diff --git a/paddle/fluid/framework/new_executor/interpreter/static_build.cc b/paddle/fluid/framework/new_executor/interpreter/static_build.cc
index 9e530761b76245..cbbe36191452bd 100644
--- a/paddle/fluid/framework/new_executor/interpreter/static_build.cc
+++ b/paddle/fluid/framework/new_executor/interpreter/static_build.cc
@@ -348,7 +348,7 @@ void FakeInitializeTensor(const phi::DeviceContext& dev_ctx,
       PADDLE_ENFORCE_EQ(place,
                         dev_ctx.GetPlace(),
                         common::errors::Unavailable(
-                            "The place %s for fack alloc is not equal to "
+                            "The place %s for fake alloc is not equal to "
                             "the place %s of DeviceContext.",
                             place,
                             dev_ctx.GetPlace()));
@@ -525,7 +525,7 @@ void RunWhileBlockPreStaticBuild(const framework::Scope& scope,
   // note(lvyongkang): The assign op in while loop may change the place of
   // variable. However, InterpreterCore fix the kernel of every ops during its
   // first run. A cpu tensor may become gpu tensor after first run. This will
-  // lead to segmetation fault when it's used in a cpu kernel. Here we record
+  // lead to segmentation fault when it's used in a cpu kernel. Here we record
   // the place of every inputs and restore their place after
   // InterpreterCore.run().
   std::map<std::string, phi::Place> input_var_original_places;
@@ -696,7 +696,7 @@ void FakeInitializeOutputsForOperatorBase(
         if (following_input_vars.count(var_name)) {
           PADDLE_THROW(common::errors::PreconditionNotMet(
               "The output %s s' dtype/place of %s is "
-              "changed after static build. Befer static build, the "
+              "changed after static build. Before static build, the "
               "dtype is %s, place is %s. After static "
               "build, the dtype is %s, place is %s.",
               op_type,
diff --git a/paddle/fluid/framework/new_executor/pir_adaptor/pir_adaptor_util.cc b/paddle/fluid/framework/new_executor/pir_adaptor/pir_adaptor_util.cc
index 7ecef0a54751f5..11ed55043c723b 100644
--- a/paddle/fluid/framework/new_executor/pir_adaptor/pir_adaptor_util.cc
+++ b/paddle/fluid/framework/new_executor/pir_adaptor/pir_adaptor_util.cc
@@ -450,8 +450,8 @@ void BuildValue(pir::Value value,
   } else {
     PADDLE_THROW(common::errors::PreconditionNotMet(
         "Output only support DenseTensorType "
-        "or SelectedRowsType or VectorType or StackType or SpasrCooTensorType "
-        "or SpasreCsrTensorType"));
+        "or SelectedRowsType or VectorType or StackType or SparseCooTensorType "
+        "or SparseCsrTensorType"));
   }
 }
 
@@ -710,9 +710,9 @@ bool IsNeedVarInplace(pir::Operation* op,
 
 // NOTE(chenxi67): Here, we only perform inplace processing for variables that
 // need to be inplaced by var (mostly, whose type is TensorArray or re-Allocated
-// Densetensor). For other types of variables, we only share the holder of
+// DenseTensor). For other types of variables, we only share the holder of
 // DenseTensor but not the var*. The reason is that vector<DenseTensor> in
-// TensorArray (or re-Allocated Densetensor) cannot be shared totally.
+// TensorArray (or re-Allocated DenseTensor) cannot be shared totally.
 void HandleForInplaceVarOp(pir::Operation* op,
                            const std::string& var_name_prefix,
                            ValueExecutionInfo* value_exe_info) {
diff --git a/paddle/fluid/framework/new_executor/workqueue/nonblocking_threadpool.h b/paddle/fluid/framework/new_executor/workqueue/nonblocking_threadpool.h
index ae0125f8204089..11d45d7d9560e7 100644
--- a/paddle/fluid/framework/new_executor/workqueue/nonblocking_threadpool.h
+++ b/paddle/fluid/framework/new_executor/workqueue/nonblocking_threadpool.h
@@ -50,7 +50,7 @@ class ThreadPoolTempl {
     // and NonEmptyQueueIndex. Iteration is based on the fact that if we take
     // a random starting thread index t and calculate num_threads - 1 subsequent
     // indices as (t + coprime) % num_threads, we will cover all threads without
-    // repetitions (effectively getting a presudo-random permutation of thread
+    // repetitions (effectively getting a pseudo-random permutation of thread
     // indices).
     assert(num_threads_ >= 1 && num_threads_ < kMaxThreads);
     all_coprimes_.reserve(num_threads_);