未验证 提交 d5ac87ec 编写于 作者: M mapingshuo 提交者: GitHub

Lookahead optimizer (#19386)

* Add lookahead optimizer

* add unittest for lookahead optimizer

test=develop

* add doc string for LookaheadOptimizer

test=develop
test=document_preview

* add API spec for lookahead

test=develop
test=document_preview

* modify api spec

test=develop
test=document_preview

* modified doc string

* modify the test file

test=develop
test=document_preview

* modify doc string

test=develop
test=document_preview
上级 12d29f4d
develop 2.0.1-rocm-post Ligoml-patch-1 OliverLPH-patch-1 OliverLPH-patch-2 PaddlePM-patch-1 PaddlePM-patch-2 ZHUI-patch-1 add_default_att add_model_benchmark_ci add_some_yaml_config addfile all_new_design_exec ascendrc ascendrelease cherry_undefined_var compile_windows cp_2.4_fix_numpy delete_2.0.1-rocm-post delete_add_default_att delete_all_new_design_exec delete_ascendrc delete_compile_windows delete_delete_addfile delete_disable_iterable_dataset_unittest delete_fix_dataloader_memory_leak delete_fix_imperative_dygraph_error delete_fix_retry_ci delete_fix_undefined_var delete_improve_sccache delete_paddle_tiny_install delete_paralleltest delete_prv-disable-more-cache delete_revert-31068-fix_conv3d_windows delete_revert-31562-mean delete_revert-33630-bug-fix delete_revert-34159-add_npu_bce_logical_dev delete_revert-34910-spinlocks_for_allocator delete_revert-35069-revert-34910-spinlocks_for_allocator delete_revert-36057-dev/read_flags_in_ut dingjiaweiww-patch-1 disable_iterable_dataset_unittest dy2static enable_eager_model_test final_state_gen_python_c final_state_intermediate fix-numpy-issue fix_concat_slice fix_dataloader_memory_leak fix_dlpack_for fix_imperative_dygraph_error fix_npu_ci fix_op_flops fix_retry_ci fix_rnn_docs fix_tensor_type fix_undefined_var fix_var_stop_gradient_error fixiscan fixiscan1 fixiscan2 fixiscan3 github/fork/123malin/netifaces github/fork/123malin/tdm_abacus github/fork/AshburnLee/dev_unique github/fork/ForFishes/fix_memory_matmul github/fork/ForFishes/rm_fluid github/fork/LielinJiang/move-2.0-api github/fork/LielinJiang/visual-dl-cb github/fork/LiuChiachi/add-transformer-generate-square-subsequent-mask-api github/fork/LiuChiachi/fix-example-code-for-hapi-Model github/fork/LiuChiachi/remove-input-requirment-in-dygraph-Model github/fork/MrChengmo/fix_ps_profiler github/fork/MrChengmo/update_ps_heter github/fork/PWhiddy/patch-1 github/fork/Shixiaowei02/dev/save_load_upgrade github/fork/TCChenlong/fix_hapi github/fork/TCChenlong/fix_inden github/fork/Thunderbrook/xpu_slice github/fork/XieYunshen/disable_ut_test_parallel_executor_fetch_isolated_var github/fork/XieYunshen/disable_ut_test_parallel_executor_fetch_isolated_var_2 github/fork/XieYunshen/disable_ut_test_parallel_executor_fetch_isolated_var_3 github/fork/XieYunshen/timeout_20S_ut github/fork/ZeyuChen/remove-nltk github/fork/arlesniak/arlesniak/selective__mkldnn_flags github/fork/baiyfbupt/code_doc_mig github/fork/chalsliu/set_timeout github/fork/chen-zhiyu/develop github/fork/chenwhql/ci/try_to_find_test_buffer_shared_memory_reuse_pass_error github/fork/chenwhql/dygraph/remove_scale_loss_and_apply_collective_grads github/fork/chenwhql/saveload/add_get_inference_program github/fork/chenwhql/saveload/remove_save_load_config github/fork/cryoco/pass-compatibility-trt github/fork/danleifeng/isempty_api2.0 github/fork/frankwhzhang/api_transfer github/fork/hbwx24/error_msg/cuda_kernel_error_msg github/fork/heavengate/cherry_yolo_box github/fork/heavengate/update_yolo_box github/fork/iclementine/rnn_fix github/fork/iducn/testestse github/fork/jczaja/prv-25537-fix github/fork/jeff41404/release/1.8 github/fork/jiweibo/api_2.0 github/fork/jiweibo/fix_lite_resnet50_test github/fork/juncaipeng/fix_doc_1 github/fork/lfchener/sample_code github/fork/littletomatodonkey/fix_reg_doc github/fork/liym27/dy2stat_update_assign_to_rc20 github/fork/luotao1/profiler_ut github/fork/mapingshuo/add_wait github/fork/mapingshuo/doc_2.0 github/fork/mapingshuo/zero-0.5 github/fork/miraiwk/dev github/fork/pangyoki/add-Categorical-class-branch github/fork/pangyoki/add-multinomial-op-branch github/fork/pangyoki/fix-test_distritbution-CI github/fork/qjing666/doublegrad github/fork/qjing666/fix_hdfs_download github/fork/sandyhouse/add_gather_etc github/fork/sandyhouse/add_send_recv_alltoall_etc github/fork/sandyhouse/pipeline_exe_run github/fork/seiriosPlus/feature/large_scale_kv_save_delta github/fork/seiriosPlus/fix/paddle_errors_fix github/fork/seiriosPlus/fix/paddle_op_errors github/fork/shangzhizhou/fix_test_activation_op_random_bug github/fork/smallv0221/yxp0924 github/fork/smallv0221/yxp0925 github/fork/swtkiwi/del-matplotlib github/fork/tianshuo78520a/kunlun_test github/fork/tianshuo78520a/update_dockerfile github/fork/wanghaoshuang/bert_fuse github/fork/wanghaoshuang/label_smooth github/fork/wanghuancoder/develop_CUDASynchronize github/fork/wanghuancoder/develop_Layer_doc github/fork/wanghuancoder/develop_ParameterList_doc github/fork/wanghuancoder/develop_Sequential_doc github/fork/wanghuancoder/develop_bilinear_tensor_product github/fork/wanghuancoder/develop_coverage_build_sh github/fork/wanghuancoder/develop_in_dynamic_mode_doc github/fork/wanghuancoder/develop_unique_name_doc github/fork/wangxicoding/fleet_meta_combine github/fork/wawltor/error_message_fix_5 github/fork/willthefrog/remove_l2_norm github/fork/windstamp/momentum_op github/fork/windstamp/mv_op_5 github/fork/windstamp/normal_api github/fork/wojtuss/wojtuss/fusion_gru_quantization github/fork/wojtuss/wojtuss/quantization-with-shift github/fork/wzzju/fix_err_info github/fork/wzzju/pure_fp16 github/fork/xiemoyuan/op_error_message github/fork/xiemoyuan/optimize_error_message github/fork/yaoxuefeng6/fix_doc github/fork/yaoxuefeng6/mod_dataset_v2 github/fork/yongqiangma/lod github/fork/ysh329/fix-clip-by-norm-error github/fork/ysh329/fix-error-clip-by-value github/fork/yukavio/error_info github/fork/zhangting2020/conv_filter_grad github/fork/zhangting2020/is_compile_with_cuda github/fork/zhangting2020/place_doc github/fork/zhangting2020/program github/fork/zhhsplendid/fix_any github/fork/zhhsplendid/refine_api2 github/fork/zhhsplendid/refine_api2_test github/fork/zhhsplendid/refine_api_test_ptb_lm github/fork/zhhsplendid/refine_api_test_resnet github/fork/zhhsplendid/refine_api_test_simnet github/fork/zhiqiu/dev/refine_initializer github/fork/zhiqiu/dev/remove_inplace_argument github/fork/zlsh80826/nvinfer_plugin_var_len_cuda11 improve_sccache incubate/frl_train_eval incubate/infrt inplace_addto layer_norm make_flag_adding_easier matmul_double_grad move_embedding_to_phi move_histogram_to_pten move_sgd_to_phi move_slice_to_pten move_temporal_shift_to_phi move_yolo_box_to_phi npu_fix_alloc numel paddle_tiny_install paralleltest preln_ernie prv-disable-more-cache prv-md-even-more prv-onednn-2.5 prv-reshape-mkldnn-ut2 pten_tensor_refactor release/1.6 release/1.7 release/1.8 release/2.0 release/2.0-alpha release/2.0-beta release/2.0-rc release/2.0-rc1 release/2.1 release/2.2 release/2.3 release/2.3-fc-ernie-fix release/2.4 revert-24981-add_device_attr_for_regulization revert-26856-strategy_example2 revert-27520-disable_pr revert-31068-fix_conv3d_windows revert-31562-mean revert-32290-develop-hardlabel revert-33037-forci revert-33475-fix_cifar_label_dimension revert-33630-bug-fix revert-34159-add_npu_bce_logical_dev revert-34406-add_copy_from_tensor revert-34910-spinlocks_for_allocator revert-35069-revert-34910-spinlocks_for_allocator revert-36057-dev/read_flags_in_ut revert-36201-refine_fast_threaded_ssa_graph_executor revert-36985-add_license revert-37318-refactor_dygraph_to_eager revert-37926-eager_coreops_500 revert-37956-revert-37727-pylayer_support_tuple revert-38100-mingdong revert-38301-allocation_rearrange_pr revert-38703-numpy_bf16_package_reupload revert-38732-remove_useless_header_in_elementwise_mul_grad revert-38959-Reduce_Grad revert-39143-adjust_empty revert-39227-move_trace_op_to_pten revert-39268-dev/remove_concat_fluid_kernel revert-40170-support_partial_grad revert-41056-revert-40727-move_some_activaion_to_phi revert-41065-revert-40993-mv_ele_floordiv_pow revert-41068-revert-40790-phi_new revert-41944-smaller_inference_api_test revert-42149-do-not-reset-default-stream-for-stream-safe-cuda-allocator revert-43155-fix_ut_tempfile revert-43882-revert-41944-smaller_inference_api_test revert-45808-phi/simplify_size_op revert-46827-deform_comment revert-47325-remove_cudnn_hardcode revert-47645-add_npu_storage_dims revert-48815-set_free_when_no_cache_hit_default_value_true revert-49654-prim_api_gen revert-49763-fix_static_composite_gen rocm_dev_0217 support-0D-sort support_weight_transpose test_benchmark_ci test_feature_precision_test_c test_for_Filtetfiles test_model_benchmark test_model_benchmark_ci zhiqiu-patch-1 v2.4.1 v2.4.0 v2.4.0-rc0 v2.3.2 v2.3.1 v2.3.0 v2.3.0-rc0 v2.2.2 v2.2.1 v2.2.0 v2.2.0-rc0 v2.2.0-bak0 v2.1.3 v2.1.2 v2.1.1 v2.1.0 v2.1.0-rc0 v2.0.2 v2.0.1 v2.0.0 v2.0.0-rc1 v2.0.0-rc0 v2.0.0-beta0 v2.0.0-alpha0 v1.8.5 v1.8.4 v1.8.3 v1.8.2 v1.8.1 v1.8.0 v1.7.2 v1.7.1 v1.7.0 v1.6.3 v1.6.2 v1.6.1 v1.6.0 v1.6.0-rc0
无相关合并请求
......@@ -982,6 +982,9 @@ paddle.fluid.optimizer.ExponentialMovingAverage.update (ArgSpec(args=['self'], v
paddle.fluid.optimizer.PipelineOptimizer ('paddle.fluid.optimizer.PipelineOptimizer', ('document', '6f85382abedb922387b08d98e8d0b69c'))
paddle.fluid.optimizer.PipelineOptimizer.__init__ (ArgSpec(args=['self', 'optimizer', 'cut_list', 'place_list', 'concurrency_list', 'queue_size', 'sync_steps', 'start_cpu_core_id'], varargs=None, keywords=None, defaults=(None, None, None, 30, 1, 0)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
paddle.fluid.optimizer.PipelineOptimizer.minimize (ArgSpec(args=['self', 'loss', 'startup_program', 'parameter_list', 'no_grad_set'], varargs=None, keywords=None, defaults=(None, None, None)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
paddle.fluid.optimizer.LookaheadOptimizer ('paddle.fluid.optimizer.LookaheadOptimizer', ('document', 'c291cadfa7452c7bf58b9e2f900a3511'))
paddle.fluid.optimizer.LookaheadOptimizer.__init__ (ArgSpec(args=['self', 'inner_optimizer', 'alpha', 'k'], varargs=None, keywords=None, defaults=(0.5, 5)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
paddle.fluid.optimizer.LookaheadOptimizer.minimize (ArgSpec(args=['self', 'loss', 'startup_program'], varargs=None, keywords=None, defaults=(None,)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))
paddle.fluid.backward.append_backward (ArgSpec(args=['loss', 'parameter_list', 'no_grad_set', 'callbacks'], varargs=None, keywords=None, defaults=(None, None, None)), ('document', '1b7b2bfb986e93048e75ba69f2f490ab'))
paddle.fluid.backward.gradients (ArgSpec(args=['targets', 'inputs', 'target_gradients', 'no_grad_set'], varargs=None, keywords=None, defaults=(None, None)), ('document', 'e2097e1e0ed84ae44951437bfe269a1b'))
paddle.fluid.regularizer.L1DecayRegularizer ('paddle.fluid.regularizer.L1DecayRegularizer', ('document', '34603757e70974d2fcc730643b382925'))
......
......@@ -43,7 +43,7 @@ __all__ = [
'AdamaxOptimizer', 'DecayedAdagradOptimizer', 'RMSPropOptimizer',
'FtrlOptimizer', 'Adadelta', 'ModelAverage', 'LarsMomentum',
'LarsMomentumOptimizer', 'DGCMomentumOptimizer', 'LambOptimizer',
'ExponentialMovingAverage', 'PipelineOptimizer'
'ExponentialMovingAverage', 'PipelineOptimizer', 'LookaheadOptimizer'
]
......@@ -2953,3 +2953,156 @@ class PipelineOptimizer(object):
"sync_steps": self._sync_steps,
"param_need_sync": param_need_sync
}
class LookaheadOptimizer(object):
"""
This implements the Lookahead optimizer of the
paper : https://arxiv.org/abs/1907.08610.
Lookahead keeps two sets of params: the fast_params and
the slow_params. inner_optimizer update fast_params every
training step. Lookahead updates the slow_params and fast_params
every k training steps as follows:
.. math::
slow\_param_t &= slow\_param_{t-1} + \\alpha * (fast\_param_{t-1} - slow\_param_{t-1})
fast\_param_t &= slow\_param_t
Args:
inner_optimizer (Optimizer): The optimizer that update fast params step by step.
alpha (float): The learning rate of Lookahead.
k (int): The slow params is updated every k steps.
Examples:
.. code-block:: python
import paddle
import paddle.fluid as fluid
import numpy as np
x = fluid.layers.data(name='x', shape=[2], dtype='float32')
label = fluid.layers.data(name="label", shape=[1], dtype="int64")
y = fluid.layers.fc(input=[x], size=2, act="softmax")
loss = fluid.layers.cross_entropy(input=y, label=label)
loss = fluid.layers.mean(x=loss)
sgd = fluid.optimizer.SGD(learning_rate=0.01)
optimizer = fluid.optimizer.LookaheadOptimizer(sgd,
alpha=0.5,
k=5)
optimizer.minimize(loss)
main_program = fluid.default_main_program()
place = fluid.CPUPlace()
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())
feeder = fluid.DataFeeder(feed_list=[x, label], place=place)
step = 0
while(step < 10):
step += 1
exe.run(fluid.default_main_program(),
feed=feeder.feed(batch_data))
"""
def __init__(self, inner_optimizer, alpha=0.5, k=5):
assert (inner_optimizer is not None), "inner optimizer can not be None"
assert (
0.0 <= alpha <= 1.0
), "alpha should be larger or equal to 0.0, and less or equal than 1.0"
assert (isinstance(k, int) and k > 0), "k should be a positive integer"
self.inner_optimizer = inner_optimizer
self.alpha = alpha
self.k = k
self.type = "lookahead"
def minimize(self, loss, startup_program=None):
# Apply inner optimizer to the main_program
mini_out = self.inner_optimizer.minimize(
loss, startup_program=startup_program)
# Get startup_program and main_program
if startup_program is None:
startup_program = default_startup_program()
main_block = loss.block
# add some vars to the main_program
params = [param.name for param in main_block.all_parameters()]
param_to_slow = {}
for param in params:
fast_var = main_block.var(param)
assert (fast_var is not None)
slow_var = main_block.create_var(
name=param + "@SLOW",
shape=fast_var.shape,
dtype=fast_var.dtype,
persistable=True)
param_to_slow[param] = slow_var
# add some vars to the startup_program
startup_block = startup_program.global_block()
for param in params:
fast_var = startup_block.var(param)
assert (fast_var is not None)
slow_var = startup_block.create_var(
name=param + "@SLOW",
shape=fast_var.shape,
dtype=fast_var.dtype,
persistable=True)
startup_block.append_op(
type="assign",
inputs={"X": fast_var},
outputs={"Out": slow_var})
# Add Var k to main prog and startup prog
k = layers.create_global_var(
name="lookahead_k",
shape=[1],
value=int(self.k),
dtype='int32',
persistable=True)
# Add Var alpha to main prog and startup prog
alpha = layers.create_global_var(
name="lookahead_alpha",
shape=[1],
value=float(self.alpha),
dtype='float32',
persistable=True)
# Add Var step
step = layers.create_global_var(
name="lookahead_step",
shape=[1],
value=int(0),
dtype='int32',
persistable=True)
layers.increment(x=step, value=1.0, in_place=True)
# lookahead
zero_var = layers.fill_constant(shape=[1], dtype='float32', value=0.0)
one_var = layers.fill_constant(shape=[1], dtype='float32', value=1.0)
mod = layers.elementwise_mod(step, k)
with layers.control_flow.Switch() as switch:
with switch.case(mod == zero_var):
for param_name in params:
fast_var = main_block.var(param_name)
slow_var = param_to_slow[param_name]
tmp_var = layers.elementwise_add(
layers.elementwise_mul(fast_var, alpha),
layers.elementwise_mul(
slow_var, layers.elementwise_sub(one_var, alpha)))
layers.assign(input=tmp_var, output=slow_var)
layers.assign(input=tmp_var, output=fast_var)
with switch.default():
pass
return mini_out
......@@ -544,5 +544,44 @@ class TestFtrlOptimizer(unittest.TestCase):
self.assertAlmostEqual(init_ops[0].attr('value'), learning_rate)
class TestLookaheadOptimizer(unittest.TestCase):
def test_lookahead_optimizer(self):
init_program = framework.Program()
program = framework.Program()
block = program.global_block()
init_block = init_program.global_block()
mul_x = block.create_parameter(
dtype="float32",
shape=[5, 10],
lod_level=0,
name="mul.x",
optimize_attr={'learning_rate': 1.1})
init_mul_x = init_block.create_parameter(
dtype="float32", shape=[5, 10], lod_level=0, name="mul.x")
mul_y = block.create_var(
dtype="float32", shape=[10, 8], lod_level=0, name="mul.y")
mul_out = block.create_var(
dtype="float32", shape=[5, 8], lod_level=0, name="mul.out")
mean_out = block.create_var(
dtype="float32", shape=[1], lod_level=0, name="mean.out")
block.append_op(
type="mul",
inputs={"X": mul_x,
"Y": mul_y},
outputs={"Out": mul_out},
attrs={"x_num_col_dims": 1})
block.append_op(
type="mean", inputs={"X": mul_out}, outputs={"Out": mean_out})
sgd = optimizer.SGD(learning_rate=0.01)
lookahead = optimizer.LookaheadOptimizer(sgd, alpha=0.5, k=5)
with framework.program_guard(program, init_program):
opts, _ = lookahead.minimize(mean_out)
self.assertEqual(len(opts), 3)
self.assertEqual([op.type for op in opts],
["fill_constant", "elementwise_mul", "sgd"])
if __name__ == '__main__':
unittest.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
反馈
建议
客服 返回
顶部