Fix Optimze Optimize (#60936)

co63oc · web-flow · commit 2ed019d7b0f9 · 2024-01-19T10:21:12.000+08:00
diff --git a/paddle/fluid/imperative/amp_auto_cast.cc b/paddle/fluid/imperative/amp_auto_cast.cc
@@ -399,7 +399,7 @@ NameVarMap<VarType> AutoCastInputs(const std::string& op_type,
           pair.first != "Mask" && dst_type == framework::proto::VarType::FP32) {
         continue;
       }
-      if ((op_type == "fused_attention" || op_type == "fused_feedforwad") &&
+      if ((op_type == "fused_attention" || op_type == "fused_feedforward") &&
           dst_type == framework::proto::VarType::FP32) {
         if (pair.first != "LnScale" && pair.first != "LnBias" &&
             pair.first != "Ln2Scale" && pair.first != "Ln2Bias" &&
diff --git a/paddle/fluid/imperative/amp_auto_cast.h b/paddle/fluid/imperative/amp_auto_cast.h
@@ -31,7 +31,7 @@ enum class AmpLevel {
   O1,      // amp, mixed fp32-fp16
   O2,      // almost fp16
   O3,      // fp16
-  OD,      // only conv and matmul use low precison.
+  OD,      // only conv and matmul use low precision.
 };
 
 std::tuple<std::unordered_set<std::string>,
diff --git a/paddle/fluid/imperative/basic_engine.cc b/paddle/fluid/imperative/basic_engine.cc
@@ -236,7 +236,7 @@ void BasicEngine::PrepareGradAccumulators(
 
             accumulator->IncreaseRefCnt();
 
-            VLOG(3) << "Prepare to acccumulate variable grad " << var->Name()
+            VLOG(3) << "Prepare to accumulate variable grad " << var->Name()
                     << "(" << var.get()
                     << ") that has grad node with reference count "
                     << accumulator->RefCnt();
@@ -267,7 +267,7 @@ void BasicEngine::PrepareGradAccumulators(
 
         accumulator->IncreaseRefCnt();
 
-        VLOG(3) << "Prepare to acccumulate variable grad " << var->Name() << "("
+        VLOG(3) << "Prepare to accumulate variable grad " << var->Name() << "("
                 << var.get()
                 << ") that don't have grad node  with reference count "
                 << accumulator->RefCnt();
@@ -445,7 +445,7 @@ void BasicEngine::Execute() {
        *
        * - construct the temp output map, avoid to disrupt graph
        * - replace the element in the map by temp var, because a
-       *   var may be coresponding to several grad var in one op
+       *   var may be corresponding to several grad var in one op
        */
       NameVarMap<VariableWrapper> tmp_outs(bwd_outs);
 
@@ -495,7 +495,7 @@ void BasicEngine::Execute() {
           }
 
           // leaf_accumulators_ : hooks and accumulate-grad for leaf tensor,
-          // it should be orderly and not reapeated.
+          // it should be orderly and not repeated.
           if (var->IsLeafGrad()) {
             if (std::find(leaf_accumulators_.begin(),
                           leaf_accumulators_.end(),
diff --git a/paddle/fluid/imperative/data_loader.cc b/paddle/fluid/imperative/data_loader.cc
@@ -45,7 +45,7 @@ void EraseLoadProcessPIDs(int64_t key) {
             << ")";
     load_process_pids.erase(it);
   } else {
-    VLOG(3) << "Dygraph Data Loader: The dygrph loader (id: " << key
+    VLOG(3) << "Dygraph Data Loader: The dygraph loader (id: " << key
             << ") you want erase does not exist.";
   }
 }
diff --git a/paddle/fluid/imperative/gradient_accumulator.cc b/paddle/fluid/imperative/gradient_accumulator.cc
@@ -138,7 +138,7 @@ TType& GetInnerTensor(const paddle::Tensor& src) {
       src.initialized(),
       true,
       platform::errors::Fatal("We only add tensor with value if a tensor is "
-                              "NOT INITILIZED, it should just move instead of "
+                              "NOT INITIALIZED, it should just move instead of "
                               "calling this method."));
   auto* src_tensor = static_cast<TType*>(src.impl().get());
   return *src_tensor;
@@ -198,7 +198,7 @@ void TensorAdd(const VarType& src, VarType* dst) {
   }
 
   // AddKernel already support inputs of different dtype. For AMP master_grad,
-  // the dtype of source tensor and destination tensor will be diferent. So the
+  // the dtype of source tensor and destination tensor will be different. So the
   // check requiring input dtypes to be the same have been removed.
 #define PADDLE_TENSOR_ADD(T, CONTEXT)                                          \
   if (data_type == framework::DataTypeTrait<T>::DataType()) {                  \
@@ -550,7 +550,7 @@ void GradientAccumulator::AccumulateGrad() {
   auto* dst = var_->MutableVar();
   if (!var_->IsEmpty()) {
     VLOG(6) << "Leaf Var(" << var_->Name()
-            << ")'s Gradient has been initizlized, will accumulate on "
+            << ")'s Gradient has been initialized, will accumulate on "
                "previous gradient.";
     if (dst->IsType<phi::DenseTensor>()) {
       if (src->IsType<phi::DenseTensor>()) {
diff --git a/paddle/fluid/imperative/gradient_accumulator.h b/paddle/fluid/imperative/gradient_accumulator.h
@@ -128,7 +128,7 @@ class GradientAccumulator {
 
  protected:
   VariableWrapper* var_;
-  // NOTE: only gradient accumulater of leaf tensor should hold
+  // NOTE: only gradient accumulator of leaf tensor should hold
   // inner_var_, So not hold it by other shared pointer.
   std::shared_ptr<VariableWrapper> inner_var_;
   size_t ref_cnt_{0};
diff --git a/paddle/fluid/imperative/infer_shape_context.h b/paddle/fluid/imperative/infer_shape_context.h
@@ -401,7 +401,7 @@ class DygraphInferShapeContext : public framework::InferShapeContext {
                       it->second.size(),
                       platform::errors::InvalidArgument(
                           "The number of dims is expected to be equal to the "
-                          "number of Outputs(%s). But receieved: the number of "
+                          "number of Outputs(%s). But received: the number of "
                           "dims = %d, the number of Outputs(%s) = %d.",
                           name,
                           dims.size(),
diff --git a/paddle/fluid/imperative/layout_autotune.cc b/paddle/fluid/imperative/layout_autotune.cc
@@ -33,7 +33,7 @@ LayoutAutoTune::LayoutAutoTune() {
       continue;
     }
 
-    // only record forwrd operators
+    // only record forward operators
     if (info.first.find("_grad") != std::string::npos) {
       continue;
     }
diff --git a/paddle/fluid/imperative/layout_transformer.h b/paddle/fluid/imperative/layout_transformer.h
@@ -85,7 +85,7 @@ class LayoutTransformer {
       const paddle::imperative::NameVarMap<VarType>& outs,
       paddle::framework::AttributeMap* attrs,
       const std::shared_ptr<paddle::imperative::Tracer>& tracer) {
-    VLOG(3) << "Optimze Layout agnostic op: " << type_;
+    VLOG(3) << "Optimize Layout agnostic op: " << type_;
     auto in_layout = DataLayout::UNDEFINED;
     for (auto& pair : ins) {
       for (auto& var : pair.second) {
@@ -101,7 +101,7 @@ class LayoutTransformer {
         }
       }
     }
-    VLOG(3) << "Optimze Layout agnostic op: " << type_ << " "
+    VLOG(3) << "Optimize Layout agnostic op: " << type_ << " "
             << common::DataLayoutToString(in_layout);
     if (in_layout != DataLayout::UNDEFINED) {
       SetVarsLayout(outs, in_layout);
@@ -179,7 +179,7 @@ class HeavilyLayoutSensitiveOpTransformer : public LayoutTransformer<VarType> {
       const paddle::imperative::NameVarMap<VarType>& outs,
       paddle::framework::AttributeMap* attrs,
       const std::shared_ptr<paddle::imperative::Tracer>& tracer) {
-    VLOG(3) << "Optimze heavily layout sensitive op " << this->Type();
+    VLOG(3) << "Optimize heavily layout sensitive op " << this->Type();
     paddle::imperative::NameVarMap<VarType> new_ins(ins);
 
     // Step 1: Adjust the data_layout attr to the desired layout
@@ -238,7 +238,7 @@ class LightlyLayoutSensitiveOpTransformer : public LayoutTransformer<VarType> {
       const paddle::imperative::NameVarMap<VarType>& outs,
       paddle::framework::AttributeMap* attrs,
       const std::shared_ptr<paddle::imperative::Tracer>& tracer) {
-    VLOG(3) << "Optimze lightly layout sensitive op " << this->Type();
+    VLOG(3) << "Optimize lightly layout sensitive op " << this->Type();
     paddle::imperative::NameVarMap<VarType> new_ins(ins);
     // If input's layout is not tuned, transformation is unnecessary.
     // If input's layout is already tuned, it will be transformed back to NCHW.
@@ -326,7 +326,7 @@ class TransposeOpTransformer
       const paddle::imperative::NameVarMap<VarType>& outs,
       paddle::framework::AttributeMap* attrs,
       const std::shared_ptr<paddle::imperative::Tracer>& tracer) {
-    VLOG(3) << "Optimze lightly layout sensitive op " << this->Type();
+    VLOG(3) << "Optimize lightly layout sensitive op " << this->Type();
     // When the input layout is the desired format, it means that there
     // is a transpose layer in the network, it is better to transpose
     // the result to the original format.
@@ -360,7 +360,7 @@ class FlattenOpTransformer
       const paddle::imperative::NameVarMap<VarType>& outs,
       paddle::framework::AttributeMap* attrs,
       const std::shared_ptr<paddle::imperative::Tracer>& tracer) {
-    VLOG(3) << "Optimze lightly layout sensitive op " << this->Type();
+    VLOG(3) << "Optimize lightly layout sensitive op " << this->Type();
     // Flatten the C, H, W dimensions will not affect functionality.
     // So transformation is unnecessary. But in other cases, it needs to
     // fall back to the LightlyLayoutSensitiveOpTransformer.
@@ -389,7 +389,7 @@ class ArgmaxOpTransformer
       const paddle::imperative::NameVarMap<VarType>& outs,
       paddle::framework::AttributeMap* attrs,
       const std::shared_ptr<paddle::imperative::Tracer>& tracer) {
-    VLOG(3) << "Optimze lightly layout sensitive op " << this->Type();
+    VLOG(3) << "Optimize lightly layout sensitive op " << this->Type();
     auto& in_var = ins.at("X")[0];
     auto var_layout = paddle::imperative::GetDataLayout(in_var);
     bool keep_dims = PADDLE_GET_CONST(bool, (*attrs)["keepdims"]);
@@ -436,21 +436,21 @@ class ConcatOpTransformer
       const paddle::imperative::NameVarMap<VarType>& outs,
       paddle::framework::AttributeMap* attrs,
       const std::shared_ptr<paddle::imperative::Tracer>& tracer) {
-    VLOG(3) << "Optimze lightly layout sensitive op " << this->Type();
+    VLOG(3) << "Optimize lightly layout sensitive op " << this->Type();
     auto& in_var = ins.at("X")[0];
     auto var_layout = paddle::imperative::GetDataLayout(in_var);
-    bool need_tranppose = false;
+    bool need_transpose = false;
     for (auto& pair : ins) {
       for (auto& var : pair.second) {
         if (var != nullptr &&
             (paddle::imperative::GetDataLayout(var) != var_layout)) {
-          need_tranppose = true;
+          need_transpose = true;
           break;
         }
       }
     }
 
-    if (need_tranppose) {
+    if (need_transpose) {
       return LightlyLayoutSensitiveOpTransformer<VarType>::Apply(
           ins, outs, attrs, tracer);
     }
@@ -463,7 +463,7 @@ class ConcatOpTransformer
       (*attrs)["axis"] = static_cast<int>(perm[axis]);
     }
     auto axis = PADDLE_GET_CONST(int, (*attrs)["axis"]);
-    VLOG(3) << "Optimze lightly layout sensitive op asdfasdfasdf axis" << axis;
+    VLOG(3) << "Optimize lightly layout sensitive op axis" << axis;
 
     this->SetVarsLayout(outs, var_layout);
     return ins;
diff --git a/paddle/fluid/imperative/partial_grad_engine.cc b/paddle/fluid/imperative/partial_grad_engine.cc
@@ -196,7 +196,7 @@ static void GetGraphInfoBetweenTargets(
     bool is_valid = false;
     for (auto &output_pair : op->GetOutsMap()) {
       if (!output_pair.second.IsGrad()) {
-        VLOG(10) << "Continueded output for : " << op->Type();
+        VLOG(10) << "Continued output for : " << op->Type();
         continue;
       }
 
@@ -220,7 +220,7 @@ static void GetGraphInfoBetweenTargets(
     is_valid = false;
     for (auto &input_pair : op->GetInsMap()) {
       if (!input_pair.second.IsGrad()) {
-        VLOG(10) << "Continueded input for : " << op->Type();
+        VLOG(10) << "Continued input for : " << op->Type();
         continue;
       }
 

Original file line number	Diff line number	Diff line change
`@@ -399,7 +399,7 @@ NameVarMap<VarType> AutoCastInputs(const std::string& op_type,`
`399`	`399`	`pair.first != "Mask" && dst_type == framework::proto::VarType::FP32) {`
`400`	`400`	`continue;`
`401`	`401`	`}`
`402`		`- if ((op_type == "fused_attention" \|\| op_type == "fused_feedforwad") &&`
	`402`	`+ if ((op_type == "fused_attention" \|\| op_type == "fused_feedforward") &&`
`403`	`403`	`dst_type == framework::proto::VarType::FP32) {`
`404`	`404`	`if (pair.first != "LnScale" && pair.first != "LnBias" &&`
`405`	`405`	`pair.first != "Ln2Scale" && pair.first != "Ln2Bias" &&`
Original file line number	Diff line number	Diff line change
`@@ -45,7 +45,7 @@ void EraseLoadProcessPIDs(int64_t key) {`
`45`	`45`	`<< ")";`
`46`	`46`	`load_process_pids.erase(it);`
`47`	`47`	`} else {`
`48`		`- VLOG(3) << "Dygraph Data Loader: The dygrph loader (id: " << key`
	`48`	`+ VLOG(3) << "Dygraph Data Loader: The dygraph loader (id: " << key`
`49`	`49`	`<< ") you want erase does not exist.";`
`50`	`50`	`}`
`51`	`51`	`}`
Original file line number	Diff line number	Diff line change
`@@ -33,7 +33,7 @@ LayoutAutoTune::LayoutAutoTune() {`
`33`	`33`	`continue;`
`34`	`34`	`}`
`35`	`35`
`36`		`- // only record forwrd operators`
	`36`	`+ // only record forward operators`
`37`	`37`	`if (info.first.find("_grad") != std::string::npos) {`
`38`	`38`	`continue;`
`39`	`39`	`}`