Remove selected rows all reduce over height check (#28460)

* remove slelected rows all reduce over height check

* polish unittest

paddle/fluid/imperative/all_reduce.cc

@@ -53,7 +53,7 @@ static void AllReduce(const framework::Tensor &src, framework::Tensor *dst,
 static void AllReduce(const framework::SelectedRows &src,
                       framework::SelectedRows *dst,
                       const ParallelStrategy &strategy, cudaStream_t stream) {
-  VLOG(0) << "SelectedRows AllReduce start";
+  VLOG(3) << "SelectedRows AllReduce start";
   const auto &src_tensor = src.value();
   const auto &place = src_tensor.place();
   PADDLE_ENFORCE_EQ(
@@ -87,18 +87,10 @@ static void AllReduce(const framework::SelectedRows &src,
                       static_cast<int64_t>(0));
   dst->set_height(src.height());
 
-  VLOG(0) << "Gather rows: " << string::join_strings(rows_num_vector, ',')
+  VLOG(3) << "Gather rows: " << string::join_strings(rows_num_vector, ',')
           << ", total rows number: " << rows_num
           << ", height: " << src.height();
 
-  PADDLE_ENFORCE_LE(
-      rows_num, src.height(),
-      platform::errors::Unimplemented(
-          "The gathered SelectedRows's rows number should less than or equal "
-          "to the SelectedRows's height, but the actual rows number is %d, the "
-          "SelectedRows's height is %d.",
-          rows_num, src.height()));
-
   auto *dst_rows = dst->mutable_rows();
   dst_rows->resize(rows_num);
   auto *dst_rows_ptr = dst_rows->CUDAMutableData(place);
@@ -130,9 +122,9 @@ static void AllReduce(const framework::SelectedRows &src,
     }
   }
 
-  VLOG(0) << "Original SelectedRows rows: "
+  VLOG(3) << "Original SelectedRows rows: "
           << string::join_strings(src_rows, ',');
-  VLOG(0) << "Result SelectedRows rows: "
+  VLOG(3) << "Result SelectedRows rows: "
           << string::join_strings(*dst_rows, ',');
 }
 #endif

python/paddle/fluid/tests/unittests/CMakeLists.txt

@@ -12,6 +12,7 @@ string(REPLACE ".py" "" DIST_TEST_OPS "${DIST_TEST_OPS}")
 list(APPEND DIST_TEST_OPS test_parallel_dygraph_mnist)
 list(APPEND DIST_TEST_OPS test_parallel_dygraph_se_resnext)
 list(APPEND DIST_TEST_OPS test_parallel_dygraph_sparse_embedding)
+list(APPEND DIST_TEST_OPS test_parallel_dygraph_sparse_embedding_over_height)
 list(APPEND DIST_TEST_OPS test_parallel_dygraph_transformer)
 list(APPEND DIST_TEST_OPS test_listen_and_serv_op)
 list(APPEND DIST_TEST_OPS test_fleet_graph_execution_meta_optimizer)
@@ -127,6 +128,7 @@ if (NOT ${WITH_GPU})
     LIST(REMOVE_ITEM TEST_OPS test_parallel_dygraph_mnist) # TODO(Yancey1989): parallel dygraph support CPU device in future
     list(REMOVE_ITEM TEST_OPS test_parallel_dygraph_se_resnext)
     LIST(REMOVE_ITEM TEST_OPS test_parallel_dygraph_sparse_embedding)
+    LIST(REMOVE_ITEM TEST_OPS test_parallel_dygraph_sparse_embedding_over_height)
     LIST(REMOVE_ITEM TEST_OPS test_parallel_dygraph_transformer)
     LIST(REMOVE_ITEM TEST_OPS test_parallel_dygraph_sync_batch_norm)
     LIST(REMOVE_ITEM TEST_OPS test_imperative_auto_mixed_precision)
@@ -139,6 +141,7 @@ endif()
 if (WITH_NCCL)
     if (${NCCL_VERSION} VERSION_LESS 2212)
         LIST(REMOVE_ITEM DIST_TEST_OPS test_parallel_dygraph_sparse_embedding)
+        LIST(REMOVE_ITEM DIST_TEST_OPS test_parallel_dygraph_sparse_embedding_over_height)
         LIST(REMOVE_ITEM DIST_TEST_OPS test_parallel_dygraph_transformer)
     endif()
 endif()

python/paddle/fluid/tests/unittests/parallel_dygraph_sparse_embedding_over_height.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import print_function
+
+import paddle
+import paddle.fluid as fluid
+from parallel_dygraph_sparse_embedding import SimpleNet, fake_sample_reader, TestSparseEmbedding
+
+from test_dist_base import runtime_main, TestParallelDyGraphRunnerBase
+
+# global configs
+# using small `vocab_size` to test rows number over height
+batch_size = 4
+batch_num = 200
+hidden_size = 10
+vocab_size = 10
+num_steps = 3
+init_scale = 0.1
+
+
+class TestSparseEmbeddingOverHeight(TestSparseEmbedding):
+    def get_model(self):
+        model = SimpleNet(
+            hidden_size=hidden_size,
+            vocab_size=vocab_size,
+            num_steps=num_steps,
+            init_scale=init_scale,
+            is_sparse=True)
+
+        train_reader = paddle.batch(
+            fake_sample_reader(), batch_size=batch_size, drop_last=True)
+
+        optimizer = fluid.optimizer.SGD(learning_rate=0.001,
+                                        parameter_list=model.parameters())
+
+        return model, train_reader, optimizer
+
+
+if __name__ == "__main__":
+    runtime_main(TestSparseEmbeddingOverHeight)

python/paddle/fluid/tests/unittests/test_parallel_dygraph_sparse_embedding_over_height.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import print_function
+
+import os
+import sys
+import unittest
+
+import paddle.fluid as fluid
+from test_dist_base import TestDistBase
+from spawn_runner_base import TestDistSpawnRunner
+from parallel_dygraph_sparse_embedding_over_height import TestSparseEmbeddingOverHeight
+
+flag_name = os.path.splitext(__file__)[0]
+
+
+class TestParallelDygraphSparseEmdeddingOverHeight(TestDistBase):
+    def _setup_config(self):
+        self._sync_mode = False
+        self._nccl2_mode = True
+        self._dygraph = True
+
+    def test_sparse_embedding(self):
+        if fluid.core.is_compiled_with_cuda():
+            self.check_with_place(
+                "parallel_dygraph_sparse_embedding_over_height.py",
+                delta=1e-5,
+                check_error_log=True,
+                log_name=flag_name)
+
+
+class TestParallelDygraphSparseEmdeddingOverHeightSpawn(TestDistSpawnRunner):
+    def test_sparse_embedding_with_spawn(self):
+        if fluid.core.is_compiled_with_cuda() and sys.version_info >= (3, 4):
+            self.check_dist_result_with_spawn(
+                test_class=TestSparseEmbeddingOverHeight, delta=1e-5)
+
+
+if __name__ == "__main__":
+    unittest.main()