未验证 提交 0a1862d1 编写于 作者: W WangXi 提交者: GitHub

fleet combine amp dgc recompute meta optimizer (#27643)

上级 8fabb1c3
...@@ -744,13 +744,13 @@ class DistributedStrategy(object): ...@@ -744,13 +744,13 @@ class DistributedStrategy(object):
strategy.adaptive_localsgd = True # by default this is false strategy.adaptive_localsgd = True # by default this is false
""" """
return self.strategy.localsgd return self.strategy.adaptive_localsgd
@adaptive_localsgd.setter @adaptive_localsgd.setter
@is_strict_auto @is_strict_auto
def adaptive_localsgd(self, flag): def adaptive_localsgd(self, flag):
if isinstance(flag, bool): if isinstance(flag, bool):
self.strategy.localsgd = flag self.strategy.adaptive_localsgd = flag
else: else:
print("WARNING: adaptive_localsgd should have value of bool type") print("WARNING: adaptive_localsgd should have value of bool type")
......
...@@ -19,16 +19,14 @@ class AMPOptimizer(MetaOptimizerBase): ...@@ -19,16 +19,14 @@ class AMPOptimizer(MetaOptimizerBase):
def __init__(self, optimizer): def __init__(self, optimizer):
super(AMPOptimizer, self).__init__(optimizer) super(AMPOptimizer, self).__init__(optimizer)
self.inner_opt = optimizer self.inner_opt = optimizer
self.amp_opt = None self.wrapped_opt = None
# we do not allow meta optimizer to be inner optimizer currently # we do not allow meta optimizer to be inner optimizer currently
self.meta_optimizers_white_list = [ self.meta_optimizers_white_list = [
"LarsOptimizer", "LarsOptimizer",
"LambOptimizer", "LambOptimizer",
"RecomputeOptimizer", "RecomputeOptimizer",
"LocalSGDOptimizer",
"GradientMergeOptimizer", "GradientMergeOptimizer",
"GraphExecutionOptimizer", "GraphExecutionOptimizer",
"AdaptiveLocalSGDOptimizer",
] ]
self.meta_optimizers_black_list = ["DGCOptimizer"] self.meta_optimizers_black_list = ["DGCOptimizer"]
...@@ -37,6 +35,24 @@ class AMPOptimizer(MetaOptimizerBase): ...@@ -37,6 +35,24 @@ class AMPOptimizer(MetaOptimizerBase):
super(AMPOptimizer, self)._set_basic_info( super(AMPOptimizer, self)._set_basic_info(
loss, role_maker, user_defined_optimizer, user_defined_strategy) loss, role_maker, user_defined_optimizer, user_defined_strategy)
def _init_wrapped_opt(self):
if self.wrapped_opt is not None:
return
config = self.user_defined_strategy.amp_configs
custom_white_list = set(config['custom_white_list'])
custom_black_list = set(config['custom_black_list'])
custom_black_varnames = set(config['custom_black_varnames'])
amp_lists = mixed_precision.AutoMixedPrecisionLists(
custom_white_list, custom_black_list, custom_black_varnames)
self.wrapped_opt = mixed_precision.decorate(
self.inner_opt, amp_lists, config['init_loss_scaling'],
config['incr_every_n_steps'], config['decr_every_n_nan_or_inf'],
config['incr_ratio'], config['decr_ratio'],
config['use_dynamic_loss_scaling'])
def _can_apply(self): def _can_apply(self):
if not self.role_maker._is_collective: if not self.role_maker._is_collective:
return False return False
...@@ -60,26 +76,31 @@ class AMPOptimizer(MetaOptimizerBase): ...@@ -60,26 +76,31 @@ class AMPOptimizer(MetaOptimizerBase):
"use_dynamic_loss_scaling": True "use_dynamic_loss_scaling": True
} }
def backward(self,
loss,
startup_program=None,
parameter_list=None,
no_grad_set=None,
callbacks=None):
# maybe inner_opt of other meta optimizer
self._init_wrapped_opt()
return self.wrapped_opt.backward(loss, startup_program, parameter_list,
no_grad_set, callbacks)
def apply_gradients(self, params_grads):
return self.wrapped_opt.apply_gradients(params_grads=params_grads)
def apply_optimize(self, loss, startup_program, params_grads):
return self.wrapped_opt.apply_optimize(
loss, startup_program=startup_program, params_grads=params_grads)
def minimize_impl(self, def minimize_impl(self,
loss, loss,
startup_program=None, startup_program=None,
parameter_list=None, parameter_list=None,
no_grad_set=None): no_grad_set=None):
if self.amp_opt is None: self._init_wrapped_opt()
config = self.user_defined_strategy.amp_configs
custom_white_list = set(config['custom_white_list'])
custom_black_list = set(config['custom_black_list'])
custom_black_varnames = set(config['custom_black_varnames'])
amp_lists = mixed_precision.AutoMixedPrecisionLists(
custom_white_list, custom_black_list, custom_black_varnames)
self.amp_opt = mixed_precision.decorate(
self.inner_opt, amp_lists, config['init_loss_scaling'],
config['incr_every_n_steps'], config['decr_every_n_nan_or_inf'],
config['incr_ratio'], config['decr_ratio'],
config['use_dynamic_loss_scaling'])
optimize_ops, params_grads = \ optimize_ops, params_grads = \
self.amp_opt.minimize(loss, startup_program, self.wrapped_opt.minimize(loss, startup_program,
parameter_list, no_grad_set) parameter_list, no_grad_set)
return optimize_ops, params_grads return optimize_ops, params_grads
...@@ -85,6 +85,13 @@ class DGCOptimizer(MetaOptimizerBase): ...@@ -85,6 +85,13 @@ class DGCOptimizer(MetaOptimizerBase):
return self.dgc_opt.backward(loss, startup_program, parameter_list, return self.dgc_opt.backward(loss, startup_program, parameter_list,
no_grad_set, callbacks) no_grad_set, callbacks)
def apply_gradients(self, params_grads):
return self.dgc_opt.apply_gradients(params_grads=params_grads)
def apply_optimize(self, loss, startup_program, params_grads):
return self.dgc_opt.apply_optimize(
loss, startup_program=startup_program, params_grads=params_grads)
def minimize_impl(self, def minimize_impl(self,
loss, loss,
startup_program=None, startup_program=None,
......
...@@ -98,6 +98,10 @@ class LambOptimizer(MetaOptimizerBase): ...@@ -98,6 +98,10 @@ class LambOptimizer(MetaOptimizerBase):
def apply_gradients(self, params_grads): def apply_gradients(self, params_grads):
return self.lamb_opt.apply_gradients(params_grads=params_grads) return self.lamb_opt.apply_gradients(params_grads=params_grads)
def apply_optimize(self, loss, startup_program, params_grads):
return self.lamb_opt.apply_optimize(
loss, startup_program=startup_program, params_grads=params_grads)
def minimize_impl(self, def minimize_impl(self,
loss, loss,
startup_program=None, startup_program=None,
......
...@@ -85,6 +85,10 @@ class LarsOptimizer(MetaOptimizerBase): ...@@ -85,6 +85,10 @@ class LarsOptimizer(MetaOptimizerBase):
def apply_gradients(self, params_grads): def apply_gradients(self, params_grads):
return self.lars_opt.apply_gradients(params_grads=params_grads) return self.lars_opt.apply_gradients(params_grads=params_grads)
def apply_optimize(self, loss, startup_program, params_grads):
return self.lars_opt.apply_optimize(
loss, startup_program=startup_program, params_grads=params_grads)
def minimize_impl(self, def minimize_impl(self,
loss, loss,
startup_program=None, startup_program=None,
......
...@@ -24,7 +24,7 @@ class LocalSGDOptimizer(MetaOptimizerBase): ...@@ -24,7 +24,7 @@ class LocalSGDOptimizer(MetaOptimizerBase):
def __init__(self, optimizer): def __init__(self, optimizer):
super(LocalSGDOptimizer, self).__init__(optimizer) super(LocalSGDOptimizer, self).__init__(optimizer)
self.inner_opt = optimizer self.inner_opt = optimizer
self.meta_optimizers_white_list = [] self.meta_optimizers_white_list = ['AMPOptimizer']
self.meta_optimizers_black_list = [ self.meta_optimizers_black_list = [
"GraphExecutionOptimizer", "GraphExecutionOptimizer",
"AdaptiveLocalSGDOptimizer", "AdaptiveLocalSGDOptimizer",
...@@ -195,7 +195,7 @@ class AdaptiveLocalSGDOptimizer(MetaOptimizerBase): ...@@ -195,7 +195,7 @@ class AdaptiveLocalSGDOptimizer(MetaOptimizerBase):
def __init__(self, optimizer): def __init__(self, optimizer):
super(AdaptiveLocalSGDOptimizer, self).__init__(optimizer) super(AdaptiveLocalSGDOptimizer, self).__init__(optimizer)
self.inner_opt = optimizer self.inner_opt = optimizer
self.meta_optimizers_white_list = [] self.meta_optimizers_white_list = ['AMPOptimizer']
self.meta_optimizers_black_list = [ self.meta_optimizers_black_list = [
"GraphExecutionOptimizer", "LocalSGDOptimizer" "GraphExecutionOptimizer", "LocalSGDOptimizer"
] ]
......
...@@ -18,15 +18,14 @@ from .meta_optimizer_base import MetaOptimizerBase ...@@ -18,15 +18,14 @@ from .meta_optimizer_base import MetaOptimizerBase
class RecomputeOptimizer(MetaOptimizerBase): class RecomputeOptimizer(MetaOptimizerBase):
def __init__(self, optimizer): def __init__(self, optimizer):
super(RecomputeOptimizer, self).__init__(optimizer) super(RecomputeOptimizer, self).__init__(optimizer)
#self.inner_opt = RO(optimizer)
self.inner_opt = optimizer self.inner_opt = optimizer
self.wrapped_opt = RO(optimizer) self.wrapped_opt = None
# we do not allow meta optimizer to be inner optimizer currently # we do not allow meta optimizer to be inner optimizer currently
self.meta_optimizers_white_list = [ self.meta_optimizers_white_list = [
"LarsOptimizer", "LarsOptimizer",
"LambOptimizer", "LambOptimizer",
"GradientMergeOptimizer",
"GraphExecutionOptimizer", "GraphExecutionOptimizer",
"DGCOptimizer",
] ]
self.meta_optimizers_black_list = [] self.meta_optimizers_black_list = []
...@@ -34,8 +33,15 @@ class RecomputeOptimizer(MetaOptimizerBase): ...@@ -34,8 +33,15 @@ class RecomputeOptimizer(MetaOptimizerBase):
user_defined_strategy): user_defined_strategy):
super(RecomputeOptimizer, self)._set_basic_info( super(RecomputeOptimizer, self)._set_basic_info(
loss, role_maker, user_defined_optimizer, user_defined_strategy) loss, role_maker, user_defined_optimizer, user_defined_strategy)
self.wrapped_opt._set_checkpoints(
list(user_defined_strategy.recompute_configs["checkpoints"])) def _init_wrapped_opt(self):
if self.wrapped_opt is not None:
return
configs = self.user_defined_strategy.recompute_configs
self.wrapped_opt = RO(self.inner_opt)
self.wrapped_opt._set_checkpoints(list(configs["checkpoints"]))
def _can_apply(self): def _can_apply(self):
if not self.role_maker._is_collective: if not self.role_maker._is_collective:
...@@ -62,14 +68,24 @@ class RecomputeOptimizer(MetaOptimizerBase): ...@@ -62,14 +68,24 @@ class RecomputeOptimizer(MetaOptimizerBase):
parameter_list=None, parameter_list=None,
no_grad_set=None, no_grad_set=None,
callbacks=None): callbacks=None):
# maybe inner_opt of other meta optimizer
self._init_wrapped_opt()
return self.wrapped_opt.backward(loss, startup_program, parameter_list, return self.wrapped_opt.backward(loss, startup_program, parameter_list,
no_grad_set, callbacks) no_grad_set, callbacks)
def apply_gradients(self, params_grads):
return self.wrapped_opt.apply_gradients(params_grads=params_grads)
def apply_optimize(self, loss, startup_program, params_grads):
return self.wrapped_opt.apply_optimize(
loss, startup_program=startup_program, params_grads=params_grads)
def minimize_impl(self, def minimize_impl(self,
loss, loss,
startup_program=None, startup_program=None,
parameter_list=None, parameter_list=None,
no_grad_set=None): no_grad_set=None):
self._init_wrapped_opt()
optimize_ops, params_grads = \ optimize_ops, params_grads = \
self.wrapped_opt.minimize(loss, startup_program, self.wrapped_opt.minimize(loss, startup_program,
parameter_list, no_grad_set) parameter_list, no_grad_set)
......
...@@ -16,6 +16,7 @@ from ... import default_main_program ...@@ -16,6 +16,7 @@ from ... import default_main_program
from ... import default_startup_program from ... import default_startup_program
from ... import layers from ... import layers
from ... import unique_name from ... import unique_name
from ... import program_guard
from . import fp16_utils from . import fp16_utils
from .fp16_utils import rewrite_program from .fp16_utils import rewrite_program
from .fp16_utils import update_role_var_grad from .fp16_utils import update_role_var_grad
...@@ -58,21 +59,40 @@ class OptimizerWithMixedPrecision(object): ...@@ -58,21 +59,40 @@ class OptimizerWithMixedPrecision(object):
self._optimizer = optimizer self._optimizer = optimizer
self._amp_lists = amp_lists self._amp_lists = amp_lists
self._param_grads = None self._param_grads = None
self._train_program = default_main_program() self._train_program = None
self._startup_prog = default_startup_program()
self._scaled_loss = None self._scaled_loss = None
self._loss_scaling = layers.create_global_var( self._loss_scaling = None
name=unique_name.generate("loss_scaling"), self._init_loss_scaling = init_loss_scaling
shape=[1],
value=init_loss_scaling,
dtype='float32',
persistable=True)
self._use_dynamic_loss_scaling = use_dynamic_loss_scaling self._use_dynamic_loss_scaling = use_dynamic_loss_scaling
if self._use_dynamic_loss_scaling: if self._use_dynamic_loss_scaling:
self._incr_every_n_steps = incr_every_n_steps self._incr_every_n_steps = incr_every_n_steps
self._decr_every_n_nan_or_inf = decr_every_n_nan_or_inf self._decr_every_n_nan_or_inf = decr_every_n_nan_or_inf
self._incr_ratio = incr_ratio self._incr_ratio = incr_ratio
self._decr_ratio = decr_ratio self._decr_ratio = decr_ratio
self._num_good_steps = None
self._num_bad_steps = None
def get_loss_scaling(self):
"""Return the real-time loss scaling factor.
"""
return self._loss_scaling
def get_scaled_loss(self):
"""Return the scaled loss.
It's useful when you feed customed loss into executor.
"""
return self._scaled_loss
def _init_amp_var(self):
self._loss_scaling = layers.create_global_var(
name=unique_name.generate("loss_scaling"),
shape=[1],
value=self._init_loss_scaling,
dtype='float32',
persistable=True)
if self._use_dynamic_loss_scaling:
self._num_good_steps = layers.create_global_var( self._num_good_steps = layers.create_global_var(
name=unique_name.generate("num_good_steps"), name=unique_name.generate("num_good_steps"),
shape=[1], shape=[1],
...@@ -86,28 +106,16 @@ class OptimizerWithMixedPrecision(object): ...@@ -86,28 +106,16 @@ class OptimizerWithMixedPrecision(object):
dtype='int32', dtype='int32',
persistable=True) persistable=True)
# Ensure the data type of learning rate vars is float32 (same as the # Ensure the data type of learning rate vars is float32 (same as the
# master parameter dtype) # master parameter dtype)
if isinstance(optimizer._learning_rate, float): if isinstance(self._optimizer._learning_rate, float):
optimizer._learning_rate_map[default_main_program()] = \ self._optimizer._learning_rate_map[default_main_program()] = \
layers.create_global_var( layers.create_global_var(
name=unique_name.generate("learning_rate"), name=unique_name.generate("learning_rate"),
shape=[1], shape=[1],
value=float(optimizer._learning_rate), value=float(self._optimizer._learning_rate),
dtype='float32', dtype='float32',
persistable=True) persistable=True)
def get_loss_scaling(self):
"""Return the real-time loss scaling factor.
"""
return self._loss_scaling
def get_scaled_loss(self):
"""Return the scaled loss.
It's useful when you feed customed loss into executor.
"""
return self._scaled_loss
def backward(self, def backward(self,
loss, loss,
...@@ -131,16 +139,21 @@ class OptimizerWithMixedPrecision(object): ...@@ -131,16 +139,21 @@ class OptimizerWithMixedPrecision(object):
A list of (param, grad), which is a tuple of a parameter and its A list of (param, grad), which is a tuple of a parameter and its
gradient respectively, and the scaled loss. gradient respectively, and the scaled loss.
""" """
rewrite_program(self._train_program, self._amp_lists) train_program = loss.block.program
self._scaled_loss = loss * self._loss_scaling self._train_program = train_program
self._params_grads = self._optimizer.backward(
self._scaled_loss, startup_program, parameter_list, no_grad_set, with program_guard(train_program, startup_program):
callbacks) self._init_amp_var()
# Change the op_role_var attr for some ops, so that gradients
# transferred across GPUs can be FP16. rewrite_program(train_program, self._amp_lists)
update_role_var_grad(self._train_program, self._params_grads) self._scaled_loss = loss * self._loss_scaling
params_grads = self._optimizer.backward(
return self._params_grads self._scaled_loss, startup_program, parameter_list, no_grad_set,
callbacks)
# Change the op_role_var attr for some ops, so that gradients
# transferred across GPUs can be FP16.
update_role_var_grad(train_program, params_grads)
return params_grads
def apply_gradients(self, params_grads): def apply_gradients(self, params_grads):
""" """
...@@ -182,6 +195,12 @@ class OptimizerWithMixedPrecision(object): ...@@ -182,6 +195,12 @@ class OptimizerWithMixedPrecision(object):
return optimize_ops return optimize_ops
def apply_optimize(self, loss, startup_program, params_grads):
program = loss.block.program
with program_guard(program, startup_program):
optimize_ops = self.apply_gradients(params_grads)
return optimize_ops
def minimize(self, def minimize(self,
loss, loss,
startup_program=None, startup_program=None,
...@@ -207,7 +226,8 @@ class OptimizerWithMixedPrecision(object): ...@@ -207,7 +226,8 @@ class OptimizerWithMixedPrecision(object):
parameter_list=parameter_list, parameter_list=parameter_list,
no_grad_set=no_grad_set) no_grad_set=no_grad_set)
optimize_ops = self.apply_gradients(scaled_params_grads) optimize_ops = self.apply_optimize(loss, startup_program,
scaled_params_grads)
return optimize_ops, scaled_params_grads return optimize_ops, scaled_params_grads
......
...@@ -731,9 +731,6 @@ class Optimizer(object): ...@@ -731,9 +731,6 @@ class Optimizer(object):
outputs={"ParamOut": param_and_grad[0]}) outputs={"ParamOut": param_and_grad[0]})
return new_param_grads, (table_param, table_grad), sgd_op return new_param_grads, (table_param, table_grad), sgd_op
def _append_dgc_ops(self, param_and_grad):
pass
def backward(self, def backward(self,
loss, loss,
startup_program=None, startup_program=None,
...@@ -801,9 +798,6 @@ class Optimizer(object): ...@@ -801,9 +798,6 @@ class Optimizer(object):
with program_guard(program, startup_program): with program_guard(program, startup_program):
params_grads = append_backward(loss, parameter_list, params_grads = append_backward(loss, parameter_list,
act_no_grad_set, callbacks) act_no_grad_set, callbacks)
# Note: since we can't use all_reduce_op now,
# dgc_op should be the last op of one grad.
self._append_dgc_ops(params_grads)
return params_grads return params_grads
def apply_gradients(self, params_grads): def apply_gradients(self, params_grads):
...@@ -1569,6 +1563,11 @@ class DGCMomentumOptimizer(Optimizer): ...@@ -1569,6 +1563,11 @@ class DGCMomentumOptimizer(Optimizer):
@imperative_base.no_grad @imperative_base.no_grad
def apply_gradients(self, params_grads): def apply_gradients(self, params_grads):
# Note: since we can't use all_reduce_op now,
# dgc_op should be the last op of one grad.
# Maybe need a grad allreduce pass.
self._append_dgc_ops(params_grads)
params_grads = sorted(params_grads, key=lambda x: x[0].name) params_grads = sorted(params_grads, key=lambda x: x[0].name)
params_grads, table_param_and_grad, table_optimize_op = \ params_grads, table_param_and_grad, table_optimize_op = \
self._process_distribute_lookuptable(params_grads) self._process_distribute_lookuptable(params_grads)
...@@ -4784,10 +4783,6 @@ class RecomputeOptimizer(Optimizer): ...@@ -4784,10 +4783,6 @@ class RecomputeOptimizer(Optimizer):
params_grads = append_backward( params_grads = append_backward(
loss, parameter_list, no_grad_set, checkpoints=checkpoint_vars) loss, parameter_list, no_grad_set, checkpoints=checkpoint_vars)
# Note: since we can't use all_reduce_op now,
# dgc_op should be the last op of one grad.
if hasattr(self._optimizer, "_append_dgc_ops"):
self._optimizer._append_dgc_ops(params_grads)
return params_grads return params_grads
def apply_optimize(self, loss, startup_program, params_grads): def apply_optimize(self, loss, startup_program, params_grads):
......
# 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
from paddle import fluid
import os
import paddle.distributed.fleet as fleet
import paddle.distributed.fleet.base.role_maker as role_maker
class TestFleetMetaOptimizer(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, main_prog, startup_prog):
with fluid.program_guard(main_prog, startup_prog):
with fluid.unique_name.guard():
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.distributed.fleet.DistributedStrategy()
return avg_cost, strategy
def optimizer(self,
loss,
strategy,
train_prog,
startup_prog,
name='momentum'):
with fluid.program_guard(train_prog, startup_prog):
with fluid.unique_name.guard():
if name == 'momentum':
optimizer = paddle.fluid.optimizer.Momentum(
learning_rate=0.01, momentum=0.9)
elif name == 'adam':
optimizer = paddle.fluid.optimizer.Adam(learning_rate=0.01)
optimizer = fleet.distributed_optimizer(
optimizer, strategy=strategy)
optimizer.minimize(loss)
def set_strategy(self, strategy, name):
if name == 'amp':
strategy.amp = True
strategy.amp_configs = {
"init_loss_scaling": 32768,
"decr_every_n_nan_or_inf": 2,
"incr_every_n_steps": 1000,
"incr_ratio": 2.0,
"use_dynamic_loss_scaling": True,
"decr_ratio": 0.5,
"custom_white_list": ['softmax'],
"custom_black_list": ['tanh'],
}
elif name == 'dgc':
strategy.dgc = True
strategy.dgc_configs = {
"rampup_begin_step": 128,
"rampup_step": 100,
"sparsity": [0.996, 0.999]
}
elif name == 'recompute':
strategy.recompute = True
strategy.recompute_configs = {
"checkpoints": ["fc_0.tmp_2", "fc_1.tmp_2"]
}
elif name == 'lars':
strategy.lars = True
strategy.lars_configs = {
"lars_coeff": 0.001,
"lars_weight_decay": 0.0005,
"epsilon": 0,
"exclude_from_weight_decay": ["batch_norm", ".b"],
}
elif name == 'lamb':
strategy.lamb = True
strategy.lamb_configs = {
'lamb_weight_decay': 0.01,
'exclude_from_weight_decay': [],
}
elif name == 'localsgd':
strategy.localsgd = True
strategy.localsgd_configs = {
'k_steps': 1,
'begin_step': 1,
}
elif name == 'adaptive_localsgd':
strategy.adaptive_localsgd = True
strategy.adaptive_localsgd_configs = {
'init_k_steps': 1,
'begin_step': 1,
}
else:
raise NotImplementedError()
...@@ -16,12 +16,14 @@ from __future__ import print_function ...@@ -16,12 +16,14 @@ from __future__ import print_function
import unittest import unittest
import paddle
import paddle.fluid.framework as framework import paddle.fluid.framework as framework
import paddle.fluid.optimizer as optimizer import paddle.fluid.optimizer as optimizer
import paddle.fluid.regularizer as regularizer import paddle.fluid.regularizer as regularizer
import paddle.fluid.clip as clip import paddle.fluid.clip as clip
import paddle.compat as cpt import paddle.compat as cpt
from paddle.fluid.backward import append_backward from paddle.fluid.backward import append_backward
paddle.enable_static()
class TestDGCMomentumOptimizer(unittest.TestCase): class TestDGCMomentumOptimizer(unittest.TestCase):
...@@ -86,13 +88,17 @@ class TestDGCMomentumOptimizer(unittest.TestCase): ...@@ -86,13 +88,17 @@ class TestDGCMomentumOptimizer(unittest.TestCase):
block.append_op( block.append_op(
type="mean", inputs={"X": mul_out}, outputs={"Out": mean_out}) type="mean", inputs={"X": mul_out}, outputs={"Out": mean_out})
# params_grads = append_backward(mean_out) # params_grads = append_backward(mean_out)
params_grads = dgc_momentum_optimizer.backward(mean_out) params_grads = dgc_momentum_optimizer.backward(
mean_out, startup_program=init_program)
with framework.program_guard(program, init_program):
opts = dgc_momentum_optimizer.apply_gradients(params_grads)
accumulator_count = 1 if name == "momentum" else 2 accumulator_count = 1 if name == "momentum" else 2
self.assertEqual(len(params_grads), 1) self.assertEqual(len(params_grads), 1)
self.assertEqual( self.assertEqual(
len(dgc_momentum_optimizer.get_accumulators()), accumulator_count) len(dgc_momentum_optimizer.get_accumulators()), accumulator_count)
with framework.program_guard(program, init_program):
opts = dgc_momentum_optimizer.apply_gradients(params_grads)
self.assertEqual(len(opts), 2) self.assertEqual(len(opts), 2)
sgd_op = opts[-1] sgd_op = opts[-1]
self.assertEqual([op.type for op in opts], ["scale", name]) self.assertEqual([op.type for op in opts], ["scale", name])
...@@ -108,8 +114,11 @@ class TestDGCMomentumOptimizer(unittest.TestCase): ...@@ -108,8 +114,11 @@ class TestDGCMomentumOptimizer(unittest.TestCase):
self.assertTrue(mul_x.name in velocity_acc) self.assertTrue(mul_x.name in velocity_acc)
# Check init_program # Check init_program
# dgc not apply include: lr, dgc(count, nranks, begin step), (u,)
# dgc apply include: lr, dgc(count, nranks, begin_step), (u,v,k,encode,gather)
init_ops_count = 5 if name == "momentum" else 9
init_ops = init_program.global_block().ops init_ops = init_program.global_block().ops
self.assertEqual(len(init_ops), 1) self.assertEqual(len(init_ops), init_ops_count)
self.assertEqual(init_ops[0].type, "fill_constant") self.assertEqual(init_ops[0].type, "fill_constant")
self.assertAlmostEqual(init_ops[0].attr('value'), learning_rate) self.assertAlmostEqual(init_ops[0].attr('value'), learning_rate)
......
...@@ -12,57 +12,97 @@ ...@@ -12,57 +12,97 @@
# See the License for the specific language governing permissions and # See the License for the specific language governing permissions and
# limitations under the License. # limitations under the License.
import paddle.distributed.fleet as fleet
import paddle.distributed.fleet.base.role_maker as role_maker
import unittest import unittest
import paddle import paddle
import paddle.fluid as fluid
import paddle.distributed.fleet as fleet
from paddle.distributed.fleet.meta_optimizers import AMPOptimizer
import os import os
from fleet_meta_optimizer_base import TestFleetMetaOptimizer
paddle.enable_static() paddle.enable_static()
class TestFleetAMPOptimizer(unittest.TestCase): class TestFleetAMPOptimizer(TestFleetMetaOptimizer):
def setUp(self): def test_amp_optimizer_backward(self):
os.environ["PADDLE_TRAINER_ID"] = "0" """ test amp optimizer backward """
os.environ["PADDLE_TRAINER_ENDPOINTS"] = "127.0.0.1:36001" train_prog, startup_prog = fluid.Program(), fluid.Program()
avg_cost, strategy = self.net(train_prog, startup_prog)
opt = fluid.optimizer.MomentumOptimizer(
learning_rate=0.001, momentum=0.9)
opt = AMPOptimizer(opt)
opt.user_defined_strategy = strategy
params_grads = opt.backward(avg_cost, startup_prog)
ops = [op.type for op in avg_cost.block.ops]
self.assertIn('cast', ops)
self.assertNotIn('check_finite_and_unscale', ops)
def test_amp_optimizer_backward_gradients(self):
""" test amp optimizer backward + gradients"""
train_prog, startup_prog = fluid.Program(), fluid.Program()
avg_cost, strategy = self.net(train_prog, startup_prog)
opt = fluid.optimizer.MomentumOptimizer(
learning_rate=0.001, momentum=0.9)
opt = AMPOptimizer(opt)
opt.user_defined_strategy = strategy
params_grads = opt.backward(avg_cost, startup_prog)
with fluid.program_guard(train_prog, startup_prog):
opt.apply_gradients(params_grads)
ops = [op.type for op in avg_cost.block.ops]
self.assertIn('cast', ops)
self.assertIn('check_finite_and_unscale', ops)
def test_amp_optimizer_backward_optimize(self):
""" test amp optimizer backward + optimizer """
train_prog, startup_prog = fluid.Program(), fluid.Program()
avg_cost, strategy = self.net(train_prog, startup_prog)
opt = fluid.optimizer.MomentumOptimizer(
learning_rate=0.001, momentum=0.9)
opt = AMPOptimizer(opt)
opt.user_defined_strategy = strategy
params_grads = opt.backward(avg_cost, startup_prog)
opt.apply_optimize(avg_cost, startup_prog, params_grads)
ops = [op.type for op in avg_cost.block.ops]
self.assertIn('cast', ops)
self.assertIn('check_finite_and_unscale', ops)
def test_amp_optimizer(self): def test_amp_optimizer(self):
role = role_maker.PaddleCloudRoleMaker(is_collective=True) """ test amp """
fleet.init(role) train_prog, startup_prog = fluid.Program(), fluid.Program()
input_x = paddle.fluid.layers.data( avg_cost, strategy = self.net(train_prog, startup_prog)
name="x", shape=[32], dtype='float32') self.set_strategy(strategy, 'amp')
input_y = paddle.fluid.layers.data(name="y", shape=[1], dtype='int64') self.optimizer(avg_cost, strategy, train_prog, startup_prog)
fc_1 = paddle.fluid.layers.fc(input=input_x, size=64, act='tanh') ops = [op.type for op in avg_cost.block.ops]
fc_2 = paddle.fluid.layers.fc(input=fc_1, size=64, act='tanh') self.assertIn('cast', ops)
prediction = paddle.fluid.layers.fc(input=[fc_2], size=2, act='softmax') self.assertIn('check_finite_and_unscale', ops)
cost = paddle.fluid.layers.cross_entropy(
input=prediction, label=input_y) def test_amp_recompute_optimizer(self):
avg_cost = paddle.fluid.layers.mean(x=cost) """ test amp + recompute """
train_prog, startup_prog = fluid.Program(), fluid.Program()
strategy = paddle.distributed.fleet.DistributedStrategy() avg_cost, strategy = self.net(train_prog, startup_prog)
strategy.amp = True self.set_strategy(strategy, 'amp')
strategy.amp_configs = { self.set_strategy(strategy, 'recompute')
"init_loss_scaling": 32768, self.optimizer(avg_cost, strategy, train_prog, startup_prog)
"decr_every_n_nan_or_inf": 2,
"incr_every_n_steps": 1000,
"incr_ratio": 2.0,
"use_dynamic_loss_scaling": True,
"decr_ratio": 0.5,
"custom_white_list": ['softmax'],
"custom_black_list": ['tanh'],
}
optimizer = paddle.fluid.optimizer.SGD(learning_rate=0.01)
optimizer = fleet.distributed_optimizer(optimizer, strategy=strategy)
optimizer.minimize(avg_cost)
strategy = fleet._final_strategy() strategy = fleet._final_strategy()
ops = [op.type for op in avg_cost.block.ops] ops = [op.type for op in avg_cost.block.ops]
outs = [
op.output('Out')[0] for op in avg_cost.block.ops if op.type == 'mul'
]
self.assertIn('cast', ops) self.assertIn('cast', ops)
self.assertIn('check_finite_and_unscale', ops) self.assertIn('check_finite_and_unscale', ops)
# recompute
self.assertIn('subprog', ''.join(outs))
if __name__ == "__main__": if __name__ == "__main__":
unittest.main() unittest.main()
...@@ -17,65 +17,82 @@ import paddle ...@@ -17,65 +17,82 @@ import paddle
from paddle import fluid from paddle import fluid
import os import os
import paddle.distributed.fleet as fleet import paddle.distributed.fleet as fleet
from fleet_meta_optimizer_base import TestFleetMetaOptimizer
from paddle.distributed.fleet.meta_optimizers import DGCOptimizer
import paddle.distributed.fleet.base.role_maker as role_maker import paddle.distributed.fleet.base.role_maker as role_maker
paddle.enable_static()
class TestFleetDGCOptimizer(unittest.TestCase):
def setUp(self): class TestFleetDGCOptimizer(TestFleetMetaOptimizer):
os.environ["PADDLE_TRAINER_ID"] = "1" def test_dgc_optimizer_backward(self):
os.environ[ """ test dgc optimizer backward """
"PADDLE_TRAINER_ENDPOINTS"] = "127.0.0.1:36001,127.0.0.1:36002" train_prog, startup_prog = fluid.Program(), fluid.Program()
avg_cost, strategy = self.net(train_prog, startup_prog)
def net(self, main_prog, startup_prog):
with fluid.program_guard(main_prog, startup_prog): self.set_strategy(strategy, 'dgc')
with fluid.unique_name.guard(): opt = fluid.optimizer.MomentumOptimizer(
role = role_maker.PaddleCloudRoleMaker(is_collective=True) learning_rate=0.001, momentum=0.9)
fleet.init(role) dgc_opt = DGCOptimizer(opt)
input_x = paddle.fluid.layers.data( role = role_maker.PaddleCloudRoleMaker(is_collective=True)
name="x", shape=[32], dtype='float32') dgc_opt._set_basic_info(avg_cost, role, opt, strategy)
input_y = paddle.fluid.layers.data( params_grads = dgc_opt.backward(avg_cost, startup_prog)
name="y", shape=[1], dtype='int64')
ops = [op.type for op in avg_cost.block.ops]
fc_1 = paddle.fluid.layers.fc(input=input_x, self.assertNotIn('dgc', ops)
size=64,
act='tanh') def test_dgc_optimizer_gradients(self):
fc_2 = paddle.fluid.layers.fc(input=fc_1, size=256, act='tanh') """ test dgc optimizer backward + gradients """
prediction = paddle.fluid.layers.fc(input=[fc_2], train_prog, startup_prog = fluid.Program(), fluid.Program()
size=2, avg_cost, strategy = self.net(train_prog, startup_prog)
act='softmax')
cost = paddle.fluid.layers.cross_entropy( self.set_strategy(strategy, 'dgc')
input=prediction, label=input_y) opt = fluid.optimizer.MomentumOptimizer(
avg_cost = paddle.fluid.layers.mean(x=cost) learning_rate=0.001, momentum=0.9)
dgc_opt = DGCOptimizer(opt)
strategy = paddle.distributed.fleet.DistributedStrategy() role = role_maker.PaddleCloudRoleMaker(is_collective=True)
strategy.dgc = True dgc_opt._set_basic_info(avg_cost, role, opt, strategy)
strategy.dgc_configs = { params_grads = dgc_opt.backward(avg_cost, startup_prog)
"rampup_begin_step": 128, with fluid.program_guard(train_prog, startup_prog):
"rampup_step": 100, dgc_opt.apply_gradients(params_grads)
"sparsity": [0.996, 0.999]
} ops = [op.type for op in avg_cost.block.ops]
return avg_cost, strategy self.assertIn('dgc', ops)
self.assertIn('dgc_momentum', ops)
def test_dgc_optimizer_optimize(self):
""" test dgc optimizer backward + optimize """
train_prog, startup_prog = fluid.Program(), fluid.Program()
avg_cost, strategy = self.net(train_prog, startup_prog)
self.set_strategy(strategy, 'dgc')
opt = fluid.optimizer.MomentumOptimizer(
learning_rate=0.001, momentum=0.9)
dgc_opt = DGCOptimizer(opt)
role = role_maker.PaddleCloudRoleMaker(is_collective=True)
dgc_opt._set_basic_info(avg_cost, role, opt, strategy)
params_grads = dgc_opt.backward(avg_cost, startup_prog)
dgc_opt.apply_optimize(avg_cost, startup_prog, params_grads)
ops = [op.type for op in avg_cost.block.ops]
self.assertIn('dgc', ops)
self.assertIn('dgc_momentum', ops)
def test_dgc_optimizer(self): def test_dgc_optimizer(self):
startup_prog = fluid.Program() train_prog, startup_prog = fluid.Program(), fluid.Program()
train_prog = fluid.Program()
avg_cost, strategy = self.net(train_prog, startup_prog) avg_cost, strategy = self.net(train_prog, startup_prog)
optimizer = paddle.fluid.optimizer.Momentum( self.set_strategy(strategy, 'dgc')
learning_rate=0.01, momentum=0.9) self.optimizer(avg_cost, strategy, train_prog, startup_prog)
optimizer = fleet.distributed_optimizer(optimizer, strategy=strategy)
optimizer.minimize(avg_cost)
ops = [op.type for op in avg_cost.block.ops] ops = [op.type for op in avg_cost.block.ops]
self.assertIn('dgc', ops) self.assertIn('dgc', ops)
self.assertIn('dgc_momentum', ops) self.assertIn('dgc_momentum', ops)
def test_dgc_not_apply_with_adam(self): def test_dgc_not_apply_with_adam(self):
startup_prog = fluid.Program() train_prog, startup_prog = fluid.Program(), fluid.Program()
train_prog = fluid.Program()
avg_cost, strategy = self.net(train_prog, startup_prog) avg_cost, strategy = self.net(train_prog, startup_prog)
optimizer = paddle.fluid.optimizer.Adam(learning_rate=0.01) self.set_strategy(strategy, 'dgc')
optimizer = fleet.distributed_optimizer(optimizer, strategy=strategy) self.optimizer(avg_cost, strategy, train_prog, startup_prog, 'adam')
optimizer.minimize(avg_cost)
ops = [op.type for op in avg_cost.block.ops] ops = [op.type for op in avg_cost.block.ops]
self.assertNotIn('dgc', ops) self.assertNotIn('dgc', ops)
...@@ -85,18 +102,32 @@ class TestFleetDGCOptimizer(unittest.TestCase): ...@@ -85,18 +102,32 @@ class TestFleetDGCOptimizer(unittest.TestCase):
os.environ["PADDLE_TRAINER_ID"] = "0" os.environ["PADDLE_TRAINER_ID"] = "0"
os.environ["PADDLE_TRAINER_ENDPOINTS"] = "127.0.0.1:36001" os.environ["PADDLE_TRAINER_ENDPOINTS"] = "127.0.0.1:36001"
startup_prog = fluid.Program() train_prog, startup_prog = fluid.Program(), fluid.Program()
train_prog = fluid.Program()
avg_cost, strategy = self.net(train_prog, startup_prog) avg_cost, strategy = self.net(train_prog, startup_prog)
optimizer = paddle.fluid.optimizer.Momentum( self.set_strategy(strategy, 'dgc')
learning_rate=0.01, momentum=0.9) self.optimizer(avg_cost, strategy, train_prog, startup_prog)
optimizer = fleet.distributed_optimizer(optimizer, strategy=strategy)
optimizer.minimize(avg_cost)
ops = [op.type for op in avg_cost.block.ops] ops = [op.type for op in avg_cost.block.ops]
self.assertNotIn('dgc', ops) self.assertNotIn('dgc', ops)
self.assertNotIn('dgc_momentum', ops) self.assertNotIn('dgc_momentum', ops)
def test_dgc_recompute_optimizer(self):
train_prog, startup_prog = fluid.Program(), fluid.Program()
avg_cost, strategy = self.net(train_prog, startup_prog)
self.set_strategy(strategy, 'dgc')
self.set_strategy(strategy, 'recompute')
self.optimizer(avg_cost, strategy, train_prog, startup_prog)
ops = [op.type for op in avg_cost.block.ops]
outs = [
op.output('Out')[0] for op in avg_cost.block.ops if op.type == 'mul'
]
self.assertIn('dgc', ops)
self.assertIn('dgc_momentum', ops)
# recompute
self.assertIn('subprog', ''.join(outs))
if __name__ == "__main__": if __name__ == "__main__":
unittest.main() unittest.main()
...@@ -16,71 +16,87 @@ import unittest ...@@ -16,71 +16,87 @@ import unittest
import paddle import paddle
import os import os
import paddle
import paddle.fluid as fluid
import paddle.distributed.fleet as fleet import paddle.distributed.fleet as fleet
import paddle.distributed.fleet.base.role_maker as role_maker import paddle.distributed.fleet.base.role_maker as role_maker
from fleet_meta_optimizer_base import TestFleetMetaOptimizer
paddle.enable_static()
class TestFleetLocalSGDMetaOptimizer(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"
class TestFleetLocalSGDMetaOptimizer(TestFleetMetaOptimizer):
def test_localsgd_optimizer(self): def test_localsgd_optimizer(self):
role = role_maker.PaddleCloudRoleMaker(is_collective=True) train_prog, startup_prog = fluid.Program(), fluid.Program()
fleet.init(role) avg_cost, strategy = self.net(train_prog, startup_prog)
input_x = paddle.fluid.layers.data( self.set_strategy(strategy, 'localsgd')
name="x", shape=[32], dtype='float32') self.optimizer(avg_cost, strategy, train_prog, startup_prog)
input_y = paddle.fluid.layers.data(name="y", shape=[1], dtype='int64')
ops = [op.type for op in avg_cost.block.ops]
fc = paddle.fluid.layers.fc(input=input_x, size=64, act='tanh') outs = [
prediction = paddle.fluid.layers.fc(input=[fc], size=2, act='softmax') ''.join(op.output('Out')) for op in avg_cost.block.ops
cost = paddle.fluid.layers.cross_entropy( if op.type == 'conditional_block'
input=prediction, label=input_y) ]
avg_cost = paddle.fluid.layers.mean(x=cost)
self.assertIn('conditional_block', ops)
strategy = paddle.distributed.fleet.DistributedStrategy() self.assertIn('@SNAPSHOT', ''.join(outs))
strategy.localsgd = True
strategy.auto = True def test_localsgd_amp_optimizer(self):
config = strategy.localsgd_configs train_prog, startup_prog = fluid.Program(), fluid.Program()
config['k_steps'] = 1 avg_cost, strategy = self.net(train_prog, startup_prog)
config['begin_step'] = 1 self.set_strategy(strategy, 'localsgd')
strategy.localsgd_configs = config self.set_strategy(strategy, 'amp')
self.optimizer(avg_cost, strategy, train_prog, startup_prog)
optimizer = paddle.fluid.optimizer.SGD(learning_rate=0.01)
optimizer = fleet.distributed_optimizer(optimizer, strategy=strategy) ops = [op.type for op in avg_cost.block.ops]
optimizer.minimize(avg_cost) outs = [
''.join(op.output('Out')) for op in avg_cost.block.ops
if op.type == 'conditional_block'
class TestFleetAdaptiveLocalSGDMetaOptimizer(unittest.TestCase): ]
def setUp(self):
os.environ["PADDLE_TRAINER_ID"] = "1" self.assertIn('conditional_block', ops)
os.environ[ self.assertIn('@SNAPSHOT', ''.join(outs))
"PADDLE_TRAINER_ENDPOINTS"] = "127.0.0.1:36001,127.0.0.1:36002"
# amp
self.assertIn('cast', ops)
self.assertIn('check_finite_and_unscale', ops)
class TestFleetAdaptiveLocalSGDMetaOptimizer(TestFleetMetaOptimizer):
def test_adaptive_localsgd_optimizer(self): def test_adaptive_localsgd_optimizer(self):
role = role_maker.PaddleCloudRoleMaker(is_collective=True) train_prog, startup_prog = fluid.Program(), fluid.Program()
fleet.init(role) avg_cost, strategy = self.net(train_prog, startup_prog)
input_x = paddle.fluid.layers.data( self.set_strategy(strategy, 'adaptive_localsgd')
name="x", shape=[32], dtype='float32') self.optimizer(avg_cost, strategy, train_prog, startup_prog)
input_y = paddle.fluid.layers.data(name="y", shape=[1], dtype='int64')
ops = [op.type for op in avg_cost.block.ops]
fc = paddle.fluid.layers.fc(input=input_x, size=64, act='tanh') outs = [
prediction = paddle.fluid.layers.fc(input=[fc], size=2, act='softmax') ''.join(op.output('Out')) for op in avg_cost.block.ops
cost = paddle.fluid.layers.cross_entropy( if op.type == 'conditional_block'
input=prediction, label=input_y) ]
avg_cost = paddle.fluid.layers.mean(x=cost)
self.assertIn('conditional_block', ops)
strategy = paddle.distributed.fleet.DistributedStrategy() self.assertIn('@SNAPSHOT', ''.join(outs))
strategy.adaptive_localsgd = True
config = strategy.adaptive_localsgd_configs def test_localsgd_amp_optimizer(self):
config['init_k_steps'] = 1 train_prog, startup_prog = fluid.Program(), fluid.Program()
config['begin_step'] = 1 avg_cost, strategy = self.net(train_prog, startup_prog)
strategy.adaptive_localsgd_configs = config self.set_strategy(strategy, 'adaptive_localsgd')
self.set_strategy(strategy, 'amp')
optimizer = paddle.fluid.optimizer.SGD(learning_rate=0.01) self.optimizer(avg_cost, strategy, train_prog, startup_prog)
optimizer = fleet.distributed_optimizer(optimizer, strategy=strategy)
optimizer.minimize(avg_cost) ops = [op.type for op in avg_cost.block.ops]
outs = [
''.join(op.output('Out')) for op in avg_cost.block.ops
if op.type == 'conditional_block'
]
self.assertIn('conditional_block', ops)
self.assertIn('@SNAPSHOT', ''.join(outs))
# amp
self.assertIn('cast', ops)
self.assertIn('check_finite_and_unscale', ops)
if __name__ == "__main__": if __name__ == "__main__":
......
...@@ -14,40 +14,144 @@ ...@@ -14,40 +14,144 @@
import unittest import unittest
import paddle import paddle
import paddle.fluid as fluid
import os import os
from fleet_meta_optimizer_base import TestFleetMetaOptimizer
from paddle.distributed.fleet.meta_optimizers import RecomputeOptimizer
paddle.enable_static()
class TestFleetRecomputeMetaOptimizer(unittest.TestCase):
def setUp(self): class TestFleetRecomputeMetaOptimizer(TestFleetMetaOptimizer):
os.environ["POD_IP"] = "127.0.0.1" def test_recompute_optimizer_backward(self):
os.environ["PADDLE_TRAINER_ENDPOINTS"] = "127.0.0.1:36001" """ test recompute optimizer backward """
os.environ["PADDLE_TRAINERS_NUM"] = "2" train_prog, startup_prog = fluid.Program(), fluid.Program()
os.environ["PADDLE_PSERVERS_IP_PORT_LIST"] = \ avg_cost, strategy = self.net(train_prog, startup_prog)
"127.0.0.1:36001,127.0.0.2:36001"
self.set_strategy(strategy, 'recompute')
opt = fluid.optimizer.MomentumOptimizer(
learning_rate=0.001, momentum=0.9)
opt = RecomputeOptimizer(opt)
opt.user_defined_strategy = strategy
params_grads = opt.backward(avg_cost, startup_prog)
outs = [
op.output('Out')[0] for op in avg_cost.block.ops if op.type == 'mul'
]
self.assertIn('subprog', ''.join(outs))
def test_recompute_optimizer_backward_gradients(self):
""" test recompute optimizer backward + gradients """
train_prog, startup_prog = fluid.Program(), fluid.Program()
avg_cost, strategy = self.net(train_prog, startup_prog)
self.set_strategy(strategy, 'recompute')
opt = fluid.optimizer.MomentumOptimizer(
learning_rate=0.001, momentum=0.9)
opt = RecomputeOptimizer(opt)
opt.user_defined_strategy = strategy
params_grads = opt.backward(avg_cost, startup_prog)
with fluid.program_guard(train_prog, startup_prog):
opt.apply_gradients(params_grads)
outs = [
op.output('Out')[0] for op in avg_cost.block.ops if op.type == 'mul'
]
self.assertIn('subprog', ''.join(outs))
def test_recompute_optimizer_backward_optimize(self):
""" test recompute optimizer backward + optimize """
train_prog, startup_prog = fluid.Program(), fluid.Program()
avg_cost, strategy = self.net(train_prog, startup_prog)
self.set_strategy(strategy, 'recompute')
opt = fluid.optimizer.MomentumOptimizer(
learning_rate=0.001, momentum=0.9)
opt = RecomputeOptimizer(opt)
opt.user_defined_strategy = strategy
params_grads = opt.backward(avg_cost, startup_prog)
opt.apply_optimize(avg_cost, startup_prog, params_grads)
outs = [
op.output('Out')[0] for op in avg_cost.block.ops if op.type == 'mul'
]
self.assertIn('subprog', ''.join(outs))
def test_recompute_optimizer_backward(self):
""" test recompute optimizer backward """
train_prog, startup_prog = fluid.Program(), fluid.Program()
avg_cost, strategy = self.net(train_prog, startup_prog)
self.set_strategy(strategy, 'recompute')
opt = fluid.optimizer.MomentumOptimizer(
learning_rate=0.001, momentum=0.9)
opt = RecomputeOptimizer(opt)
opt.user_defined_strategy = strategy
params_grads = opt.backward(avg_cost, startup_prog)
outs = [
op.output('Out')[0] for op in avg_cost.block.ops if op.type == 'mul'
]
self.assertIn('subprog', ''.join(outs))
def test_recompute_optimizer_backward(self):
""" test recompute optimizer backward """
train_prog, startup_prog = fluid.Program(), fluid.Program()
avg_cost, strategy = self.net(train_prog, startup_prog)
self.set_strategy(strategy, 'recompute')
opt = fluid.optimizer.MomentumOptimizer(
learning_rate=0.001, momentum=0.9)
opt = RecomputeOptimizer(opt)
opt.user_defined_strategy = strategy
params_grads = opt.backward(avg_cost, startup_prog)
outs = [
op.output('Out')[0] for op in avg_cost.block.ops if op.type == 'mul'
]
self.assertIn('subprog', ''.join(outs))
def test_recompute_optimizer(self): def test_recompute_optimizer(self):
import paddle.distributed.fleet as fleet train_prog, startup_prog = fluid.Program(), fluid.Program()
import paddle.distributed.fleet.base.role_maker as role_maker avg_cost, strategy = self.net(train_prog, startup_prog)
role = role_maker.PaddleCloudRoleMaker(is_collective=True) self.set_strategy(strategy, 'recompute')
fleet.init(role) self.optimizer(avg_cost, strategy, train_prog, startup_prog)
input_x = paddle.fluid.layers.data(
name="x", shape=[32], dtype='float32') outs = [
input_y = paddle.fluid.layers.data(name="y", shape=[1], dtype='int64') op.output('Out')[0] for op in avg_cost.block.ops if op.type == 'mul'
]
fc_1 = paddle.fluid.layers.fc(input=input_x, size=64, act='tanh')
fc_2 = paddle.fluid.layers.fc(input=fc_1, size=64, act='tanh') self.assertIn('subprog', ''.join(outs))
prediction = paddle.fluid.layers.fc(input=[fc_2], size=2, act='softmax')
cost = paddle.fluid.layers.cross_entropy( def test_recompute_lars_optimizer(self):
input=prediction, label=input_y) train_prog, startup_prog = fluid.Program(), fluid.Program()
avg_cost = paddle.fluid.layers.mean(x=cost) avg_cost, strategy = self.net(train_prog, startup_prog)
self.set_strategy(strategy, 'recompute')
strategy = paddle.distributed.fleet.DistributedStrategy() self.set_strategy(strategy, 'lars')
strategy.recompute = True self.optimizer(avg_cost, strategy, train_prog, startup_prog)
strategy.recompute_configs = {"checkpoints": ["fc_1.tmp_0"]}
ops = [op.type for op in avg_cost.block.ops]
optimizer = paddle.fluid.optimizer.SGD(learning_rate=0.01) outs = [
optimizer = fleet.distributed_optimizer(optimizer, strategy=strategy) op.output('Out')[0] for op in avg_cost.block.ops if op.type == 'mul'
optimizer.minimize(avg_cost) ]
self.assertIn('subprog', ''.join(outs))
self.assertIn('lars_momentum', ops)
def test_recompute_lamb_optimizer(self):
train_prog, startup_prog = fluid.Program(), fluid.Program()
avg_cost, strategy = self.net(train_prog, startup_prog)
self.set_strategy(strategy, 'recompute')
self.set_strategy(strategy, 'lamb')
self.optimizer(avg_cost, strategy, train_prog, startup_prog, 'adam')
ops = [op.type for op in avg_cost.block.ops]
outs = [
op.output('Out')[0] for op in avg_cost.block.ops if op.type == 'mul'
]
self.assertIn('subprog', ''.join(outs))
self.assertIn('lamb', ops)
if __name__ == "__main__": if __name__ == "__main__":
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册