Fix typo (allreudce -> allreduce)(#61761)

co63oc · web-flow · commit 12fd0601102d · 2024-02-19T16:27:49.000+08:00
diff --git a/paddle/fluid/framework/ir/fused_attention_pass.cc b/paddle/fluid/framework/ir/fused_attention_pass.cc
@@ -267,9 +267,9 @@ PDNode* FusedAttentionPattern::operator()(PDNode* x,
 
   PDNode* mp_allreduce_out_node{nullptr};
   if (use_mp) {
-    mp_allreduce_out_node = pattern->NewNode(mp_allreudce_sum_out_repr())
+    mp_allreduce_out_node = pattern->NewNode(mp_allreduce_sum_out_repr())
                                 ->assert_is_op_output("mp_allreduce_sum");
-    auto* mp_allreduce_node = pattern->NewNode(mp_allreudce_sum_op_repr())
+    auto* mp_allreduce_node = pattern->NewNode(mp_allreduce_sum_op_repr())
                                   ->assert_is_op("mp_allreduce_sum");
     out_linear_ele_add_out_node->assert_is_op_input("mp_allreduce_sum");
     mp_allreduce_node->LinksFrom({out_linear_ele_add_out_node})
@@ -460,9 +460,9 @@ PDNode* FusedAttentionGradPattern::operator()(PDNode* x,
 
   PDNode* mp_c_identity_out_node{nullptr};
   if (use_mp) {
-    mp_c_identity_out_node = pattern->NewNode(mp_allreudce_sum_grad_out_repr())
+    mp_c_identity_out_node = pattern->NewNode(mp_allreduce_sum_grad_out_repr())
                                  ->assert_is_op_output("c_identity", "Out");
-    auto* mp_c_identity_node = pattern->NewNode(mp_allreudce_sum_grad_op_repr())
+    auto* mp_c_identity_node = pattern->NewNode(mp_allreduce_sum_grad_op_repr())
                                    ->assert_is_op("c_identity");
     out_linear_dropout_grad_out_node->assert_is_op_input("c_identity");
     mp_c_identity_node->LinksFrom({out_linear_dropout_grad_out_node})
@@ -989,13 +989,13 @@ ir::Graph* FusedAttentionsPass::ForwardHandlerHelper(
     if (use_mp) {
       GET_IR_NODE_FROM_SUBGRAPH(
           c_identity_op_node, c_identity_op, fused_attention_pattern);
-      GET_IR_NODE_FROM_SUBGRAPH(mp_allreudce_sum_op_node,
-                                mp_allreudce_sum_op,
+      GET_IR_NODE_FROM_SUBGRAPH(mp_allreduce_sum_op_node,
+                                mp_allreduce_sum_op,
                                 fused_attention_pattern);
       remove_nodes.insert(c_identity_op_node);
-      remove_nodes.insert(mp_allreudce_sum_op_node);
+      remove_nodes.insert(mp_allreduce_sum_op_node);
       ring_id = PADDLE_GET_CONST(
-          int, mp_allreudce_sum_op_node->Op()->GetAttr("ring_id"));
+          int, mp_allreduce_sum_op_node->Op()->GetAttr("ring_id"));
     }
 
     std::string cache_anchor_name = fuse_qkv_matmul_w_node->Var()->Name();
@@ -1367,16 +1367,16 @@ ir::Graph* FusedAttentionsPass::BackwardHandlerHelper(
 
     int ring_id = -1;
     if (use_mp) {
-      GET_IR_NODE_FROM_SUBGRAPH(mp_allreudce_sum_grad_op_node,
-                                mp_allreudce_sum_grad_op,
+      GET_IR_NODE_FROM_SUBGRAPH(mp_allreduce_sum_grad_op_node,
+                                mp_allreduce_sum_grad_op,
                                 fused_attention_grad_pattern);
       GET_IR_NODE_FROM_SUBGRAPH(c_identity_grad_op_node,
                                 c_identity_grad_op,
                                 fused_attention_grad_pattern);
-      remove_nodes.insert(mp_allreudce_sum_grad_op_node);
+      remove_nodes.insert(mp_allreduce_sum_grad_op_node);
       remove_nodes.insert(c_identity_grad_op_node);
       ring_id = PADDLE_GET_CONST(
-          int, mp_allreudce_sum_grad_op_node->Op()->GetAttr("ring_id"));
+          int, mp_allreduce_sum_grad_op_node->Op()->GetAttr("ring_id"));
     }
 
     OpDesc fused_attention_grad_op_desc(
diff --git a/paddle/fluid/framework/ir/fused_attention_pass.h b/paddle/fluid/framework/ir/fused_attention_pass.h
@@ -117,9 +117,9 @@ struct FusedAttentionPattern : public PatternBase {
   PATTERN_DECL_NODE(out_linear_ele_add_bias);
   PATTERN_DECL_NODE(out_linear_ele_add_out);
 
-  // allreudce for mp
-  PATTERN_DECL_NODE(mp_allreudce_sum_op);
-  PATTERN_DECL_NODE(mp_allreudce_sum_out);
+  // allreduce for mp
+  PATTERN_DECL_NODE(mp_allreduce_sum_op);
+  PATTERN_DECL_NODE(mp_allreduce_sum_out);
 
   PATTERN_DECL_NODE(out_linear_dropout_op);
   PATTERN_DECL_NODE(out_linear_dropout_out);
@@ -174,8 +174,8 @@ struct FusedAttentionGradPattern : public PatternBase {
   PATTERN_DECL_NODE(out_linear_dropout_grad_out);
 
   // c_identity for mp
-  PATTERN_DECL_NODE(mp_allreudce_sum_grad_op);  // c_identity
-  PATTERN_DECL_NODE(mp_allreudce_sum_grad_out);
+  PATTERN_DECL_NODE(mp_allreduce_sum_grad_op);  // c_identity
+  PATTERN_DECL_NODE(mp_allreduce_sum_grad_out);
 
   PATTERN_DECL_NODE(out_linear_ele_add_grad_op);
   PATTERN_DECL_NODE(out_linear_ele_add_grad_x);
diff --git a/paddle/fluid/imperative/reducer.cc b/paddle/fluid/imperative/reducer.cc
@@ -572,7 +572,7 @@ void Reducer::TraverseBackwardGraph(
 }
 
 // After each batch is calculated, the counter of each group(group.pending_)
-// and allreudce sequence counter(next_group_) will be cleaned up again.
+// and allreduce sequence counter(next_group_) will be cleaned up again.
 void Reducer::PrepareForBackward(
     const std::vector<std::shared_ptr<imperative::VarBase>> &outputs) {
   VLOG(3) << "after forward, then reset count for backward.";
diff --git a/python/paddle/distributed/fleet/utils/tensor_fusion_helper.py b/python/paddle/distributed/fleet/utils/tensor_fusion_helper.py
@@ -359,7 +359,7 @@ def __init__(
             assert dst != -1
         else:
             raise ValueError(
-                "The act should be allreudce for dp or reduce for sharding."
+                "The act should be allreduce for dp or reduce for sharding."
             )
         self._dst = dst
 
diff --git a/test/collective/fleet/test_fleet_sharding_meta_optimizer.py b/test/collective/fleet/test_fleet_sharding_meta_optimizer.py
@@ -880,7 +880,7 @@ def test_sharding_hybrid_dp(self):
                 loss_scale = 1.0 / scale
                 self.assertAlmostEqual(float(op.attr('value')), loss_scale)
 
-        # check program (allreudce)
+        # check program (allreduce)
         ops = [op.type for op in main_prog_ops]
         self.assertEqual(
             ops,

Original file line number	Diff line number	Diff line change
`@@ -572,7 +572,7 @@ void Reducer::TraverseBackwardGraph(`
`572`	`572`	`}`
`573`	`573`
`574`	`574`	`// After each batch is calculated, the counter of each group(group.pending_)`
`575`		`-// and allreudce sequence counter(next_group_) will be cleaned up again.`
	`575`	`+// and allreduce sequence counter(next_group_) will be cleaned up again.`
`576`	`576`	`void Reducer::PrepareForBackward(`
`577`	`577`	`const std::vector<std::shared_ptr<imperative::VarBase>> &outputs) {`
`578`	`578`	`VLOG(3) << "after forward, then reset count for backward.";`
Original file line number	Diff line number	Diff line change
`@@ -359,7 +359,7 @@ def __init__(`
`359`	`359`	`assert dst != -1`
`360`	`360`	`else:`
`361`	`361`	`raise ValueError(`
`362`		`- "The act should be allreudce for dp or reduce for sharding."`
	`362`	`+ "The act should be allreduce for dp or reduce for sharding."`
`363`	`363`	`)`
`364`	`364`	`self._dst = dst`
`365`	`365`