distributed_strategy.py 21.2 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14
#   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.

15
import paddle
16 17
from paddle.fleet.proto import distributed_strategy_pb2
from paddle.fluid.framework import Variable
18
import google.protobuf.text_format
19 20


21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
def get_msg_dict(msg):
    res_dict = {}
    fields = msg.DESCRIPTOR.fields
    for f in fields:
        res_dict[f.name] = getattr(msg, f.name)
    return res_dict


def assign_configs_value(msg, config):
    fields = msg.DESCRIPTOR.fields
    for key in config:
        for f in fields:
            if key == f.name:
                if f.label == 3:
                    getattr(msg, f.name).extend(config[f.name])
                elif f.label == 1 or f.label == 2:
                    setattr(msg, f.name, config[f.name])


def check_configs_key(msg, config, field_name):
    key_list = msg.DESCRIPTOR.fields_by_name.keys()
    for key in config:
        assert key in key_list, "key:{} not in {}".format(key, field_name)


46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84
class DistributedJobInfo(object):
    """
    DistributedJobInfo will serialize all distributed training information
    Just for inner use: 1) debug 2) replicate experiments
    """

    def __init__(self):
        self.job_info = distributed_strategy_pb2.DistributedJobInfo()

    def _set_worker_num(self, worker_num):
        self.job_info.worker_num = worker_num

    def _set_server_num(self, server_num):
        self.job_info.server_num = server_num

    def _set_worker_ips(self, worker_ips):
        self.job_info.worker_ips.extend(worker_ips)

    def _set_server_endpoints(self, server_endpoints):
        self.job_info.server_endpoints.extend(server_endpoints)

    def _set_origin_startup(self, origin_startup_prog):
        self.job_info.origin_startup = str(origin_startup_prog)

    def _set_origin_main(self, origin_main_prog):
        self.job_info.origin_main = str(origin_main_prog)

    def _distributed_main(self, distributed_main_prog):
        self.job_info.distributed_main = str(distributed_main_prog)

    def _optimizer_name(self, optimizer_name):
        self.job_info.optimizer_name = optimizer_name

    def _set_distributed_strategy(self, dist_strategy):
        self.job_info.strategy = dist_strategy


class DistributedStrategy(object):
    def __init__(self):
85 86 87 88 89 90 91 92 93 94 95 96
        """
        DistributedStrategy is the main configuration entry for distributed training of Paddle.
        All of the distributed training configurations can be configured in DistributedStrategy,
        such as automatic mixed precision (AMP), Layer-wise Adaptive Rate Scaling (LARS), 
        asynchronous update parameter server(ASGD), etc.
        
        DistributedStrategy can be serialized into protobuf file or deserialized from protobuf file

        Users who run local training usually configure BuildStrategy and ExecutionStrategy, and 
        DistributedStrategy supports configurations from BuildStrategy and ExecutionStrategy

        """
97 98
        self.strategy = distributed_strategy_pb2.DistributedStrategy()

99
    def save_to_prototxt(self, output):
100 101 102 103 104 105 106 107 108 109 110 111 112
        """
        Serialize current DistributedStrategy to string and save to output file

        Examples:
          .. code-block:: python
        
            import paddle.fleet as fleet
            strategy = fleet.DistributedStrategy()
            strategy.dgc = True
            strategy.recompute = True
            strategy.recompute_configs = {"checkpoint": ["x"]}
            strategy.save_to_prototxt("dist_strategy.prototxt")
        """
113 114 115 116
        with open(output, "w") as fout:
            fout.write(str(self.strategy))

    def load_from_prototxt(self, pb_file):
117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203
        """
        Load from prototxt file for DistributedStrategy initialization

        Examples:
          .. code-block:: python

            import paddle.fleet as fleet
            strategy = fleet.DistributedStrategy()
            strategy.load_from_prototxt("dist_strategy.protoxt")
        """
        with open(pb_file, 'r') as f:
            self.strategy = google.protobuf.text_format.Merge(
                str(f.read()), self.strategy)

    @property
    def execution_strategy(self):
        """
        Configure ExecutionStrategy for DistributedStrategy

        Examples:
          .. code-block:: python

            exe_strategy = paddle.fluid.ExecutionStrategy()
            exe_strategy.num_threads = 10
            exe_strategy.num_iteration_per_drop_scope = 10
            exe_strategy.num_iteration_per_run = 10

            strategy = paddle.fleet.DistributedStrategy()
            strategy.execution_strategy = exe_strategy
        """
        execution_strategy = paddle.fluid.ExecutionStrategy()
        fields = self.strategy.execution_strategy.DESCRIPTOR.fields
        for f in fields:
            setattr(execution_strategy, f.name,
                    getattr(self.strategy.execution_strategy, f.name))
        return execution_strategy

    @execution_strategy.setter
    def execution_strategy(self, strategy):
        fields = self.strategy.execution_strategy.DESCRIPTOR.fields
        for f in fields:
            setattr(self.strategy.execution_strategy, f.name,
                    getattr(strategy, f.name))

    @property
    def build_strategy(self):
        """
        Configure BuildStrategy for DistributedStrategy
        Note that the properties of BuildStrategy are valid in DistributedStrategy
        only if the property is non-distributed strategy.

        Examples:
          .. code-block:: python

            build_strategy = paddle.fluid.BuildStrategy()
            build_strategy.enable_sequential_execution = True
            build_strategy.fuse_elewise_add_act_ops = True
            build_strategy.fuse_bn_act_ops = True
            build_strategy.enable_auto_fusion = True
            build_strategy.fuse_relu_depthwise_conv = True
            build_strategy.fuse_broadcast_ops = True
            build_strategy.fuse_all_optimizer_ops = True
            build_strategy.enable_inplace = True
            
            strategy = paddle.fleet.DistributedStrategy()
            strategy.build_strategy = build_strategy
        """

        build_strategy = paddle.fluid.BuildStrategy()
        fields = self.strategy.build_strategy.DESCRIPTOR.fields
        for f in fields:
            setattr(build_strategy, f.name,
                    getattr(self.strategy.build_strategy, f.name))
        return build_strategy

    @build_strategy.setter
    def build_strategy(self, strategy):
        fields = self.strategy.build_strategy.DESCRIPTOR.fields
        for f in fields:
            if f.label == 1 or f.label == 2:  # optional and required field
                setattr(self.strategy.build_strategy, f.name,
                        getattr(strategy, f.name))
            elif f.label == 3:  # repeated field
                getattr(self.strategy.build_strategy,
                        f.name).extend(getattr(strategy, f.name))

    @property
D
Dong Daxiang 已提交
204
    def a_sync(self):
205 206 207 208 209 210 211 212 213 214 215 216 217 218
        """
        Indicating whether we are using asynchronous stocastic gradient descent updates
        for training. This property is valid when we are using parameter server training, 
        which is implied by setting approperate RoleMaker
        Default value: True

        Examples:
          .. code-block:: python

            import paddle.fleet as fleet
            role_maker = fleet.PaddleCloudRoleMaker()
            fleet.init(role_maker)

            strategy = fleet.DistributedStrategy()
D
Dong Daxiang 已提交
219
            strategy.a_sync = True  # by default this is True
220 221 222 223
            
            # code block for defining loss and local optimizer
            # sgd = fleet.distributed_optimizer(optimizer, strategy)
        """
D
Dong Daxiang 已提交
224
        return self.strategy.a_sync
225

D
Dong Daxiang 已提交
226 227
    @a_sync.setter
    def a_sync(self, flag):
228
        if isinstance(flag, bool):
D
Dong Daxiang 已提交
229
            self.strategy.a_sync = flag
230
        else:
D
Dong Daxiang 已提交
231
            print("WARNING: a_sync should have value of bool type")
232 233

    @property
D
Dong Daxiang 已提交
234
    def a_sync_configs(self):
235
        """
D
Dong Daxiang 已提交
236
        Set a_sync update configurations. In general, asynchronous parameter server
237 238
        training has serveral configurable settings that can be configured through
        a dict.
239

240
        **Notes**:
D
Dong Daxiang 已提交
241
            **Detailed arguments for a_sync_configs**
242 243 244 245 246 247 248
            **k_step**: number of local optimization updates before communication
            **max_merge_var_num**: maximum number of merged gradients before communication
            **send_queue_size**: a buffer size of worker communication
            **independent_recv_thread**: if we are using independent recv thread for communication
            **thread_pool_size**: number of thread pool
            **send_wait_times**: waiting time for sending gradients
            **runtime_split_send_recv**: if we are using Tensor split for send and recv during runtime
249

250 251
        Examples:
          .. code-block:: python
252

253 254 255
            import paddle.fleet as fleet
            role_maker = fleet.PaddleCloudRoleMaker()
            fleet.init(role_maker)
256

257
            strategy = fleet.DistributedStrategy()
D
Dong Daxiang 已提交
258
            strategy.a_sync = True  # by default this is True
259
            configs = {"k_step": 10000, "send_queue_size": 32}
D
Dong Daxiang 已提交
260
            strategy.a_sync_configs = configs
261

262 263 264
            # code block for defining loss and local optimizer
            # sgd = fleet.distributed_optimizer(optimizer, strategy)
        """
D
Dong Daxiang 已提交
265
        return get_msg_dict(self.strategy.a_sync_configs)
266

D
Dong Daxiang 已提交
267 268 269 270 271
    @a_sync_configs.setter
    def a_sync_configs(self, configs):
        check_configs_key(self.strategy.a_sync_configs, configs,
                          "a_sync_configs")
        assign_configs_value(self.strategy.a_sync_configs, configs)
272

273
    @property
274 275 276 277
    def amp(self):
        """
        Indicating whether we are using automatic mixed precision training
        Default Value: False
278

279 280
        Examples:
          .. code-block:: python
281

282 283 284
            import paddle.fleet as fleet
            strategy = fleet.DistributedStrategy()
            strategy.amp = True # by default this is false
285

286 287
        """
        return self.strategy.amp
288

289 290
    @amp.setter
    def amp(self, flag):
291
        if isinstance(flag, bool):
292
            self.strategy.amp = flag
293
        else:
294
            print("WARNING: amp should have value of bool type")
295 296

    @property
297 298
    def amp_configs(self):
        return get_msg_dict(self.strategy.amp_configs)
299

300 301 302 303
    @amp_configs.setter
    def amp_configs(self, configs):
        check_configs_key(self.strategy.amp_configs, configs, "amp_configs")
        assign_configs_value(self.strategy.amp_configs, configs)
304 305

    @property
306 307 308 309 310 311 312 313 314 315 316 317 318 319 320
    def recompute(self):
        """
        Indicating whether we are using forward recomputation for memory optimization
        Default value: False

        Examples:
          .. code-block:: python

            import paddle.fleet as fleet
            strategy = fleet.DistributedStrategy()
            strategy.recompute = True
            # suppose x and y are names of checkpoint tensors for recomputation
            strategy.recompute_configs = {"checkpoints": ["x", "y"]}
        """
        return self.strategy.recompute
321

322 323 324 325 326 327 328 329 330
    @property
    def sync_nccl_allreduce(self):
        return self.strategy.sync_nccl_allreduce

    @sync_nccl_allreduce.setter
    def sync_nccl_allreduce(self, flag):
        if isinstance(flag, bool):
            self.strategy.sync_nccl_allreduce = flag
        else:
331
            print("WARNING: sync_nccl_allreduce should have value of bool type")
332

333
    @property
334 335
    def use_hierarchical_allreduce(self):
        return self.strategy.use_hierarchical_allreduce
336

337 338
    @use_hierarchical_allreduce.setter
    def use_hierarchical_allreduce(self, flag):
339
        if isinstance(flag, bool):
340
            self.strategy.use_hierarchical_allreduce = flag
341 342
        else:
            print(
343
                "WARNING: use_hierarchical_allreduce should have value of bool type"
344 345 346
            )

    @property
347 348
    def hierarchical_allreduce_inter_nranks(self):
        return self.strategy.hierarchical_allreduce_inter_nranks
349

350 351 352 353
    @hierarchical_allreduce_inter_nranks.setter
    def hierarchical_allreduce_inter_nranks(self, value):
        if isinstance(value, int):
            self.strategy.hierarchical_allreduce_inter_nranks = value
354 355
        else:
            print(
356
                "WARNING: hierarchical_allreduce_inter_nranks should have value of int type"
357 358
            )

359
    @property
360 361
    def sync_batch_norm(self):
        return self.strategy.sync_batch_norm
362

363 364
    @sync_batch_norm.setter
    def sync_batch_norm(self, flag):
365
        if isinstance(flag, bool):
366
            self.strategy.sync_batch_norm = flag
367
        else:
368
            print("WARNING: sync_batch_norm should have value of bool type")
369 370 371 372 373 374 375 376 377 378 379 380 381

    @property
    def fuse_all_reduce_ops(self):
        return self.strategy.fuse_all_reduce_ops

    @fuse_all_reduce_ops.setter
    def fuse_all_reduce_ops(self, flag):
        if isinstance(flag, bool):
            self.strategy.fuse_all_reduce_ops = flag
        else:
            print("WARNING: fuse_all_reduce_ops should have value of bool type")

    @property
382 383
    def nccl_comm_num(self):
        return self.strategy.nccl_comm_num
384

385 386
    @nccl_comm_num.setter
    def nccl_comm_num(self, value):
387
        if isinstance(value, int):
388
            self.strategy.nccl_comm_num = value
389
        else:
390
            print("WARNING: nccl_comm_num should have value of int type")
391

392 393
    @recompute.setter
    def recompute(self, flag):
394
        if isinstance(flag, bool):
395
            self.strategy.recompute = flag
396
        else:
397
            print("WARNING: recompute should have value of bool type")
398 399

    @property
400 401 402 403
    def recompute_configs(self):
        """
        Set recompute configurations. In general, the recompute strategy of current
        implementation should have some manually assign checkpoints
404

405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420
        Examples:
          .. code-block:: python
        
            import paddle.fleet as fleet
            strategy = fleet.DistributedStrategy()
            strategy.recompute = True
            strategy.recompute_configs = {"checkpionts": ["x", "y"]}

        """
        return get_msg_dict(self.strategy.recompute_configs)

    @recompute_configs.setter
    def recompute_configs(self, configs):
        check_configs_key(self.strategy.recompute_configs, configs,
                          "checkpoint_configs")
        assign_configs_value(self.strategy.recompute_configs, configs)
421 422

    @property
423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438
    def pipeline(self):
        """
        Indicating whether we are using pipeline parallelism for distributed training.
        Current implementation mainly focus on single GPU machine pipeline parallelism and
        data parallelism across GPU machine. The pipeline information is indicated through
        device_guard information in user-defined program.

        Examples:
          .. code-block:: python
        
            import paddle.fleet as fleet
            strategy = fleet.DistributedStrategy()
            strategy.pipeline = True

        """
        return self.strategy.pipeline
439

440 441
    @pipeline.setter
    def pipeline(self, flag):
442
        if isinstance(flag, bool):
443
            self.strategy.pipeline = flag
444
        else:
445
            print("WARNING: pipeline should have value of bool type")
446 447

    @property
448 449 450 451 452 453 454 455 456 457
    def pipeline_configs(self):
        """
        Set pipeline parallelism configurations. In pipeline parallelism,
        different parts of neural networks are running on different GPUS.
        There are Tensor queue buffer between each pair of neighborhood GPUS 
        that are responsible for synchronizing hidden Tensor results between
        GPUs. Pipeline parallelism consists of serveral producer-consumer style
        hardware pairs, such as GPU-GPU, CPU-GPU, GPU-XPU. The best way to speedup
        pipeline parallelism is to make the size of Tensor in Tensor queue smaller, 
        so that we will have a faster producer for downstream consumers.
458

459 460 461
        **Notes**:
            **Detailed arguments for pipeline_configs**
            **micro_batch**: the number of small batches in each user defined batch
462

463 464 465 466 467 468 469
        Examples:
          .. code-block:: python
        
            import paddle.fleet as fleet
            strategy = fleet.DistributedStrategy()
            strategy.pipeline = True
            strategy.pipeline_configs = {"micro_batch": 12}
470

471
        """
472

473
        return get_msg_dict(self.strategy.pipeline_configs)
474

475 476 477 478 479
    @pipeline_configs.setter
    def pipeline_configs(self, configs):
        check_configs_key(self.strategy.pipeline_configs, configs,
                          "pipeline_configs")
        assign_configs_value(self.strategy.pipeline_configs, configs)
480 481

    @property
482 483
    def localsgd(self):
        return self.strategy.localsgd
484

485 486 487 488
    @localsgd.setter
    def localsgd(self, flag):
        if isinstance(flag, bool):
            self.strategy.localsgd = flag
489
        else:
490
            print("WARNING: localsgd should have value of bool type")
491 492

    @property
493 494
    def localsgd_configs(self):
        return get_msg_dict(self.strategy.localsgd_configs)
495

496 497 498 499 500
    @localsgd_configs.setter
    def localsgd_configs(self, configs):
        check_configs_key(self.strategy.localsgd_configs, configs,
                          "localsgd_configs")
        assign_configs_value(self.strategy.localsgd_configs, configs)
501 502

    @property
503 504
    def dgc(self):
        return self.strategy.dgc
505

506 507 508 509
    @dgc.setter
    def dgc(self, flag):
        if isinstance(flag, bool):
            self.strategy.dgc = flag
510
        else:
511
            print("WARNING: dgc should have value of bool type")
512 513

    @property
514 515
    def dgc_configs(self):
        return get_msg_dict(self.strategy.dgc_configs)
516

517 518 519 520
    @dgc_configs.setter
    def dgc_configs(self, configs):
        check_configs_key(self.strategy.dgc_configs, configs, "dgc_configs")
        assign_configs_value(self.strategy.dgc_configs, configs)
521 522

    @property
523
    def gradient_merge(self):
524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540
        """
        Gradient Merge, also called as Gradient Accumulation,
        is a strategy for large batch training. With this strategy,
        model parameter will not be updated until user-defined steps.
        For each step, the forward network and the backward network
        will run to calculate the gradient of model parameters.
        For every k step, the optimization network will run,
        applying a specific optimization method (such as SGD, Adam)
        to model parameters.

        Examples:
        .. code-block:: python
            import paddle.fleet as fleet
            strategy = fleet.DistributedStrategy()
            strategy.gradient_merge = True
            strategy.gradient_merge_configs = {"k_steps": 4, "avg": True}
        """
541
        return self.strategy.gradient_merge
542

543 544
    @gradient_merge.setter
    def gradient_merge(self, flag):
545
        if isinstance(flag, bool):
546
            self.strategy.gradient_merge = flag
547
        else:
548 549 550 551
            print("WARNING: gradient_merge should have value of bool type")

    @property
    def gradient_merge_configs(self):
552 553 554 555 556 557 558 559 560 561 562 563
        """
        the key-value configs of distribute_strategy
        Keys: 
            k_steps (int): the update period of the parameters
            avg (bool): whether to average the gradients of each mini-batch,
                the default value is `True`
        Example:
            import paddle.fleet as fleet
            strategy = fleet.DistributedStrategy()
            strategy.gradient_merge = True
            strategy.gradient_merge_configs = {"k_steps": 4, "avg": True}
        """
564 565 566 567 568 569 570
        return get_msg_dict(self.strategy.gradient_merge_configs)

    @gradient_merge_configs.setter
    def gradient_merge_configs(self, configs):
        check_configs_key(self.strategy.gradient_merge_configs, configs,
                          "gradient_configs")
        assign_configs_value(self.strategy.gradient_merge_configs, configs)
571 572

    @property
573 574
    def lars(self):
        return self.strategy.lars
575

576 577
    @lars.setter
    def lars(self, flag):
578
        if isinstance(flag, bool):
579
            self.strategy.lars = flag
580
        else:
581
            print("WARNING: lars should have value of bool type")
582 583

    @property
584 585
    def lamb(self):
        return self.strategy.lamb
586

587 588
    @lamb.setter
    def lamb(self, flag):
589
        if isinstance(flag, bool):
590
            self.strategy.lamb = flag
591
        else:
592
            print("WARNING: lamb should have value of bool type")
593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616

    @property
    def elastic(self):
        return self.strategy.elastic

    @elastic.setter
    def elastic(self, flag):
        if isinstance(flag, bool):
            self.strategy.elastic = flag
        else:
            print("WARNING: elastic should have value of bool type")

    @property
    def auto(self):
        return self.strategy.auto

    @auto.setter
    def auto(self, flag):
        if isinstance(flag, bool):
            self.strategy.auto = flag
        else:
            print("WARNING: auto should have value of bool type")

    def __repr__(self):
D
Dong Daxiang 已提交
617 618 619
        fields = self.strategy.DESCRIPTOR.fields
        for f in fields:
            print("{}: {}".format(f.name, f.default_value))
620
        return str(self.strategy)