【paddle.fleet】Add dgc to fleet meta optimizer (#25738)

Add dgc to fleet meta optimizer, rm dgc from optimizer all

paddle/fluid/framework/distributed_strategy.proto

@@ -43,6 +43,12 @@ message GradientMergeConfig {
   optional bool avg = 2 [ default = true ];
 }
 
+message DGCConfig {
+  optional int32 rampup_begin_step = 1 [ default = 0 ];
+  optional int32 rampup_step = 2 [ default = 1 ];
+  repeated float sparsity = 3;
+}
+
 message LarsConfig {
   optional float lars_coeff = 1 [ default = 0.001 ];
   optional float lars_weight_decay = 2 [ default = 0.0005 ];
@@ -114,6 +120,7 @@ message DistributedStrategy {
   optional AMPConfig amp_configs = 102;
   optional LocalSGDConfig localsgd_configs = 103;
   optional GradientMergeConfig gradient_merge_configs = 104;
+  optional DGCConfig dgc_configs = 105;
   optional PipelineConfig pipeline_configs = 106;
   optional AsyncConfig a_sync_configs = 107;
   optional LarsConfig lars_configs = 108;

python/paddle/fleet/base/distributed_strategy.py

@@ -604,6 +604,15 @@ class DistributedStrategy(object):
         else:
             print("WARNING: lars should have value of bool type")
 
+    @property
+    def lars_configs(self):
+        return get_msg_dict(self.strategy.lars_configs)
+
+    @lars_configs.setter
+    def lars_configs(self, configs):
+        check_configs_key(self.strategy.lars_configs, configs, "lars_configs")
+        assign_configs_value(self.strategy.lars_configs, configs)
+
     @property
     def lamb(self):
         return self.strategy.lamb

python/paddle/fleet/base/meta_optimizer_factory.py

@@ -19,6 +19,7 @@ from ..meta_optimizers import GraphExecutionOptimizer
 from ..meta_optimizers import PipelineOptimizer
 from ..meta_optimizers import LocalSGDOptimizer
 from ..meta_optimizers import LarsOptimizer
+from ..meta_optimizers import DGCOptimizer
 
 __all__ = ["MetaOptimizerFactory"]
 
@@ -30,6 +31,7 @@ meta_optimizer_names = [
     "PipelineOptimizer",
     "LocalSGDOptimizer",
     "LarsOptimizer",
+    "DGCOptimizer",
 ]
 
 

python/paddle/fleet/meta_optimizers/__init__.py

@@ -18,6 +18,7 @@ from .graph_execution_optimizer import GraphExecutionOptimizer
 from .pipeline_optimizer import PipelineOptimizer
 from .localsgd_optimizer import LocalSGDOptimizer
 from .lars_optimizer import LarsOptimizer
+from .dgc_optimizer import DGCOptimizer
 
 __all__ = [
     'AMPOptimizer',
@@ -26,4 +27,5 @@ __all__ = [
     'PipelineOptimizer',
     'LocalSGDOptimizer',
     'LarsOptimizer',
+    'DGCOptimizer',
 ]

python/paddle/fleet/meta_optimizers/dgc_optimizer.py

@@ -0,0 +1,94 @@
+#   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
+
+from paddle.fluid.optimizer import Momentum, DGCMomentumOptimizer
+from .meta_optimizer_base import MetaOptimizerBase
+import logging
+
+__all__ = ["DGCOptimizer"]
+
+
+class DGCOptimizer(MetaOptimizerBase):
+    def __init__(self, optimizer):
+        super(DGCOptimizer, self).__init__(optimizer)
+        self.inner_opt = optimizer
+        self.dgc_opt = None
+        # we do not allow meta optimizer to be inner optimizer currently
+        self.meta_optimizers_white_list = []
+
+    def _set_basic_info(self, loss, role_maker, user_defined_optimizer,
+                        user_defined_strategy):
+        super(DGCOptimizer, self)._set_basic_info(
+            loss, role_maker, user_defined_optimizer, user_defined_strategy)
+
+        opt = self.inner_opt
+        if not isinstance(opt, Momentum):
+            return
+
+        configs = self.user_defined_strategy.dgc_configs
+        if len(configs['sparsity']) == 0:
+            # default is [0.999]
+            configs['sparsity'] = [0.999]
+
+        self.dgc_opt = DGCMomentumOptimizer(
+            learning_rate=opt._learning_rate,
+            momentum=opt._momentum,
+            rampup_begin_step=configs['rampup_begin_step'],
+            rampup_step=configs['rampup_step'],
+            sparsity=configs['sparsity'],
+            parameter_list=opt._parameter_list,
+            use_nesterov=opt._use_nesterov,
+            num_trainers=self.role_maker.worker_num(),
+            regularization=opt.regularization,
+            grad_clip=opt._grad_clip,
+            name=opt._name)
+
+    def _can_apply(self):
+        if self.user_defined_strategy.dgc:
+            if not isinstance(self.inner_opt, Momentum):
+                logging.warn("dgc only works on Momentum optimizer")
+                return False
+            if self.role_maker.worker_num() <= 1:
+                logging.warn("dgc only works on multi cards")
+                return False
+
+            return True
+
+        return False
+
+    def _disable_strategy(self, dist_strategy):
+        dist_strategy.dgc = False
+        dist_strategy.dgc_configs = {
+            'rampup_begin_step': 0,
+            'rampup_step': 1,
+            'sparsity': [0.999]
+        }
+
+    def backward(self,
+                 loss,
+                 startup_program=None,
+                 parameter_list=None,
+                 no_grad_set=None,
+                 callbacks=None):
+        return self.dgc_opt.backward(loss, startup_program, parameter_list,
+                                     no_grad_set, callbacks)
+
+    def minimize_impl(self,
+                      loss,
+                      startup_program=None,
+                      parameter_list=None,
+                      no_grad_set=None):
+        optimize_ops, params_grads = \
+            self.dgc_opt.minimize(loss, startup_program,
+                                      parameter_list, no_grad_set)
+        return optimize_ops, params_grads

python/paddle/fleet/meta_optimizers/meta_optimizer_base.py

@@ -40,7 +40,8 @@ class MetaOptimizerBase(object):
             return True
 
     def _disable_strategy(self, dist_strategy):
-        raise NotImplementedError("you should implement disable strategy")
+        raise NotImplementedError("you should implement disable strategy in {}".
+                                  format(type(self).__name__))
 
     def minimize_impl(self,
                       loss,

python/paddle/fluid/optimizer.py

@@ -47,9 +47,8 @@ __all__ = [
     'AdamOptimizer', 'AdamaxOptimizer', 'DpsgdOptimizer',
     'DecayedAdagradOptimizer', 'RMSPropOptimizer', 'FtrlOptimizer', 'Adadelta',
     'AdadeltaOptimizer', 'ModelAverage', 'LarsMomentum',
-    'LarsMomentumOptimizer', 'DGCMomentumOptimizer', 'LambOptimizer',
-    'ExponentialMovingAverage', 'PipelineOptimizer', 'LookaheadOptimizer',
-    'RecomputeOptimizer'
+    'LarsMomentumOptimizer', 'LambOptimizer', 'ExponentialMovingAverage',
+    'PipelineOptimizer', 'LookaheadOptimizer', 'RecomputeOptimizer'
 ]
 
 

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

@@ -39,6 +39,7 @@ list(APPEND MIXED_DIST_TEST_OPS test_fleet_amp_meta_optimizer)
 list(APPEND MIXED_DIST_TEST_OPS test_fleet_gradient_merge_meta_optimizer)
 list(APPEND MIXED_DIST_TEST_OPS test_fleet_localsgd_meta_optimizer)
 list(APPEND MIXED_DIST_TEST_OPS test_fleet_lars_meta_optimizer)
+list(APPEND MIXED_DIST_TEST_OPS test_fleet_dgc_meta_optimizer)
 list(APPEND MIXED_DIST_TEST_OPS test_fleet_private_function)
 list(APPEND MIXED_DIST_TEST_OPS test_fleet_graph_executor)
 foreach(TEST_OP ${MIXED_DIST_TEST_OPS})
@@ -388,6 +389,7 @@ if(WITH_DISTRIBUTE)
         py_test_modules(test_dgc_op MODULES test_dgc_op)
         py_test_modules(test_dgc_momentum_op MODULES test_dgc_momentum_op)
         py_test_modules(test_dgc_optimizer MODULES test_dgc_optimizer)
+        py_test_modules(test_fleet_dgc_meta_optimizer MODULES test_fleet_dgc_meta_optimizer)
     else()
         # if not with dgc, must close all dgc tests
         list(REMOVE_ITEM DIST_TEST_OPS "test_dist_mnist_dgc_nccl")

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

@@ -39,7 +39,6 @@ class TestDistMnistNCCL2DGC(TestDistBase):
         self._nccl2_mode = True
         self._use_dgc = True
 
-    @unittest.skip(reason="Skip unstable ut")
     def test_dist_train(self):
         import paddle.fluid as fluid
         if fluid.core.is_compiled_with_cuda():
@@ -69,7 +68,6 @@ class TestDistMnistNCCL2DGCMultiCards(TestDistBase):
         self._nccl2_mode = True
         self._use_dgc = True
 
-    @unittest.skip(reason="Skip unstable ut")
     def test_dist_train(self):
         import paddle.fluid as fluid
         if fluid.core.is_compiled_with_cuda():

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

@@ -0,0 +1,86 @@
+# 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.
+
+import unittest
+import paddle
+import os
+import paddle.fleet as fleet
+import paddle.fluid.incubate.fleet.base.role_maker as role_maker
+
+
+class TestFleetDGCOptimizer(unittest.TestCase):
+    def setUp(self):
+        os.environ["PADDLE_TRAINER_ID"] = "1"
+        os.environ[
+            "PADDLE_TRAINER_ENDPOINTS"] = "127.0.0.1:36001,127.0.0.1:36002"
+
+    def net(self):
+        role = role_maker.PaddleCloudRoleMaker(is_collective=True)
+        fleet.init(role)
+        input_x = paddle.fluid.layers.data(
+            name="x", shape=[32], dtype='float32')
+        input_y = paddle.fluid.layers.data(name="y", shape=[1], dtype='int64')
+
+        fc_1 = paddle.fluid.layers.fc(input=input_x, size=64, act='tanh')
+        fc_2 = paddle.fluid.layers.fc(input=fc_1, size=256, act='tanh')
+        prediction = paddle.fluid.layers.fc(input=[fc_2], size=2, act='softmax')
+        cost = paddle.fluid.layers.cross_entropy(
+            input=prediction, label=input_y)
+        avg_cost = paddle.fluid.layers.mean(x=cost)
+
+        strategy = paddle.fleet.DistributedStrategy()
+        strategy.dgc = True
+        strategy.dgc_configs = {
+            "rampup_begin_step": 128,
+            "rampup_step": 100,
+            "sparsity": [0.996, 0.999]
+        }
+        return avg_cost, strategy
+
+    def test_dgc_optimizer(self):
+        avg_cost, strategy = self.net()
+        optimizer = paddle.optimizer.Momentum(learning_rate=0.01, momentum=0.9)
+        optimizer = fleet.distributed_optimizer(optimizer, strategy=strategy)
+        optimizer.minimize(avg_cost)
+
+        ops = [op.type for op in avg_cost.block.ops]
+        self.assertIn('dgc', ops)
+        self.assertIn('dgc_momentum', ops)
+
+    def test_dgc_not_apply_with_adam(self):
+        avg_cost, strategy = self.net()
+        optimizer = paddle.optimizer.Adam(learning_rate=0.01)
+        optimizer = fleet.distributed_optimizer(optimizer, strategy=strategy)
+        optimizer.minimize(avg_cost)
+
+        ops = [op.type for op in avg_cost.block.ops]
+        self.assertNotIn('dgc', ops)
+        self.assertNotIn('dgc_momentum', ops)
+
+    def test_dgc_not_apply_with_one_worker(self):
+        os.environ["PADDLE_TRAINER_ID"] = "0"
+        os.environ["PADDLE_TRAINER_ENDPOINTS"] = "127.0.0.1:36001"
+
+        avg_cost, strategy = self.net()
+        optimizer = paddle.optimizer.Momentum(learning_rate=0.01, momentum=0.9)
+        optimizer = fleet.distributed_optimizer(optimizer, strategy=strategy)
+        optimizer.minimize(avg_cost)
+
+        ops = [op.type for op in avg_cost.block.ops]
+        self.assertNotIn('dgc', ops)
+        self.assertNotIn('dgc_momentum', ops)
+
+
+if __name__ == "__main__":
+    unittest.main()