it can reset
   // the newly added relational guard when the guard eval fails.
   py::cast<GuardManager*>(guard_managers[0])
       ->get_root()
@@ -6309,7 +6309,7 @@ PyObject* torch_c_dynamo_guards_init() {
             self.add_permitted_leaf_guard(std::make_shared<NO_HASATTR>(
                 std::move(attr_name), std::move(verbose_code_parts)));
           })
-      // Not permitted accesssors
+      // Not permitted accessors
       .def("lambda_manager", &DictGuardManager::fail_on_get_child_manager)
       .def("getitem_manager", &DictGuardManager::fail_on_get_child_manager)
       .def("dict_getitem_manager", &DictGuardManager::fail_on_get_child_manager)

torch/csrc/inductor/aoti_eager/kernel_holder.cpp

@@ -110,7 +110,7 @@ std::vector<ParameterMetadata> unpack_input_parameters(
     }
 
     if (stack[idx].isScalar()) {
-      // Beyond c10::Scalar, the floating value and interger value are also
+      // Beyond c10::Scalar, the floating value and integer value are also
       // represented as Scalar.
       inputs_metadata.emplace_back(stack[idx].toScalar(), arg_order);
     } else if (stack[idx].isTensorList()) {
@@ -528,7 +528,7 @@ std::string AOTIPythonKernelHolder::produce_aoti_kernel_lib(
   auto kernel_lib_path = py::cast<std::string>(result);
   TORCH_CHECK(
       !kernel_lib_path.empty(),
-      "Failed to produce kernel libarary by using AOTI for ",
+      "Failed to produce kernel library by using AOTI for ",
       c10::DeviceTypeName(device_.type()),
       ". Operator Name is ",
       op.operator_name().name,

torch/csrc/inductor/aoti_runner/model_container_runner.h

@@ -121,8 +121,8 @@ TORCH_API std::unordered_map<std::string, CreateAOTIModelRunnerFunc>&
 getAOTIModelRunnerRegistry();
 
 // To register a new external backend in AOTI one needs to create an instance of
-// this struct. It is not thread-safe. Becase it is expected to be called during
-// the initialization of the program.
+// this struct. It is not thread-safe. Because it is expected to be called
+// during the initialization of the program.
 struct TORCH_API RegisterAOTIModelRunner{RegisterAOTIModelRunner(
     const std::string& name,
     CreateAOTIModelRunnerFunc create_aoti_model_runner_fn){

torch/csrc/inductor/aoti_runtime/model.h

@@ -659,7 +659,7 @@ class AOTInductorModelBase {
     AOTI_RUNTIME_CHECK(
         reinterpret_cast<uint64_t*>(
             self_mmap + weights_size - sizeof(uint64_t))[0] == magic_number,
-        "Weigths data seems corrupt");
+        "Weights data seems corrupt");
     return self_mmap;
 #endif
   }
@@ -707,7 +707,7 @@ class AOTInductorModelBase {
   bool include_weights;
 
   // Record if the model finishes an inference run so that its owning
-  // AOTModelContainer can re-use this instance.
+  // AOTModelContainer can reuse this instance.
 #ifdef USE_CUDA
   std::optional<cudaEvent_t> run_finished_;
 #elif defined(USE_XPU)

torch/csrc/inductor/aoti_runtime/model_container.h

@@ -18,7 +18,7 @@ namespace torch::aot_inductor {
 // when model_container is created and no constants are being loaded or updated.
 // (2) INITIALIZED state: This state get set whenever we load the constants into
 // the buffer. This could be done by load_constants or update_constants_buffer.
-// (3) FOLDED state: This state should transition from INITIALILZED after
+// (3) FOLDED state: This state should transition from INITIALIZED after
 // const_fold is being invoked.
 enum class ConstantState : uint8_t { NONE, INITIALIZED, FOLDED, UNKNOWN };
 

torch/csrc/inductor/aoti_torch/oss_proxy_executor.cpp

@@ -872,7 +872,7 @@ void OSSProxyExecutor::call_function(
       auto serialized_int_value = flatten_int_args[int_id++];
       TORCH_CHECK(
           returned_int_value == serialized_int_value,
-          "Expect returned int value to match the serialized int value, but got retured int value: ",
+          "Expect returned int value to match the serialized int value, but got returned int value: ",
           returned_int_value,
           " and serialized int value: ",
           serialized_int_value);