From a348a423ce5238773355b07b504871cf152863f5 Mon Sep 17 00:00:00 2001
From: yu wentao <yuwentao126@126.com>
Date: Wed, 1 Mar 2023 16:21:00 +0800
Subject: [PATCH] rename distributed_fused_lamb attr ring_id->ring_ids (#51000)

---
 .../fluid/operators/optimizers/distributed_fused_lamb_op.cc   | 4 ++--
 .../fluid/operators/optimizers/distributed_fused_lamb_op.cu   | 2 +-
 python/paddle/incubate/optimizer/distributed_fused_lamb.py    | 2 +-
 3 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/paddle/fluid/operators/optimizers/distributed_fused_lamb_op.cc b/paddle/fluid/operators/optimizers/distributed_fused_lamb_op.cc
index f7b8dacfc5a..77d8682e335 100644
--- a/paddle/fluid/operators/optimizers/distributed_fused_lamb_op.cc
+++ b/paddle/fluid/operators/optimizers/distributed_fused_lamb_op.cc
@@ -151,8 +151,8 @@ class DistributedFusedLambOpMaker : public framework::OpProtoAndCheckerMaker {
                   "Whether the input gradient has been scaled by nranks.")
         .SetDefault(true);
     AddAttr<int64_t>("nranks", "The world size.").SetDefault(1);
-    AddAttr<std::vector<int>>("ring_id",
-                              "The ring id of the NCCL communicator.")
+    AddAttr<std::vector<int>>("ring_ids",
+                              "The ring ids of the NCCL communicator.")
         .SetDefault({0});
     AddAttr<bool>("use_hierarchical_allreduce",
                   "Whether to use hierarchical allreduce")
diff --git a/paddle/fluid/operators/optimizers/distributed_fused_lamb_op.cu b/paddle/fluid/operators/optimizers/distributed_fused_lamb_op.cu
index 908be3cd41d..0289e70d1da 100644
--- a/paddle/fluid/operators/optimizers/distributed_fused_lamb_op.cu
+++ b/paddle/fluid/operators/optimizers/distributed_fused_lamb_op.cu
@@ -1644,7 +1644,7 @@ class DistributedFusedLambOpKernel<phi::GPUContext, T>
             "The nranks must be exactly divided by num_devices."));
     bool local_shard = (nranks > num_devices);
 
-    const auto &ring_ids = ctx.Attr<std::vector<int>>("ring_id");
+    const auto &ring_ids = ctx.Attr<std::vector<int>>("ring_ids");
     auto use_master_param_norm = ctx.Attr<bool>("use_master_param_norm");
     auto is_grad_scaled_by_nranks = ctx.Attr<bool>("is_grad_scaled_by_nranks");
     auto use_hierarchical_allreduce =
diff --git a/python/paddle/incubate/optimizer/distributed_fused_lamb.py b/python/paddle/incubate/optimizer/distributed_fused_lamb.py
index 3964f431ac5..9a76db3be3d 100644
--- a/python/paddle/incubate/optimizer/distributed_fused_lamb.py
+++ b/python/paddle/incubate/optimizer/distributed_fused_lamb.py
@@ -479,7 +479,7 @@ class DistributedFusedLamb(Optimizer):
                 'clip_after_allreduce': self._clip_after_allreduce,
                 'rank': rank,
                 'nranks': nranks,
-                'ring_id': ring_ids,
+                'ring_ids': ring_ids,
                 'use_master_param_norm': self._use_master_param_norm,
                 'is_grad_scaled_by_nranks': self._is_grad_scaled_by_nranks,
                 'acc_steps': self._gradient_accumulation_steps,
-- 
GitLab