diff --git a/paddle/fluid/eager/pylayer/py_layer_node.cc b/paddle/fluid/eager/pylayer/py_layer_node.cc
index e601f17d88fd391725cd4355827a479d375c5a92..3be77e5984313b97d66a4118739dc7d90976bf09 100644
--- a/paddle/fluid/eager/pylayer/py_layer_node.cc
+++ b/paddle/fluid/eager/pylayer/py_layer_node.cc
@@ -165,7 +165,7 @@ GradNodePyLayer::operator()(
           this->OutputMeta()[i][0].IsStopGradient(),
           true,
           paddle::platform::errors::InvalidArgument(
-              "%s's backward function should not return empyt at %d position.",
+              "%s's backward function should not return empty at %d position.",
               name(),
               i));
       grad_out.push_back({});
diff --git a/paddle/fluid/framework/details/all_reduce_op_handle.cc b/paddle/fluid/framework/details/all_reduce_op_handle.cc
index 4a7b1d602d6ae18fbea24ca7a4aaa5c7889f7b5e..2afeef1efec403ed0ac1d7a845f3f57802983efd 100644
--- a/paddle/fluid/framework/details/all_reduce_op_handle.cc
+++ b/paddle/fluid/framework/details/all_reduce_op_handle.cc
@@ -135,7 +135,7 @@ void AllReduceOpHandle::AllReduceImpl(
 
     if (i == 0) {
       numel = static_cast<int64_t>(lod_tensor.numel());
-      // only enforce place0, we will enforce other palce numel == place0 numel
+      // only enforce place0, we will enforce other place numel == place0 numel
       PADDLE_ENFORCE_GT(
           numel,
           0,
diff --git a/paddle/fluid/framework/details/bind_threaded_ssa_graph_executor.cc b/paddle/fluid/framework/details/bind_threaded_ssa_graph_executor.cc
index 577d458ba7da5800607affae6f529bc37e1bc9db..f36f29a5b02177292bd0987da924f60e97b9f378 100644
--- a/paddle/fluid/framework/details/bind_threaded_ssa_graph_executor.cc
+++ b/paddle/fluid/framework/details/bind_threaded_ssa_graph_executor.cc
@@ -144,7 +144,7 @@ FetchResultType BindThreadedSSAGraphExecutor::RunMainStream(
   while (cur_count < op_deps->size()) {
     cur_count++;
     auto cur_op = ready_ops->Pop();
-    // when execption, get cur_op == nullptr
+    // when exception, get cur_op == nullptr
     if (cur_op == nullptr) {
       std::lock_guard<std::mutex> lock(mutex_);
       exec_op_count_ = op_deps->size();
diff --git a/paddle/fluid/framework/details/eager_deletion_op_handle.cc b/paddle/fluid/framework/details/eager_deletion_op_handle.cc
index ade3a6f842e04a210087e9c20b53c390e4e29c25..4012263f688cb5cc3875df8f9614045b931bd2d0 100644
--- a/paddle/fluid/framework/details/eager_deletion_op_handle.cc
+++ b/paddle/fluid/framework/details/eager_deletion_op_handle.cc
@@ -59,7 +59,7 @@ EagerDeletionOpHandle::EagerDeletionOpHandle(
 #endif
       PADDLE_ENFORCE_NOT_NULL(
           event_,
-          platform::errors::InvalidArgument("The cuda envet created is NULL."));
+          platform::errors::InvalidArgument("The cuda event created is NULL."));
     }
   }
 #endif
diff --git a/paddle/fluid/framework/details/fetch_async_op_handle.cc b/paddle/fluid/framework/details/fetch_async_op_handle.cc
index 922dbb488018e76080bece86b608c8220e7c28ca..5088843bb98d3979699fe6c56bfc80a218613aaa 100644
--- a/paddle/fluid/framework/details/fetch_async_op_handle.cc
+++ b/paddle/fluid/framework/details/fetch_async_op_handle.cc
@@ -203,7 +203,7 @@ void FetchAsyncOpHandle::FetchMergedLodTensor(
   }
 
   // slice and memcpy
-  // for 0D tensor, can't concat eath tensor, stack them. for 1+D tensor, concat
+  // for 0D tensor, can't concat each tensor, stack them. for 1+D tensor, concat
   // them
   int begin = 0;
   int end = 0;
diff --git a/paddle/fluid/framework/details/multi_devices_helper.h b/paddle/fluid/framework/details/multi_devices_helper.h
index a3a3b993b7ec14fb95f4d59ea6bd86edec13207e..60c3f35a4f7f720c90742e7dfae5f58add5b92d0 100644
--- a/paddle/fluid/framework/details/multi_devices_helper.h
+++ b/paddle/fluid/framework/details/multi_devices_helper.h
@@ -44,8 +44,8 @@ namespace details {
 // all variable in each devices.
 // The outside vector is the device vector. Each element of this vector is a
 // map from variable name to variables. The variables, who have the same name,
-// will have a differsent version. The offset in the
-// `std::vector<VarHandle*>` is the version of varaibles.
+// will have a different version. The offset in the
+// `std::vector<VarHandle*>` is the version of variables.
 typedef std::vector<std::unordered_map<std::string, std::vector<VarHandle *>>>
     GraphVars;
 constexpr char kGraphVars[] = "vars";
diff --git a/paddle/fluid/framework/details/op_handle_base.h b/paddle/fluid/framework/details/op_handle_base.h
index c9021c84cdf2de8b04b7ec725356b5585dcddee8..9afe56e4babd45bbb0680981bf7d20317b0e8a9a 100644
--- a/paddle/fluid/framework/details/op_handle_base.h
+++ b/paddle/fluid/framework/details/op_handle_base.h
@@ -154,10 +154,10 @@ class OpHandleBase {
 
   std::vector<Scope *> local_exec_scopes_;
   bool skip_running_ = false;
-  // NOTE(Aurelius84): Indicate whether scope held in OpHandle is chanageable.
-  // Ophandle's scope noramlly keep same in most cases, except running
+  // NOTE(Aurelius84): Indicate whether scope held in OpHandle is changeable.
+  // Ophandle's scope normally keep same in most cases, except running
   // run_program_op from @to_static.
-  // The scope may be chanaged while each training iteration.
+  // The scope may be changed while each training iteration.
   // See https://github.com/PaddlePaddle/Paddle/pull/32283
   bool is_variant_scope_ = false;