io.py 94.5 KB
Newer Older
1
#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
D
dzhwinter 已提交
2
#
D
dzhwinter 已提交
3 4 5
# 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
D
dzhwinter 已提交
6
#
D
dzhwinter 已提交
7
#     http://www.apache.org/licenses/LICENSE-2.0
D
dzhwinter 已提交
8
#
D
dzhwinter 已提交
9 10 11 12 13 14
# 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 16
from __future__ import print_function

17
import os
T
bug fix  
tangwei12 已提交
18
import errno
D
dzhwinter 已提交
19
import warnings
20
import six
21
import logging
Y
Yang Zhang 已提交
22
import pickle
H
hong 已提交
23
import contextlib
24
from functools import reduce
25
import sys
26
from io import BytesIO
27

H
hong 已提交
28
import numpy as np
29
import math
30
import paddle
31
from paddle.fluid import layers
H
hong 已提交
32
from paddle.fluid.executor import Executor, global_scope
33
from paddle.fluid.evaluator import Evaluator
T
tangwei12 已提交
34
from paddle.fluid.framework import Program, Parameter, default_main_program, default_startup_program, Variable, \
35
    program_guard, dygraph_not_support, static_only
36 37
from paddle.reader import cache, map_readers, buffered, compose, chain, shuffle, \
    ComposeNotAligned, firstn, xmap_readers, multiprocess_reader
38
from .wrapped_decorator import signature_safe_contextmanager
T
tangwei12 已提交
39
from paddle.fluid.compiler import CompiledProgram
40
from paddle.fluid.log_helper import get_logger
S
sneaxiy 已提交
41
from . import reader
42
from . import unique_name
S
sneaxiy 已提交
43
from .reader import *
44 45
from . import dataloader
from .dataloader import *
K
fix bug  
Kexin Zhao 已提交
46
from . import core
47
from .. import compat as cpt
48 49
from paddle.utils import deprecated
from paddle.fluid.framework import static_only
50

51 52
batch = paddle.batch

53
__all__ = [
54 55 56 57 58 59 60 61 62 63 64 65 66
    'save_vars',
    'save_params',
    'save_persistables',
    'load_vars',
    'load_params',
    'load_persistables',
    'save_inference_model',
    'load_inference_model',
    'batch',
    'save',
    'load',
    'load_program_state',
    'set_program_state',
H
hong 已提交
67 68
    'get_program_parameter',
    'get_program_persistable_vars',
69
] + reader.__all__
70

71 72 73
_logger = get_logger(__name__,
                     logging.INFO,
                     fmt='%(asctime)s-%(levelname)s: %(message)s')
74

75

76
class _open_buffer(object):
77

78 79 80 81 82 83 84 85
    def __init__(self, buffer):
        self.buffer = buffer

    def __enter__(self):
        return self.buffer


class _buffer_reader(_open_buffer):
86

87 88 89 90 91 92 93 94 95 96 97
    def __init__(self, buffer):
        super(_buffer_reader, self).__init__(buffer)
        self.initial_tell = self.buffer.tell()

    def __exit__(self, *args):
        # `args[0]` is type of exception. When the `read` is abnormal, the file pointer returns to the initial position.
        if args[0] is not None:
            self.buffer.seek(self.initial_tell)


class _buffer_writer(_open_buffer):
98

99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116
    def __exit__(self, *args):
        self.buffer.flush()


def _is_file_path(path):
    return isinstance(path, str)


def _open_file_buffer(path_or_buffer, mode):

    if _is_file_path(path_or_buffer):
        return open(path_or_buffer, mode)
    else:
        if 'w' in mode:
            return _buffer_writer(path_or_buffer)
        elif 'r' in mode:
            return _buffer_reader(path_or_buffer)
        else:
117 118
            raise ValueError(
                "Expected 'r' or 'w' in mode but got {}".format(mode))
119 120 121 122 123 124


def _is_memory_buffer(buffer):
    return isinstance(buffer, BytesIO)


125
def is_parameter(var):
F
fengjiayi 已提交
126 127
    """
    Check whether the given variable is an instance of Parameter.
128 129

    Args:
F
fengjiayi 已提交
130
        var(Variable): The variable to be checked.
131 132

    Returns:
F
fengjiayi 已提交
133 134 135 136 137 138
        bool: True if the given `var` is an instance of Parameter,
        False if not.

    Examples:
        .. code-block:: python

139
            import paddle
140
            import paddle.fluid as fluid
141 142

            paddle.enable_static()
F
fengjiayi 已提交
143 144
            param = fluid.default_main_program().global_block().var('fc.w')
            res = fluid.io.is_parameter(param)
145
    """
146 147 148 149
    return isinstance(var, Parameter)


def is_persistable(var):
F
fengjiayi 已提交
150 151 152 153 154 155 156 157 158 159 160 161 162
    """
    Check whether the given variable is persistable.

    Args:
        var(Variable): The variable to be checked.

    Returns:
        bool: True if the given `var` is persistable
        False if not.

    Examples:
        .. code-block:: python

163
            import paddle
164
            import paddle.fluid as fluid
165 166

            paddle.enable_static()
167
            param = fluid.default_main_program().global_block().var('fc.b')
F
fengjiayi 已提交
168 169
            res = fluid.io.is_persistable(param)
    """
170
    if var.desc.type() == core.VarDesc.VarType.FEED_MINIBATCH or \
171 172
                    var.desc.type() == core.VarDesc.VarType.FETCH_LIST or \
                    var.desc.type() == core.VarDesc.VarType.READER:
173
        return False
174 175 176
    return var.persistable


H
hong 已提交
177
def is_belong_to_optimizer(var):
178
    if not (isinstance(var, Parameter) or var.desc.need_check_feed()):
179 180 181
        return is_persistable(var)

    return False
H
hong 已提交
182 183


184
@dygraph_not_support
H
hong 已提交
185 186 187 188 189 190 191 192 193 194 195 196 197
def get_program_parameter(program):
    """
    Get all the parameters from Program.

    Args:
        var(Program): The Program to get parameters

    Returns:
        list: The list contains all parameters in the program

    Examples:
        .. code-block:: python

198
            import paddle
H
hong 已提交
199
            import paddle.fluid as fluid
200 201

            paddle.enable_static()
H
hong 已提交
202 203 204 205 206 207 208 209
            data = fluid.data(name="img", shape=[64, 784])
            w = fluid.layers.create_parameter(shape=[784, 200], dtype='float32', name='fc_w')
            b = fluid.layers.create_parameter(shape=[200], dtype='float32', name='fc_b')
            list_para  = fluid.io.get_program_parameter(  fluid.default_main_program() )
    """
    return list(filter(is_parameter, program.list_vars()))


210
@dygraph_not_support
H
hong 已提交
211 212 213 214 215 216 217 218 219 220 221 222 223
def get_program_persistable_vars(program):
    """
    Get all the persistable vars from Program.

    Args:
        var(Program): The Program to get persistable vars

    Returns:
        list: The list contains all persistable vars in the program

    Examples:
        .. code-block:: python

224
            import paddle
H
hong 已提交
225
            import paddle.fluid as fluid
226 227

            paddle.enable_static()
H
hong 已提交
228 229 230 231 232 233 234 235
            data = fluid.data(name="img", shape=[64, 784])
            w = fluid.layers.create_parameter(shape=[784, 200], dtype='float32', name='fc_w')
            b = fluid.layers.create_parameter(shape=[200], dtype='float32', name='fc_b')
            list_para  = fluid.io.get_program_persistable_vars(  fluid.default_main_program() )
    """
    return list(filter(is_persistable, program.list_vars()))


236 237
def _clone_var_in_block_(block, var):
    assert isinstance(var, Variable)
238
    if var.desc.type() == core.VarDesc.VarType.LOD_TENSOR:
239 240 241 242 243 244
        return block.create_var(name=var.name,
                                shape=var.shape,
                                dtype=var.dtype,
                                type=var.type,
                                lod_level=var.lod_level,
                                persistable=True)
245
    else:
246 247 248 249 250
        return block.create_var(name=var.name,
                                shape=var.shape,
                                dtype=var.dtype,
                                type=var.type,
                                persistable=True)
251 252


253
@signature_safe_contextmanager
H
hong 已提交
254 255 256 257 258 259 260
def _load_program_scope(main=None, startup=None, scope=None):
    prog = main if main else paddle.fluid.Program()
    startup_prog = startup if startup else paddle.fluid.Program()
    scope = scope if scope else paddle.fluid.core.Scope()
    with paddle.fluid.scope_guard(scope):
        with paddle.fluid.program_guard(prog, startup_prog):
            with paddle.fluid.unique_name.guard():
261 262
                with paddle.fluid.framework._dygraph_guard(None):
                    yield
H
hong 已提交
263 264


265
def _get_valid_program(main_program=None):
C
chengduo 已提交
266 267 268 269 270
    if main_program is None:
        main_program = default_main_program()
    elif isinstance(main_program, CompiledProgram):
        main_program = main_program._program
        if main_program is None:
271 272 273
            raise TypeError(
                "The type of input main_program is invalid, expected tyep is Program, but received None"
            )
C
chengduo 已提交
274 275 276
        warnings.warn(
            "The input is a CompiledProgram, this is not recommended.")
    if not isinstance(main_program, Program):
277 278 279
        raise TypeError(
            "The type of input main_program is invalid, expected type is fluid.Program, but received %s"
            % type(main_program))
C
chengduo 已提交
280 281 282
    return main_program


283
@dygraph_not_support
284 285 286 287 288
def save_vars(executor,
              dirname,
              main_program=None,
              vars=None,
              predicate=None,
289
              filename=None):
290
    """
291
    Save specific variables in the `Program` to files.
F
fengjiayi 已提交
292

293
    There are two ways to specify the variables to be saved: set variables in
294 295
    a list and assign it to the `vars`, or use the `predicate` function to select
    variables that make `predicate(variable) == True`. The first way has a higher priority.
296

297
    The `dirname` is used to specify the folder where to save variables.
T
tianshuo78520a 已提交
298
    If you prefer to save variables in separate files in the `dirname` folder,
299
    do not set `filename`. If you prefer to save all variables in a single file,
F
fengjiayi 已提交
300
    use `filename` to specify it.
301

F
fengjiayi 已提交
302 303
    Args:
        executor(Executor): The executor to run for saving variables.
304 305
        dirname(str, optional): The folder where to save variables.
                            When you need to save the parameter to the memory, set it to None.
306
        main_program(Program, optional): The program whose variables will be saved.
307
                                    If it is None, the default main program will
F
fengjiayi 已提交
308 309
                                    be used automatically.
                                    Default: None
310 311 312
        vars(list[Variable], optional): The list contains all variables to be saved.
                                        Default: None
        predicate(function, optional): The function selects the variables that make
313
                                       `predicate(variable) == True`.
314 315
                                       Default: None
        filename(str, optional): If you prefer to save all variables in a single file,
316
                                 use `filename` to specify it. Otherwise, let `filename` be None.
317
                                 Default: None
F
fengjiayi 已提交
318 319

    Returns:
320 321
        str: When saving parameters to a file, returns None.
             When saving parameters to memory, returns a binary string containing parameters.
F
fengjiayi 已提交
322 323 324 325 326 327 328

    Raises:
        TypeError: If `main_program` is not an instance of Program nor None.

    Examples:
        .. code-block:: python

329
            import paddle
330
            import paddle.fluid as fluid
331

332
            paddle.enable_static()
333 334 335 336 337 338 339 340 341 342 343
            main_prog = fluid.Program()
            startup_prog = fluid.Program()
            with fluid.program_guard(main_prog, startup_prog):
                data = fluid.layers.data(name="img", shape=[64, 784], append_batch_size=False)
                w = fluid.layers.create_parameter(shape=[784, 200], dtype='float32', name='fc_w')
                b = fluid.layers.create_parameter(shape=[200], dtype='float32', name='fc_b')
                hidden_w = fluid.layers.matmul(x=data, y=w)
                hidden_b = fluid.layers.elementwise_add(hidden_w, b)
            place = fluid.CPUPlace()
            exe = fluid.Executor(place)
            exe.run(startup_prog)
F
fengjiayi 已提交
344

345
            # The first usage: use `vars` to set the saved variables.
346 347
            var_list = [w, b]
            path = "./my_paddle_vars"
348
            fluid.io.save_vars(executor=exe, dirname=path, vars=var_list,
349 350 351 352 353 354 355 356 357 358
                            filename="vars_file")
            # w and b will be save in a file named "var_file".

            # The second usage: use `predicate` to select the saved variable.
            def name_has_fc(var):
                res = "fc" in var.name
                return res
            param_path = "./my_paddle_model"
            fluid.io.save_vars(executor=exe, dirname=param_path, main_program=main_prog, vars=None, predicate = name_has_fc)
            # all variables whose names contain "fc " are saved.
359
    """
360 361 362 363
    save_to_memory = False
    if dirname is None and filename is None:
        save_to_memory = True

C
chengduo 已提交
364
    main_program = _get_valid_program(main_program)
T
tangwei12 已提交
365

366
    if vars is None:
367 368 369 370 371
        return save_vars(executor,
                         main_program=main_program,
                         dirname=dirname,
                         vars=list(filter(predicate, main_program.list_vars())),
                         filename=filename)
372
    else:
石晓伟 已提交
373
        params_var_name = "saved_params"
374 375 376 377 378 379 380
        # give warning when there is no var in model
        if len(list(vars)) == 0:
            warnings.warn(
                "no variable in your model, please ensure there are any variables in your model to save"
            )
            return None

381 382
        save_program = Program()
        save_block = save_program.global_block()
383 384

        save_var_map = {}
385
        for each_var in vars:
386 387 388
            # NOTE: don't save the variable which type is RAW
            if each_var.type == core.VarDesc.VarType.RAW:
                continue
389
            new_var = _clone_var_in_block_(save_block, each_var)
390
            if filename is None and save_to_memory is False:
391 392
                save_file_path = os.path.join(os.path.normpath(dirname),
                                              new_var.name)
393 394 395 396
                save_block.append_op(
                    type='save',
                    inputs={'X': [new_var]},
                    outputs={},
397
                    attrs={'file_path': os.path.normpath(save_file_path)})
398 399 400
            else:
                save_var_map[new_var.name] = new_var

401
        if filename is not None or save_to_memory:
402 403 404 405
            save_var_list = []
            for name in sorted(save_var_map.keys()):
                save_var_list.append(save_var_map[name])

406 407 408 409
            save_path = str()
            if save_to_memory is False:
                save_path = os.path.join(os.path.normpath(dirname), filename)

410 411
            saved_params = save_block.create_var(type=core.VarDesc.VarType.RAW,
                                                 name=params_var_name)
412
            saved_params.desc.set_persistable(True)
413 414 415 416 417 418 419
            save_block.append_op(type='save_combine',
                                 inputs={'X': save_var_list},
                                 outputs={'Y': saved_params},
                                 attrs={
                                     'file_path': save_path,
                                     'save_to_memory': save_to_memory
                                 })
420

421
        # NOTE(zhiqiu): save op will add variable kLookupTablePath in save_program.desc,
422 423 424
        # which leads to diff on save_program and its desc. Call _sync_with_cpp
        # to keep consistency.
        save_program._sync_with_cpp()
425
        executor.run(save_program)
426 427
        if save_to_memory:
            return global_scope().find_var(params_var_name).get_bytes()
428 429


430
@dygraph_not_support
431
def save_params(executor, dirname, main_program=None, filename=None):
432
    """
433
    Save all parameters from the :code:`main_program` to
434
    the folder :code:`dirname` or file :code:`filename`. You can refer to
G
guofei 已提交
435
    :ref:`api_guide_model_save_reader_en` for more details.
F
fengjiayi 已提交
436

G
guofei 已提交
437 438 439
    Use the :code:`dirname` to specify the saving folder. If you would like to
    save parameters in separate files, set :code:`filename` None; if you would
    like to save all parameters in a single file, use :code:`filename` to specify
F
fengjiayi 已提交
440 441
    the file name.

442
    Note:
G
guofei 已提交
443
        Some variables are not Parameter while they are necessary for
444
        training, such as learning rate, global step, etc. So you can NOT save
G
guofei 已提交
445 446
        and continue your training just by :ref:`api_fluid_io_save_params`
        and :ref:`api_fluid_io_load_params`. Please use :ref:`api_fluid_io_save_persistables`
447 448 449
        and :ref:`api_fluid_io_load_persistables` instead.

        If you want to save your model for the inference, please use the
G
guofei 已提交
450 451
        :ref:`api_fluid_io_save_inference_model`. You can refer to
        :ref:`api_guide_model_save_reader_en` for more details.
F
fengjiayi 已提交
452 453

    Args:
454
        executor(Executor): The executor to run for saving parameters, You can
G
guofei 已提交
455
                            refer to :ref:`api_guide_executor_en`.
456 457
        dirname(str, optional): The saving directory path.
                            When you need to save the parameter to the memory, set it to None.
G
guofei 已提交
458
        main_program(Program, optional): The program whose parameters will be
459 460
                                         saved. You can refer to
                                         :ref:`api_guide_Program_en` for more
G
guofei 已提交
461 462 463 464 465 466 467
                                         details. If it is None, the default main
                                         program will be used.
                                         Default: None
        filename(str, optional): The file to save all parameters. If you prefer
                                 to save parameters in different files, set it
                                 to None.
                                 Default: None
F
fengjiayi 已提交
468 469

    Returns:
470 471
        str: When saving parameters to a file, returns None.
             When saving parameters to memory, returns a binary string containing parameters.
F
fengjiayi 已提交
472 473 474 475

    Examples:
        .. code-block:: python

476
            import paddle
H
Huihuang Zheng 已提交
477
            import paddle.fluid as fluid
478

479 480

            paddle.enable_static()
G
guofei 已提交
481 482 483 484 485
            params_path = "./my_paddle_model"
            image = fluid.data(name='img', shape=[None, 28, 28], dtype='float32')
            label = fluid.data(name='label', shape=[None, 1], dtype='int64')
            feeder = fluid.DataFeeder(feed_list=[image, label], place=fluid.CPUPlace())
            predict = fluid.layers.fc(input=image, size=10, act='softmax')
486

G
guofei 已提交
487
            loss = fluid.layers.cross_entropy(input=predict, label=label)
488
            avg_loss = paddle.mean(loss)
489

F
fengjiayi 已提交
490
            exe = fluid.Executor(fluid.CPUPlace())
G
guofei 已提交
491 492
            exe.run(fluid.default_startup_program())
            fluid.io.save_params(executor=exe, dirname=params_path)
493 494
            # The parameters weights and bias of the fc layer in the network are going to
            # be saved in different files in the path "./my_paddle_model"
495
    """
496 497 498 499 500 501
    return save_vars(executor,
                     dirname=dirname,
                     main_program=main_program,
                     vars=None,
                     predicate=is_parameter,
                     filename=filename)
502 503


504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525
def _save_distributed_persistables(executor, dirname, main_program):
    """
    save_persistables for distributed training.
    the method will do things listed below:
    1.save part of persistable variables on trainer.
    2.receive "remote prefetch variables" from parameter servers and merge them.
    3.save "distributed lookup table" on parameter servers.
    4.receive "optimizer variables" from parameter servers and merge them.

    Args:
        executor(Executor): The executor to run for saving parameters.
        dirname(str): The saving directory path.
        main_program(Program): The program whose parameters will be
                            saved. the main_program must be the trainer_program
                            get after transpiler.

    Returns:
        None

    Examples:
        .. code-block:: python

526
            import paddle
527
            import paddle.fluid as fluid
528 529

            paddle.enable_static()
530 531 532 533 534 535 536 537 538 539
            exe = fluid.Executor(fluid.CPUPlace())
            param_path = "./my_paddle_model"
            t = distribute_transpiler.DistributeTranspiler()
            t.transpile(...)
            train_program = t.get_trainer_program()
            _save_distributed_persistables(executor=exe, dirname=param_path, main_program=train_program)
    """

    def __save_remote_params(executor, dirname, remote_params_map):
        """
T
tianshuo78520a 已提交
540
        receive params on pserver through rpc.
541 542 543 544 545 546 547 548 549 550
        if the params are be sliced, will concat them to one, then save it.
        """
        if not remote_params_map:
            return

        prog = Program()
        block = prog.global_block()

        # recv optimize vars from pserver
        for name, remote_params in remote_params_map.items():
T
tangwei12 已提交
551 552 553 554 555 556 557
            origin = remote_params[0].origin
            is_slice = remote_params[0].is_slice

            slices = [None] * len(remote_params)
            slice_varnames = [None] * len(remote_params)
            remote_varnames = [None] * len(remote_params)
            endpoints = [None] * len(remote_params)
558 559 560

            for idx, optimizer in enumerate(remote_params):
                block_id = optimizer.block_id
T
tangwei12 已提交
561
                slice = optimizer.slice
562 563 564
                endpoint = optimizer.endpoint

                index = block_id if is_slice else idx
T
tangwei12 已提交
565 566 567
                slices[index] = slice
                slice_varnames[index] = "{}.slice.{}".format(slice.name, idx)
                remote_varnames[index] = slice.name
568 569
                endpoints[index] = endpoint

T
tangwei12 已提交
570 571 572 573 574
            slice_shapes = []
            for slice in slices:
                tmp = [str(dim) for dim in slice.shape]
                slice_shapes.append(",".join(tmp))

575 576 577 578 579 580 581 582 583 584
            block.append_op(type='recv_save',
                            attrs={
                                "trainer_id": 0,
                                "shape": origin.shape,
                                "slice_shapes": slice_shapes,
                                "slice_varnames": slice_varnames,
                                "remote_varnames": remote_varnames,
                                "endpoints": endpoints,
                                "file_path": os.path.join(dirname, origin.name)
                            })
T
tangwei12 已提交
585

586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605
        executor.run(prog)

    def __save_distributed_lookup_tables(executor, dirname,
                                         distributed_lookup_table, endpoints):
        """
        because the distributed lookup table may too huge to merge and save at one place,
        it will be saved at parameter server independent respectively.

        the save directory is dirname/"__lookup_table__".

        """
        prog = Program()
        block = prog.global_block()

        # if there is lookup table, the trainer 0 will notify all pserver to save.
        lookup_table_filename = os.path.join(dirname, "__lookup_table__")
        attrs = {}
        attrs['epmap'] = endpoints
        attrs['dir'] = lookup_table_filename
        attrs['lookup_table'] = distributed_lookup_table
606 607 608 609
        block.append_op(type='checkpoint_notify',
                        inputs={},
                        outputs={},
                        attrs=attrs)
610 611 612
        executor.run(prog)

    def __exclude_vars(exclude_var_names=[]):
613

614 615 616 617
        def is_valid(var):
            if var.name in exclude_var_names:
                return False
            if var.desc.type() == core.VarDesc.VarType.FEED_MINIBATCH or \
618 619
                            var.desc.type() == core.VarDesc.VarType.FETCH_LIST or \
                            var.desc.type() == core.VarDesc.VarType.READER:
620 621 622 623 624 625
                return False
            return var.persistable

        return is_valid

    if not isinstance(main_program, Program):
T
tangwei12 已提交
626
        raise TypeError("'main_program' should be an instance of Program.")
627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647

    if not main_program._is_distributed:
        raise ValueError(
            "'_save_distributed_persistables' just be designed for distributed training."
        )

    remote_params_map = main_program._parameters_on_pservers.get_distributed_vars_by_vtypes(
        ["Optimizer", "RemotePrefetch"], groupby=True)

    exclude_var_names = []
    if remote_params_map:
        exclude_var_names.extend(remote_params_map.keys())

    if main_program._distributed_lookup_table:
        if isinstance(main_program._distributed_lookup_table, list):
            exclude_var_names.extend(main_program._distributed_lookup_table)
        else:
            exclude_var_names.append(main_program._distributed_lookup_table)

    local_vars = list(
        filter(__exclude_vars(exclude_var_names), main_program.list_vars()))
648 649 650 651
    save_vars(executor,
              main_program=main_program,
              dirname=dirname,
              vars=local_vars)
652 653 654 655 656 657 658 659 660 661

    if main_program._is_chief:
        if remote_params_map:
            __save_remote_params(executor, dirname, remote_params_map)
        if main_program._distributed_lookup_table:
            __save_distributed_lookup_tables(
                executor, dirname, main_program._distributed_lookup_table,
                main_program._endpoints)


662
@dygraph_not_support
663
def save_persistables(executor, dirname, main_program=None, filename=None):
664
    """
665
    Save all persistable variables from :code:`main_program` to 
G
guofei 已提交
666 667
    the folder :code:`dirname` or file :code:`filename`. You can refer to 
    :ref:`api_guide_model_save_reader_en` for more details. And then
668 669
    saves these persistables variables to the folder :code:`dirname` or file
    :code:`filename`.
F
fengjiayi 已提交
670

G
guofei 已提交
671
    The :code:`dirname` is used to specify the folder where persistable variables
672
    are going to be saved. If you would like to save variables in separate
G
guofei 已提交
673 674
    files, set :code:`filename` None; if you would like to save all variables in a
    single file, use :code:`filename` to specify the file name.
F
fengjiayi 已提交
675 676 677

    Args:
        executor(Executor): The executor to run for saving persistable variables.
678
                            You can refer to :ref:`api_guide_executor_en` for
G
guofei 已提交
679
                            more details.
680

681 682 683
        dirname(str, optional): The saving directory path.
                            When you need to save the parameter to the memory, set it to None.
        main_program(Program, optional): The program whose persistbale variables will
G
guofei 已提交
684 685
                                         be saved. You can refer to 
                                         :ref:`api_guide_Program_en` for more details.
686
                                         If it is None, the default main program will
G
guofei 已提交
687 688 689 690 691
                                         be used.
                                         Default: None.
        filename(str, optional): The file to save all variables. If you prefer to
                                 save variables in different files, set it to None.
                                 Default: None.
F
fengjiayi 已提交
692 693

    Returns:
694 695
        str: When saving parameters to a file, returns None.
             When saving parameters to memory, returns a binary string containing parameters.
F
fengjiayi 已提交
696 697 698 699

    Examples:
        .. code-block:: python

700
            import paddle
H
Huihuang Zheng 已提交
701
            import paddle.fluid as fluid
702

703
            paddle.enable_static()
G
guofei 已提交
704 705 706 707 708
            dir_path = "./my_paddle_model"
            file_name = "persistables"
            image = fluid.data(name='img', shape=[None, 28, 28], dtype='float32')
            label = fluid.data(name='label', shape=[None, 1], dtype='int64')
            feeder = fluid.DataFeeder(feed_list=[image, label], place=fluid.CPUPlace())
709

G
guofei 已提交
710 711
            predict = fluid.layers.fc(input=image, size=10, act='softmax')
            loss = fluid.layers.cross_entropy(input=predict, label=label)
712
            avg_loss = paddle.mean(loss)
F
fengjiayi 已提交
713
            exe = fluid.Executor(fluid.CPUPlace())
G
guofei 已提交
714 715
            exe.run(fluid.default_startup_program())
            fluid.io.save_persistables(executor=exe, dirname=dir_path, filename=file_name)
716
            # The persistables variables weights and bias in the fc layer of the network
G
guofei 已提交
717 718
            # are going to be saved in the same file named "persistables" in the path
            # "./my_paddle_model"
719
    """
720
    if main_program and main_program._is_distributed:
721 722 723
        return _save_distributed_persistables(executor,
                                              dirname=dirname,
                                              main_program=main_program)
724
    else:
725 726 727 728 729 730
        return save_vars(executor,
                         dirname=dirname,
                         main_program=main_program,
                         vars=None,
                         predicate=is_persistable,
                         filename=filename)
731 732


733 734 735 736 737
def load_vars(executor,
              dirname,
              main_program=None,
              vars=None,
              predicate=None,
738
              filename=None):
739
    """
740 741
    :api_attr: Static Graph

742
    This API loads variables from files by executor.
F
fengjiayi 已提交
743

744
    There are two ways to specify the variables to be loaded: the first way, set
745 746
    variables in a list and assign it to the `vars`; the second way, use the
    `predicate` function to select variables that make `predicate(variable) == True`.
747
    The first way has a higher priority.
F
fengjiayi 已提交
748

749
    The `dirname` is used to specify the folder where to load variables.
750
    If variables were saved in separate files in the folder `dirname`,
751
    set `filename` None. If all variables were saved in a single file,
F
fengjiayi 已提交
752
    use `filename` to specify it.
753

F
fengjiayi 已提交
754 755
    Args:
        executor(Executor): The executor to run for loading variables.
756 757
        dirname(str): The folder where to load the variables.
        main_program(Program, optional): The program whose variables will be loaded.
758
                                    If it is None, the default main program will
F
fengjiayi 已提交
759 760
                                    be used automatically.
                                    Default: None
761
        vars(list[Variable], optional): The list that contains all variables to be loaded.
F
fengjiayi 已提交
762
                                   Default: None
763
        predicate(function, optional): The function selects variables that make
764 765 766 767 768
                                        `predicate(variable) == True`.
                                        Default: None
        filename(str, optional): The file which saved all required variables. If variables
                                were saved in separate files, set it to be None.
                                Default: None
F
fengjiayi 已提交
769 770 771 772 773 774 775

    Returns:
        None

    Examples:
        .. code-block:: python

776
            import paddle
777
            import paddle.fluid as fluid
778

779
            paddle.enable_static()
780 781 782 783 784 785 786 787 788 789 790
            main_prog = fluid.Program()
            startup_prog = fluid.Program()
            with fluid.program_guard(main_prog, startup_prog):
                data = fluid.layers.data(name="img", shape=[64, 784], append_batch_size=False)
                w = fluid.layers.create_parameter(shape=[784, 200], dtype='float32', name='fc_w')
                b = fluid.layers.create_parameter(shape=[200], dtype='float32', name='fc_b')
                hidden_w = fluid.layers.matmul(x=data, y=w)
                hidden_b = fluid.layers.elementwise_add(hidden_w, b)
            place = fluid.CPUPlace()
            exe = fluid.Executor(place)
            exe.run(startup_prog)
F
fengjiayi 已提交
791

792 793 794 795 796 797 798 799 800 801 802
            # The first usage: using `vars` to specify the variables.
            path = "./my_paddle_vars"
            var_list = [w, b]
            fluid.io.save_vars(executor=exe, dirname=path, vars=var_list,
                               filename="vars_file")
            fluid.io.load_vars(executor=exe, dirname=path, vars=var_list,
                               filename="vars_file")
            # w and b will be loaded, and they are supposed to
            # be saved in the same file named 'var_file' in the path "./my_paddle_vars".

            # The second usage: using the `predicate` function to select variables
803
            param_path = "./my_paddle_model"
F
fengjiayi 已提交
804 805 806
            def name_has_fc(var):
                res = "fc" in var.name
                return res
807 808 809
            fluid.io.save_vars(executor=exe, dirname=param_path, main_program=main_prog,
                              vars=None, predicate=name_has_fc)
            fluid.io.load_vars(executor=exe, dirname=param_path, main_program=main_prog,
C
chengduo 已提交
810
                               vars=None, predicate=name_has_fc)
811 812
            # Load All variables in the `main_program` whose name includes "fc".
            # And all the variables are supposed to be saved in separate files.
F
fengjiayi 已提交
813

814
    """
815 816 817 818 819
    vars_from_memory = False
    if dirname is not None:
        dirname = os.path.normpath(dirname)
    else:
        vars_from_memory = True
T
tangwei12 已提交
820

821
    if vars is None:
822
        if main_program is None:
Y
Yu Yang 已提交
823
            main_program = default_main_program()
824
        if not isinstance(main_program, Program):
825 826 827
            raise TypeError(
                "The type of input main_program is invalid, expected type is fluid.Program, but received %s"
                % type(main_program))
828

829 830 831 832 833
        load_vars(executor,
                  dirname=dirname,
                  main_program=main_program,
                  vars=list(filter(predicate, main_program.list_vars())),
                  filename=filename)
834 835 836
    else:
        load_prog = Program()
        load_block = load_prog.global_block()
837

838 839
        if main_program is None:
            main_program = default_main_program()
T
tangwei12 已提交
840

841
        if not isinstance(main_program, Program):
842 843 844
            raise TypeError(
                "The type of input main_program is invalid, expected type is fluid.Program, but received %s"
                % type(main_program))
845

T
tangwei12 已提交
846
        # save origin param shape
H
hong 已提交
847
        orig_para_shape = {}
848
        load_var_map = {}
849 850 851 852

        check_vars = []
        sparse_vars = []

853 854
        for each_var in vars:
            assert isinstance(each_var, Variable)
855

T
tangwei12 已提交
856 857
            if each_var.type == core.VarDesc.VarType.RAW:
                continue
H
hong 已提交
858 859

            if isinstance(each_var, Parameter):
860 861
                orig_para_shape[each_var.name] = tuple(
                    each_var.desc.get_shape())
862 863 864 865 866

            if each_var.type == core.VarDesc.VarType.SELECTED_ROWS:
                sparse_vars.append(each_var)
                continue

867
            new_var = _clone_var_in_block_(load_block, each_var)
868 869
            check_vars.append(each_var)

870
            if filename is None:
871 872 873 874
                if dirname is None:
                    raise ValueError(
                        "The directory path and params cannot be None at the same time."
                    )
875 876 877 878
                load_block.append_op(
                    type='load',
                    inputs={},
                    outputs={'Out': [new_var]},
879
                    attrs={'file_path': os.path.join(dirname, new_var.name)})
880 881 882
            else:
                load_var_map[new_var.name] = new_var

883 884 885 886 887 888 889 890 891 892 893
        for each_var in sparse_vars:
            assert isinstance(each_var, Variable)

            if filename is not None:
                raise ValueError(
                    "SelectedRows can not be load with load_combine")

            new_var = _clone_var_in_block_(load_block, each_var)

            var_path = os.path.join(dirname, new_var.name)
            if not os.path.exists(var_path):
894 895 896
                raise ValueError(
                    "SelectedRows var {} can not find at {}".format(
                        new_var.name, var_path))
897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913

            if os.path.isfile(var_path):
                load_block.append_op(
                    type='load',
                    inputs={},
                    outputs={'Out': [new_var]},
                    attrs={'file_path': os.path.join(dirname, new_var.name)})
            else:
                blocks = []
                block_paths = os.listdir(var_path)

                for block in block_paths:
                    if block.startswith(new_var.name):
                        blocks.append(block)

                slices = []
                for block in blocks:
914 915 916 917 918
                    slice = load_block.create_var(name=block,
                                                  type=new_var.type,
                                                  shape=new_var.shape,
                                                  dtype=new_var.dtype,
                                                  persistable=False)
919 920 921
                    slices.append(slice)

                    file_path = os.path.join(var_path, block, "Param")
922 923 924 925
                    load_block.append_op(type='load',
                                         inputs={},
                                         outputs={'Out': [slice]},
                                         attrs={'file_path': file_path})
926

927 928 929 930
                load_block.append_op(type='lookup_sparse_table_merge',
                                     inputs={'X': slices},
                                     outputs={'Out': new_var},
                                     attrs={})
931

932
        if filename is not None:
933 934 935 936
            load_var_list = []
            for name in sorted(load_var_map.keys()):
                load_var_list.append(load_var_map[name])

937 938 939
            if vars_from_memory is False:
                filename = os.path.join(dirname, filename)

940 941 942 943 944 945 946
            load_block.append_op(type='load_combine',
                                 inputs={},
                                 outputs={"Out": load_var_list},
                                 attrs={
                                     'file_path': filename,
                                     'model_from_memory': vars_from_memory
                                 })
947 948
        executor.run(load_prog)

T
tangwei12 已提交
949
        # check var shape
950
        for each_var in check_vars:
H
hong 已提交
951 952 953 954 955
            if not isinstance(each_var, Parameter):
                continue
            var_temp = paddle.fluid.global_scope().find_var(each_var.name)
            assert var_temp != None, "can't not find var: " + each_var.name
            new_shape = (np.array(var_temp.get_tensor())).shape
956
            assert each_var.name in orig_para_shape, each_var.name + "MUST in var list"
H
hong 已提交
957 958 959
            orig_shape = orig_para_shape.get(each_var.name)
            if new_shape != orig_shape:
                raise RuntimeError(
960
                    "Variable's shape does not match, the Program requires a parameter with the shape of ({}), "
961 962
                    "while the loaded parameter (namely [ {} ]) has a shape of  ({})."
                    .format(orig_shape, each_var.name, new_shape))
H
hong 已提交
963

964

965
@dygraph_not_support
966
def load_params(executor, dirname, main_program=None, filename=None):
967
    """
968 969
    :api_attr: Static Graph

970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988
    This API filters out all parameters from the give ``main_program``
    and then tries to load these parameters from the directory ``dirname`` or
    the file ``filename``.

    Use the ``dirname`` to specify the directory where parameters were saved. If
    parameters were saved in separate files under the directory `dirname`, set
    ``filename`` as None; if all parameters were saved in a single file, use
    ``filename`` to specify the file name.

    **Note**:
        Some variables are not Parameter while they are necessary for
        training, such as learning rate, global step, etc. So you cannot save and
        continue your training just by using :ref:`api_fluid_io_save_params` and
        :ref:`api_fluid_io_load_params`. Please use :ref:`api_fluid_io_save_persistables`
        and :ref:`api_fluid_io_load_persistables` instead.

        If you want to load the pre-trained model structure and parameters
        for the inference, please use the :ref:`api_fluid_io_load_inference_model` API. You can
        refer to :ref:`api_guide_model_save_reader_en` for more details.
F
fengjiayi 已提交
989 990

    Args:
991 992
        executor(Executor): The executor used for loading parameters.
                            See :ref:`api_guide_executor_en` for more details about it.
F
fengjiayi 已提交
993
        dirname(str): The directory path.
994 995 996 997 998 999 1000 1001
        main_program(Program, optional): The program whose parameters will be
                                    loaded. If it is None, the ``default_main_program``
                                    will be used automatically. See :ref:`api_guide_Program_en`
                                    for more about ``Program``.
                                    Default: None.
        filename(str, optional): The file which saved all parameters. If parameters
                            were saved in separated files, set it to None.
                            Default: None.
F
fengjiayi 已提交
1002 1003 1004 1005 1006 1007 1008

    Returns:
        None

    Examples:
        .. code-block:: python

1009
            import paddle
1010
            import paddle.fluid as fluid
1011

1012
            paddle.enable_static()
F
fengjiayi 已提交
1013 1014 1015
            exe = fluid.Executor(fluid.CPUPlace())
            param_path = "./my_paddle_model"
            prog = fluid.default_main_program()
1016
            fluid.io.load_params(executor=exe, dirname=param_path,
F
fengjiayi 已提交
1017
                                main_program=None)
1018
    """
1019 1020 1021 1022 1023
    load_vars(executor,
              dirname=dirname,
              main_program=main_program,
              predicate=is_parameter,
              filename=filename)
1024 1025


1026
@dygraph_not_support
1027
def load_persistables(executor, dirname, main_program=None, filename=None):
1028
    """
1029
    :api_attr: Static Graph
1030

1031 1032
    This API filters out all variables with ``persistable==True`` from the
    given ``main_program`` and then tries to load these variables from the
T
tianshuo78520a 已提交
1033
    directory ``dirname`` or the file ``filename``.
F
fengjiayi 已提交
1034

1035 1036 1037 1038
    Use the ``dirname`` to specify the directory where persistable variables
    (refer to :ref:`api_guide_model_save_reader_en`) were saved. If variables
    were saved in separate files, set ``filename`` as None; if all variables
    were saved in a single file, use ``filename`` to specify the file name.
F
fengjiayi 已提交
1039 1040

    Args:
1041 1042
        executor(Executor): The executor used for loading persistable variables.
                            See :ref:`api_guide_executor_en` for more details about it.
F
fengjiayi 已提交
1043
        dirname(str): The directory path.
T
tianshuo78520a 已提交
1044
        main_program(Program, optional): The program whose persistable variables will
1045 1046 1047 1048 1049 1050 1051
                                    be loaded. If it is None, the ``default_main_program``
                                    will be used automatically. See :ref:`api_guide_Program_en`
                                    for more about ``Program``.
                                    Default: None.
        filename(str, optional): The file which saved all persistable variables. If variables
                                 were saved in separated files, set it to None.
                                 Default: None.
F
fengjiayi 已提交
1052 1053 1054 1055 1056 1057 1058

    Returns:
        None

    Examples:
        .. code-block:: python

1059
            import paddle
1060
            import paddle.fluid as fluid
1061

1062
            paddle.enable_static()
F
fengjiayi 已提交
1063 1064 1065
            exe = fluid.Executor(fluid.CPUPlace())
            param_path = "./my_paddle_model"
            prog = fluid.default_main_program()
1066
            fluid.io.load_persistables(executor=exe, dirname=param_path,
F
fengjiayi 已提交
1067
                                       main_program=None)
1068
    """
1069 1070

    if main_program and main_program._is_distributed:
1071 1072 1073
        _load_distributed_persistables(executor,
                                       dirname=dirname,
                                       main_program=main_program)
1074
    else:
1075 1076 1077 1078 1079
        load_vars(executor,
                  dirname=dirname,
                  main_program=main_program,
                  predicate=is_persistable,
                  filename=filename)
1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099


def _load_distributed_persistables(executor, dirname, main_program=None):
    """
    customized load_persistables for distributed training.
    it should be used on parameter server,

    Args:
        executor(Executor): The executor to run for saving parameters.
        dirname(str): The load directory path.
        main_program(Program): The program whose parameters will be
                            loaded. the main_program must be the pserver_program
                            get after transpiler.

    Returns:
        None

    Examples:
        .. code-block:: python

1100
            import paddle
1101
            import paddle.fluid as fluid
1102 1103

            paddle.enable_static()
1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128
            exe = fluid.Executor(fluid.CPUPlace())
            param_path = "./my_paddle_model"
            t = distribute_transpiler.DistributeTranspiler()
            t.transpile(...)
            pserver_prog = t.get_pserver_program(...)
            _load_distributed_persistables(executor=exe, dirname=param_path, main_program=pserver_prog)
    """

    def __is_distributed_part_var(varname):
        trainer_idx = varname.find(".trainer_")
        block_idx = varname.find(".block")
        return trainer_idx or block_idx

    def __load_persistable_vars(executor, dirname, need_load_vars):
        load_prog = Program()
        load_block = load_prog.global_block()
        need_delete_vars = []

        for param in need_load_vars:
            origin_var = param.origin
            slice_var = param.slice
            is_slice = param.is_slice
            offset = param.offset

            if is_slice:
1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145
                slice = load_block.create_var(name=slice_var.name,
                                              type=slice_var.type,
                                              shape=slice_var.shape,
                                              dtype=slice_var.dtype,
                                              persistable=True)

                load_block.append_op(type='load',
                                     inputs={},
                                     outputs={'Out': [slice]},
                                     attrs={
                                         'file_path':
                                         os.path.join(dirname, origin_var.name),
                                         'seek':
                                         offset,
                                         'shape':
                                         slice.shape
                                     })
1146
            else:
1147 1148 1149 1150 1151 1152
                origin = load_block.create_var(name="{}".format(
                    origin_var.name),
                                               type=origin_var.type,
                                               shape=origin_var.shape,
                                               dtype=origin_var.dtype,
                                               persistable=True)
1153 1154 1155 1156
                load_block.append_op(
                    type='load',
                    inputs={},
                    outputs={'Out': [origin]},
1157
                    attrs={'file_path': os.path.join(dirname, origin_var.name)})
1158 1159 1160

        load_block.append_op(
            type='delete_var',
1161 1162
            inputs={'X': need_delete_vars},
        )
1163 1164 1165 1166

        executor.run(load_prog)

    if not isinstance(main_program, Program):
T
tangwei12 已提交
1167
        raise TypeError("'main_program' should be an instance of Program.")
1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181

    if not main_program._is_distributed:
        raise ValueError(
            "'_load_distributed_persistables' just be designed for distributed training."
        )

    if not main_program._ps_endpoint:
        raise ValueError(
            "'_load_distributed_persistables' need current_endpoint set in DistributeTranspiler.transpile"
        )

    need_load_vars = main_program._parameters_on_pservers.get_distributed_vars_by_ep(
        main_program._ps_endpoint)
    __load_persistable_vars(executor, dirname, need_load_vars)
1182 1183


1184 1185 1186
def prepend_feed_ops(inference_program,
                     feed_target_names,
                     feed_holder_name='feed'):
Q
Qiao Longfei 已提交
1187 1188 1189
    if len(feed_target_names) == 0:
        return

K
Kexin Zhao 已提交
1190
    global_block = inference_program.global_block()
1191 1192 1193
    feed_var = global_block.create_var(name=feed_holder_name,
                                       type=core.VarDesc.VarType.FEED_MINIBATCH,
                                       persistable=True)
K
Kexin Zhao 已提交
1194

1195
    for i, name in enumerate(feed_target_names):
1196 1197 1198 1199 1200
        if not global_block.has_var(name):
            raise ValueError(
                "The feeded_var_names[{i}]: '{name}' doesn't exist in pruned inference program. "
                "Please check whether '{name}' is a valid feed_var name, or remove it from feeded_var_names "
                "if '{name}' is not involved in the target_vars calculation.".
1201
                format(i=i, name=name))
K
fix bug  
Kexin Zhao 已提交
1202
        out = global_block.var(name)
1203 1204 1205 1206
        global_block._prepend_op(type='feed',
                                 inputs={'X': [feed_var]},
                                 outputs={'Out': [out]},
                                 attrs={'col': i})
K
Kexin Zhao 已提交
1207 1208


1209 1210 1211
def append_fetch_ops(inference_program,
                     fetch_target_names,
                     fetch_holder_name='fetch'):
K
Kexin Zhao 已提交
1212
    global_block = inference_program.global_block()
1213 1214 1215
    fetch_var = global_block.create_var(name=fetch_holder_name,
                                        type=core.VarDesc.VarType.FETCH_LIST,
                                        persistable=True)
K
Kexin Zhao 已提交
1216

1217
    for i, name in enumerate(fetch_target_names):
1218 1219 1220 1221
        global_block.append_op(type='fetch',
                               inputs={'X': [name]},
                               outputs={'Out': [fetch_var]},
                               attrs={'col': i})
K
Kexin Zhao 已提交
1222 1223


1224 1225
@static_only
@deprecated(since="2.0.0", update_to="paddle.static.save_inference_model")
1226 1227 1228 1229
def save_inference_model(dirname,
                         feeded_var_names,
                         target_vars,
                         executor,
1230
                         main_program=None,
1231
                         model_filename=None,
1232
                         params_filename=None,
T
tangwei12 已提交
1233
                         export_for_deployment=True,
1234 1235
                         program_only=False,
                         clip_extra=False):
1236
    """
F
fengjiayi 已提交
1237
    Prune the given `main_program` to build a new program especially for inference,
G
guofei 已提交
1238
    and then save it and all related parameters to given `dirname` .
1239
    If you just want to save parameters of your trained model, please use the
G
guofei 已提交
1240 1241
    :ref:`api_fluid_io_save_params` . You can refer to :ref:`api_guide_model_save_reader_en`
    for more details.
1242

G
guofei 已提交
1243
    Note:
1244
        The :code:`dirname` is used to specify the folder where inference model
G
guofei 已提交
1245
        structure and parameters are going to be saved. If you would like to save params of
1246
        Program in separate files, set `params_filename` None; if you would like to save all
G
guofei 已提交
1247
        params of Program in a single file, use `params_filename` to specify the file name.
F
fengjiayi 已提交
1248 1249 1250

    Args:
        dirname(str): The directory path to save the inference model.
T
tianshuo78520a 已提交
1251
        feeded_var_names(list[str]): list of string. Names of variables that need to be fed
G
guofei 已提交
1252
                                     data during inference.
1253
        target_vars(list[Variable]): list of Variable. Variables from which we can get
G
guofei 已提交
1254
                                     inference results.
1255
        executor(Executor): The executor that saves the inference model. You can refer
G
guofei 已提交
1256 1257
                            to :ref:`api_guide_executor_en` for more details.
        main_program(Program, optional): The original program, which will be pruned to
T
tianshuo78520a 已提交
1258
                                         build the inference model. If is set None,
G
guofei 已提交
1259 1260 1261
                                         the global default :code:`_main_program_` will be used.
                                         Default: None.
        model_filename(str, optional): The name of file to save the inference program
T
tianshuo78520a 已提交
1262
                                       itself. If is set None, a default filename
G
guofei 已提交
1263 1264
                                       :code:`__model__` will be used.
        params_filename(str, optional): The name of file to save all related parameters.
T
tianshuo78520a 已提交
1265
                                        If it is set None, parameters will be saved
G
guofei 已提交
1266
                                        in separate files .
1267
        export_for_deployment(bool, optional): If True, programs are modified to only support
X
Xin Pan 已提交
1268 1269 1270 1271
                                     direct inference deployment. Otherwise,
                                     more information will be stored for flexible
                                     optimization and re-training. Currently, only
                                     True is supported.
G
guofei 已提交
1272
                                     Default: True.
1273
        program_only(bool, optional): If True, It will save inference program only, and do not
G
guofei 已提交
1274 1275
                                      save params of Program.
                                      Default: False.
1276

F
fengjiayi 已提交
1277
    Returns:
1278
        list, The fetch variables' name list.
F
fengjiayi 已提交
1279 1280 1281

    Examples:
        .. code-block:: python
F
fengjiayi 已提交
1282

1283
            import paddle
1284 1285
            import paddle.fluid as fluid

1286
            paddle.enable_static()
F
fengjiayi 已提交
1287 1288
            path = "./infer_model"

T
tianshuo78520a 已提交
1289
            # User defined network, here a softmax regession example
G
guofei 已提交
1290 1291
            image = fluid.data(name='img', shape=[None, 28, 28], dtype='float32')
            label = fluid.data(name='label', shape=[None, 1], dtype='int64')
1292 1293 1294 1295
            feeder = fluid.DataFeeder(feed_list=[image, label], place=fluid.CPUPlace())
            predict = fluid.layers.fc(input=image, size=10, act='softmax')

            loss = fluid.layers.cross_entropy(input=predict, label=label)
1296
            avg_loss = paddle.mean(loss)
1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308

            exe = fluid.Executor(fluid.CPUPlace())
            exe.run(fluid.default_startup_program())

            # Feed data and train process

            # Save inference model. Note we don't save label and loss in this example
            fluid.io.save_inference_model(dirname=path,
                                          feeded_var_names=['img'],
                                          target_vars=[predict],
                                          executor=exe)

G
guofei 已提交
1309
            # In this example, the save_inference_mode inference will prune the default
1310
            # main program according to the network's input node (img) and output node(predict).
G
guofei 已提交
1311
            # The pruned inference program is going to be saved in the "./infer_model/__model__"
F
fengjiayi 已提交
1312
            # and parameters are going to be saved in separate files under folder
1313
            # "./infer_model".
1314 1315

    """
M
minqiyang 已提交
1316
    if isinstance(feeded_var_names, six.string_types):
F
fengjiayi 已提交
1317
        feeded_var_names = [feeded_var_names]
X
Xin Pan 已提交
1318
    elif export_for_deployment:
Q
Qiao Longfei 已提交
1319
        if len(feeded_var_names) > 0:
1320
            # TODO(paddle-dev): polish these code blocks
Q
Qiao Longfei 已提交
1321
            if not (bool(feeded_var_names) and all(
M
minqiyang 已提交
1322
                    isinstance(name, six.string_types)
1323
                    for name in feeded_var_names)):
M
minqiyang 已提交
1324
                raise ValueError("'feed_var_names' should be a list of str.")
F
fengjiayi 已提交
1325 1326

    if isinstance(target_vars, Variable):
F
fengjiayi 已提交
1327
        target_vars = [target_vars]
X
Xin Pan 已提交
1328
    elif export_for_deployment:
1329 1330
        if not (bool(target_vars)
                and all(isinstance(var, Variable) for var in target_vars)):
F
fengjiayi 已提交
1331 1332
            raise ValueError("'target_vars' should be a list of Variable.")

C
chengduo 已提交
1333
    main_program = _get_valid_program(main_program)
T
tangwei12 已提交
1334

1335
    # remind user to set auc_states to zeros if the program contains auc op
1336 1337
    all_ops = main_program.global_block().ops
    for op in all_ops:
1338 1339 1340
        # clear device of Op
        device_attr_name = core.op_proto_and_checker_maker.kOpDeviceAttrName()
        op._set_attr(device_attr_name, "")
1341 1342 1343 1344 1345 1346
        if op.type == 'auc':
            warnings.warn(
                "please ensure that you have set the auc states to zeros before saving inference model"
            )
            break

1347 1348
    with program_guard(main_program):
        uniq_target_vars = []
F
flame 已提交
1349 1350
        for i, var in enumerate(target_vars):
            uniq_target_vars.append(var)
1351
        target_vars = uniq_target_vars
F
flame 已提交
1352
    target_var_name_list = [var.name for var in target_vars]
1353

1354
    # when a pserver and a trainer running on the same machine, mkdir may conflict
L
lujun 已提交
1355
    save_dirname = dirname
1356
    try:
L
lujun 已提交
1357 1358
        save_dirname = os.path.normpath(dirname)
        os.makedirs(save_dirname)
1359 1360 1361 1362
    except OSError as e:
        if e.errno != errno.EEXIST:
            raise

X
Xin Pan 已提交
1363 1364 1365 1366
    if model_filename is not None:
        model_basename = os.path.basename(model_filename)
    else:
        model_basename = "__model__"
L
lujun 已提交
1367
    model_basename = os.path.join(save_dirname, model_basename)
1368

X
Xin Pan 已提交
1369 1370 1371 1372
    # When export_for_deployment is true, we modify the program online so that
    # it can only be loaded for inference directly. If it's false, the whole
    # original program and related meta are saved so that future usage can be
    # more flexible.
1373 1374 1375

    origin_program = main_program.clone()

X
Xin Pan 已提交
1376
    if export_for_deployment:
X
Xin Pan 已提交
1377 1378
        main_program = main_program.clone()
        global_block = main_program.global_block()
1379
        need_to_remove_op_index = []
X
Xin Pan 已提交
1380 1381 1382
        for i, op in enumerate(global_block.ops):
            op.desc.set_is_target(False)
            if op.type == "feed" or op.type == "fetch":
1383 1384 1385 1386 1387
                need_to_remove_op_index.append(i)

        for index in need_to_remove_op_index[::-1]:
            global_block._remove_op(index)

X
Xin Pan 已提交
1388
        main_program.desc.flush()
X
Xin Pan 已提交
1389

1390 1391
        main_program = main_program._prune_with_input(
            feeded_var_names=feeded_var_names, targets=target_vars)
X
Xin Pan 已提交
1392
        main_program = main_program._inference_optimize(prune_read_op=True)
X
Xin Pan 已提交
1393 1394
        fetch_var_names = [v.name for v in target_vars]

1395 1396 1397 1398 1399
        for target_v in target_vars:
            if not main_program.global_block().has_var(target_v.name):
                main_program.global_block().create_var(
                    name=target_v.name,
                    shape=target_v.shape,
1400 1401
                    dtype=target_v.dtype,
                    persistable=target_v.persistable)
1402

X
Xin Pan 已提交
1403 1404 1405
        prepend_feed_ops(main_program, feeded_var_names)
        append_fetch_ops(main_program, fetch_var_names)

1406
        main_program.desc._set_version()
1407
        paddle.fluid.core.save_op_version_info(main_program.desc)
X
Xin Pan 已提交
1408
        with open(model_basename, "wb") as f:
1409
            f.write(
1410 1411
                main_program._remove_training_info(
                    clip_extra=clip_extra).desc.serialize_to_string())
X
Xin Pan 已提交
1412 1413 1414
    else:
        # TODO(panyx0718): Save more information so that it can also be used
        # for training and more flexible post-processing.
X
Xin Pan 已提交
1415
        with open(model_basename + ".main_program", "wb") as f:
1416
            f.write(
1417 1418
                main_program._remove_training_info(
                    clip_extra=clip_extra).desc.serialize_to_string())
T
tangwei12 已提交
1419

T
tangwei12 已提交
1420 1421 1422 1423 1424 1425
    if program_only:
        warnings.warn(
            "save_inference_model specified the param `program_only` to True, It will not save params of Program."
        )
        return target_var_name_list

1426 1427
    main_program._copy_dist_param_info_from(origin_program)

X
fix  
Xin Pan 已提交
1428 1429
    if params_filename is not None:
        params_filename = os.path.basename(params_filename)
1430

L
lujun 已提交
1431
    save_persistables(executor, save_dirname, main_program, params_filename)
F
flame 已提交
1432
    return target_var_name_list
X
fix  
Xin Pan 已提交
1433

1434

1435 1436
@static_only
@deprecated(since="2.0.0", update_to="paddle.static.load_inference_model")
1437 1438 1439
def load_inference_model(dirname,
                         executor,
                         model_filename=None,
T
tangwei12 已提交
1440 1441
                         params_filename=None,
                         pserver_endpoints=None):
1442
    """
1443 1444 1445
    Load the inference model from a given directory. By this API, you can get the model
    structure(Inference Program) and model parameters. If you just want to load
    parameters of the pre-trained model, please use the :ref:`api_fluid_io_load_params` API.
1446
    You can refer to :ref:`api_guide_model_save_reader_en` for more details.
1447

F
fengjiayi 已提交
1448
    Args:
1449 1450 1451
        dirname(str): One of the following:
          - The given directory path.
          - Set to None when reading the model from memory.
F
fengjiayi 已提交
1452
        executor(Executor): The executor to run for loading inference model.
1453
                            See :ref:`api_guide_executor_en` for more details about it.
1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464
        model_filename(str, optional): One of the following:
          - The name of file to load the inference program.
          - If it is None, the default filename ``__model__`` will be used.
          - When ``dirname`` is ``None``, it must be set to a string containing model.
          Default: ``None``.
        params_filename(str, optional): It is only used for the case that all
            parameters were saved in a single binary file. One of the following:
          - The name of file to load all parameters.  
          - When ``dirname`` is ``None``, it must be set to a string containing all the parameters.
          - If parameters were saved in separate files, set it as ``None``.
            Default: ``None``.
1465 1466 1467 1468

        pserver_endpoints(list, optional): It is only needed by the distributed inference.
                                    If using a distributed look up table during the training,
                                    this table is also needed by the inference process. Its value is
1469
                                    a list of pserver endpoints.
F
fengjiayi 已提交
1470 1471

    Returns:
1472
        list: The return of this API is a list with three elements:
1473
        (program, feed_target_names, fetch_targets). The `program` is a
1474 1475 1476 1477 1478
        ``Program`` (refer to :ref:`api_guide_Program_en`), which is used for inference.
        The `feed_target_names` is a list of ``str``, which contains names of variables
        that need to feed data in the inference program. The `fetch_targets` is a list of
        ``Variable`` (refer to :ref:`api_guide_Program_en`). It contains variables from which
        we can get inference results.
F
fengjiayi 已提交
1479 1480 1481 1482 1483


    Examples:
        .. code-block:: python

1484
            import paddle
1485 1486
            import paddle.fluid as fluid
            import numpy as np
1487

1488
            paddle.enable_static()
1489
            # Build the model
1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500
            main_prog = fluid.Program()
            startup_prog = fluid.Program()
            with fluid.program_guard(main_prog, startup_prog):
                data = fluid.layers.data(name="img", shape=[64, 784], append_batch_size=False)
                w = fluid.layers.create_parameter(shape=[784, 200], dtype='float32')
                b = fluid.layers.create_parameter(shape=[200], dtype='float32')
                hidden_w = fluid.layers.matmul(x=data, y=w)
                hidden_b = fluid.layers.elementwise_add(hidden_w, b)
            place = fluid.CPUPlace()
            exe = fluid.Executor(place)
            exe.run(startup_prog)
1501 1502

            # Save the inference model
F
fengjiayi 已提交
1503
            path = "./infer_model"
1504 1505
            fluid.io.save_inference_model(dirname=path, feeded_var_names=['img'],
                         target_vars=[hidden_b], executor=exe, main_program=main_prog)
1506 1507 1508

            # Demo one. Not need to set the distributed look up table, because the
            # training doesn't use a distributed look up table.
1509 1510
            [inference_program, feed_target_names, fetch_targets] = (
                fluid.io.load_inference_model(dirname=path, executor=exe))
1511
            tensor_img = np.array(np.random.random((1, 64, 784)), dtype=np.float32)
F
fengjiayi 已提交
1512 1513 1514 1515
            results = exe.run(inference_program,
                          feed={feed_target_names[0]: tensor_img},
                          fetch_list=fetch_targets)

1516 1517 1518
            # Demo two. If the training uses a distributed look up table, the pserver
            # endpoints list should be supported when loading the inference model.
            # The below is just an example.
1519
            endpoints = ["127.0.0.1:2023","127.0.0.1:2024"]
1520
            [dist_inference_program, dist_feed_target_names, dist_fetch_targets] = (
1521 1522
                fluid.io.load_inference_model(dirname=path,
                                              executor=exe,
1523
                                              pserver_endpoints=endpoints))
1524

1525
            # In this example, the inference program was saved in the file
1526
            # "./infer_model/__model__" and parameters were saved in
1527 1528 1529 1530
            # separate files under the directory "./infer_model".
            # By the inference program, feed_target_names and
            # fetch_targets, we can use an executor to run the inference
            # program for getting the inference result.
1531
    """
1532 1533 1534 1535
    load_from_memory = False
    if dirname is not None:
        load_dirname = os.path.normpath(dirname)
        if not os.path.isdir(load_dirname):
1536
            raise ValueError("There is no directory named '%s'" % dirname)
1537

1538 1539
        if model_filename is None:
            model_filename = '__model__'
1540

1541 1542
        model_filename = os.path.join(load_dirname,
                                      os.path.basename(model_filename))
1543

1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557
        if params_filename is not None:
            params_filename = os.path.basename(params_filename)

        with open(model_filename, "rb") as f:
            program_desc_str = f.read()
    else:
        load_from_memory = True
        if params_filename is None:
            raise ValueError(
                "The path of params cannot be None when the directory path is None."
            )
        load_dirname = dirname
        program_desc_str = model_filename
        params_filename = params_filename
1558

1559
    program = Program.parse_from_string(program_desc_str)
X
Xin Pan 已提交
1560
    if not core._is_program_version_supported(program._version()):
X
version  
Xin Pan 已提交
1561 1562 1563
        raise ValueError("Unsupported program version: %d\n" %
                         program._version())
    # Binary data also need versioning.
L
lujun 已提交
1564
    load_persistables(executor, load_dirname, program, params_filename)
1565

T
tangwei12 已提交
1566
    if pserver_endpoints:
T
tangwei12 已提交
1567
        program = _endpoints_replacement(program, pserver_endpoints)
T
tangwei12 已提交
1568

1569 1570
    feed_target_names = program.desc.get_feed_target_names()
    fetch_target_names = program.desc.get_fetch_target_names()
1571 1572 1573 1574 1575
    fetch_targets = [
        program.global_block().var(name) for name in fetch_target_names
    ]

    return [program, feed_target_names, fetch_targets]
X
xuwei06 已提交
1576 1577


T
tangwei12 已提交
1578 1579 1580
def _endpoints_replacement(program, endpoints):
    ENDPOINT_MAP = "epmap"
    for op in program.global_block().ops:
T
tangwei12 已提交
1581 1582
        if op.has_attr(ENDPOINT_MAP):
            op.set_attr(ENDPOINT_MAP, endpoints)
T
fix  
tangwei12 已提交
1583
    program._sync_with_cpp()
T
tangwei12 已提交
1584
    return program
T
tangwei12 已提交
1585 1586


X
xuwei06 已提交
1587 1588
def get_parameter_value(para, executor):
    """
F
fengjiayi 已提交
1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599
    Get the LoDTensor value of the given parameter.

    Args:
        para(Parameter): The parameter to get value from.
        executor(Executor): The executor to run for retrieving the value.

    Returns:
        numpy.array: The given parameter's values.

    Raises:
        AssertionError: If the `para` is not an instance of Parameter.
X
xuwei06 已提交
1600

F
fengjiayi 已提交
1601 1602
    Examples:
        .. code-block:: python
X
xuwei06 已提交
1603

1604
            import paddle
1605
            import paddle.fluid as fluid
1606 1607

            paddle.enable_static()
F
fengjiayi 已提交
1608 1609 1610
            exe = fluid.Executor(fluid.CPUPlace())
            param = fluid.default_main_program().global_block().var('fc.w')
            p = fluid.io.get_parameter_value(param, exe)
1611

X
xuwei06 已提交
1612
    """
1613
    assert is_parameter(para), "The input variable is not parameter."
X
xuwei06 已提交
1614

X
xuwei06 已提交
1615 1616 1617 1618 1619 1620 1621 1622
    get_program = Program()
    block = get_program.global_block()
    new_var = _clone_var_in_block_(block, para)
    return executor.run(get_program, feed={}, fetch_list=[new_var])[0]


def get_parameter_value_by_name(name, executor, program=None):
    """
F
fengjiayi 已提交
1623
    Get the LoDTensor value of a certain parameter by its name.
X
xuwei06 已提交
1624

F
fengjiayi 已提交
1625 1626 1627 1628 1629 1630 1631
    Args:
        name(str): The parameter's name.
        executor(Executor): The executor to run for retrieving the value.
        program(Program | None): The program where to find the parameter.
                               If it's set to be None, the function will
                               try to find the parameter in the default
                               main program.
X
xuwei06 已提交
1632

F
fengjiayi 已提交
1633 1634
    Returns:
        numpy.array: The parameter's values.
1635

F
fengjiayi 已提交
1636 1637 1638
    Examples:
        .. code-block:: python

1639
            import paddle
1640
            import paddle.fluid as fluid
1641 1642

            paddle.enable_static()
F
fengjiayi 已提交
1643 1644
            exe = fluid.Executor(fluid.CPUPlace())
            p = fluid.io.get_parameter_value('fc.w', exe)
X
xuwei06 已提交
1645 1646
    """
    if program is None:
Y
Yu Yang 已提交
1647
        program = default_main_program()
X
xuwei06 已提交
1648 1649
    var = program.global_block().var(name)
    return get_parameter_value(var, executor)
1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673


def _save_persistable_nodes(executor, dirname, graph):
    """
    Save persistable nodes to the given directory by the executor.

    Args:
        executor(Executor): The executor to run for saving node values.
        dirname(str): The directory path.
        graph(IrGraph): All the required persistable nodes in the graph will be saved.
    """
    persistable_node_names = set()
    persistable_nodes = []
    all_persistable_nodes = graph.all_persistable_nodes()
    for node in all_persistable_nodes:
        name = cpt.to_text(node.name())
        if name not in persistable_node_names:
            persistable_node_names.add(name)
            persistable_nodes.append(node)
    program = Program()
    var_list = []
    for node in persistable_nodes:
        var_desc = node.var()
        if var_desc.type() == core.VarDesc.VarType.RAW or \
1674
                        var_desc.type() == core.VarDesc.VarType.READER:
1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712
            continue
        var = program.global_block().create_var(
            name=var_desc.name(),
            shape=var_desc.shape(),
            dtype=var_desc.dtype(),
            type=var_desc.type(),
            lod_level=var_desc.lod_level(),
            persistable=var_desc.persistable())
        var_list.append(var)
    save_vars(executor=executor, dirname=dirname, vars=var_list)


def _load_persistable_nodes(executor, dirname, graph):
    """
    Load persistable node values from the given directory by the executor.

    Args:
        executor(Executor): The executor to run for loading node values.
        dirname(str): The directory path.
        graph(IrGraph): All the required persistable nodes in the graph will be loaded.
    """
    persistable_node_names = set()
    persistable_nodes = []
    all_persistable_nodes = graph.all_persistable_nodes()
    for node in all_persistable_nodes:
        name = cpt.to_text(node.name())
        if name not in persistable_node_names:
            persistable_node_names.add(name)
            persistable_nodes.append(node)
    program = Program()
    var_list = []

    def _exist(var):
        return os.path.exists(os.path.join(dirname, var.name))

    for node in persistable_nodes:
        var_desc = node.var()
        if var_desc.type() == core.VarDesc.VarType.RAW or \
1713
                        var_desc.type() == core.VarDesc.VarType.READER:
1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726
            continue
        var = program.global_block().create_var(
            name=var_desc.name(),
            shape=var_desc.shape(),
            dtype=var_desc.dtype(),
            type=var_desc.type(),
            lod_level=var_desc.lod_level(),
            persistable=var_desc.persistable())
        if _exist(var):
            var_list.append(var)
        else:
            _logger.warn("Cannot find the var %s!!!" % (node.name()))
    load_vars(executor=executor, dirname=dirname, vars=var_list)
H
hong 已提交
1727 1728


W
WeiXin 已提交
1729
def _unpack_saved_dict(saved_obj, protocol):
1730 1731
    temp_saved_obj = {}
    unpack_infor = {}
W
WeiXin 已提交
1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751
    # When pickle protocol=2 or protocol=3 the serialized object cannot be larger than 4G.
    if 1 < protocol < 4:
        if isinstance(saved_obj, dict):
            for key, value in saved_obj.items():
                if isinstance(value, np.ndarray):
                    MAX_NUMBER_OF_ELEMENT = int(
                        (2**30 - 1) / value.dtype.itemsize)
                    num_element = np.prod(value.shape)
                    if num_element > MAX_NUMBER_OF_ELEMENT:
                        unpack_infor[key] = {}
                        unpack_infor[key]["OriginShape"] = value.shape
                        unpack_infor[key]["slices"] = []
                        value = value.flatten()
                        for i in range(
                                int(
                                    math.ceil(num_element * 1.0 /
                                              MAX_NUMBER_OF_ELEMENT))):
                            part_name = key + "@@." + str(i)
                            unpack_infor[key]["slices"].append(part_name)
                            temp_saved_obj[part_name] = value[
1752 1753 1754
                                i *
                                MAX_NUMBER_OF_ELEMENT:MAX_NUMBER_OF_ELEMENT *
                                (i + 1)]
1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766

    if unpack_infor:
        for key, value in unpack_infor.items():
            if key in saved_obj:
                saved_obj.pop(key)
                for part in value['slices']:
                    saved_obj[part] = temp_saved_obj[part]
        saved_obj['UnpackBigParamInfor@@'] = unpack_infor
    return saved_obj


def _pack_loaded_dict(load_obj):
W
WeiXin 已提交
1767 1768 1769 1770 1771 1772
    if isinstance(load_obj, dict):
        unpack_info = 'UnpackBigParamInfor@@'
        if unpack_info in load_obj:
            removes = []
            for key, value in load_obj[unpack_info].items():
                slices = [load_obj[part] for part in value["slices"]]
1773 1774
                load_obj[key] = np.concatenate(slices).reshape(
                    value["OriginShape"])
W
WeiXin 已提交
1775 1776 1777 1778 1779
                removes += value["slices"]
            for key in removes:
                load_obj.pop(key)
            load_obj.pop(unpack_info)

1780 1781 1782
    return load_obj


1783
@static_only
1784
def _legacy_save(param_dict, model_path, protocol=2):
1785

1786
    def get_tensor(var):
J
Jiabin Yang 已提交
1787
        if isinstance(var, (core.VarBase, core.eager.Tensor)):
1788 1789 1790 1791 1792 1793 1794 1795
            return var.numpy()
        elif isinstance(var, core.LoDTensor):
            return np.array(var)
        return var

    param_dict = {name: get_tensor(param_dict[name]) for name in param_dict}

    # When value of dict is lager than 4GB ,there is a Bug on 'MAC python3'
1796 1797 1798
    if _is_file_path(
            model_path
    ) and sys.platform == 'darwin' and sys.version_info.major == 3:
1799 1800 1801 1802 1803 1804
        pickle_bytes = pickle.dumps(param_dict, protocol=protocol)
        with open(model_path, 'wb') as f:
            max_bytes = 2**30
            for i in range(0, len(pickle_bytes), max_bytes):
                f.write(pickle_bytes[i:i + max_bytes])
    else:
1805
        with _open_file_buffer(model_path, 'wb') as f:
1806 1807 1808 1809
            pickle.dump(param_dict, f, protocol=protocol)


@static_only
1810
def save(program, model_path, protocol=4, **configs):
H
hong 已提交
1811
    """
1812
    
1813
    This function save parameters, optimizer information and network description to model_path.
H
hong 已提交
1814

1815 1816
    The parameters contains all the trainable Tensor, will save to a file with suffix ".pdparams".
    The optimizer information contains all the Tensor used by optimizer. For Adam optimizer, contains beta1, beta2, momentum etc. All the information will save to a file with suffix ".pdopt". (If the optimizer have no Tensor need to save (like SGD), the fill will not generated).
H
hong 已提交
1817
    The network description is the description of the program. It's only used for deployment. The description  will save to a file with a suffix ".pdmodel".
1818

H
hong 已提交
1819 1820 1821
    Args:
        program(Program) : The program to saved.
        model_path(str): the file prefix to save the program. The format is "dirname/file_prefix". If file_prefix is empty str. A exception will be raised
1822
        protocol(int, optional): The protocol version of pickle module must be greater than 1 and less than 5.
1823
                                 Default: 4
1824
        configs(dict, optional) : optional keyword arguments.                        
H
hong 已提交
1825 1826 1827 1828 1829 1830 1831

    Returns:
        None

    Examples:
        .. code-block:: python

1832
            import paddle
1833
            import paddle.static as static
H
hong 已提交
1834

1835
            paddle.enable_static()
H
hong 已提交
1836

1837 1838 1839 1840 1841 1842 1843 1844 1845 1846
            x = static.data(name="x", shape=[10, 10], dtype='float32')
            y = static.nn.fc(x, 10)
            z = static.nn.fc(y, 10)

            place = paddle.CPUPlace()
            exe = static.Executor(place)
            exe.run(static.default_startup_program())
            prog = static.default_main_program()

            static.save(prog, "./temp")
H
hong 已提交
1847 1848 1849 1850
    """

    base_name = os.path.basename(model_path)
    assert base_name != "", \
1851
        "The input model_path MUST be format of dirname/filename [dirname\\filename in Windows system], but received model_path is empty string."
1852 1853 1854 1855 1856
    if 'pickle_protocol' in configs:
        protocol = configs['pickle_protocol']
        warnings.warn(
            "'pickle_protocol' is a deprecated argument. Please use 'protocol' instead."
        )
H
hong 已提交
1857

1858
    if not isinstance(protocol, int):
W
WeiXin 已提交
1859
        raise ValueError("The 'protocol' MUST be `int`, but received {}".format(
1860
            type(protocol)))
W
WeiXin 已提交
1861

1862
    if protocol < 2 or protocol > 4:
1863 1864 1865
        raise ValueError(
            "Expected 1<'protocol'<5, but received protocol={}".format(
                protocol))
W
WeiXin 已提交
1866

1867 1868 1869 1870
    dir_name = os.path.dirname(model_path)
    if dir_name and not os.path.exists(dir_name):
        os.makedirs(dir_name)

Y
Yang Zhang 已提交
1871 1872 1873 1874
    def get_tensor(var):
        t = global_scope().find_var(var.name).get_tensor()
        return np.array(t)

H
hong 已提交
1875
    parameter_list = list(filter(is_parameter, program.list_vars()))
Y
Yang Zhang 已提交
1876
    param_dict = {p.name: get_tensor(p) for p in parameter_list}
W
WeiXin 已提交
1877

1878
    param_dict = _unpack_saved_dict(param_dict, protocol)
1879

1880 1881 1882
    # When value of dict is lager than 4GB ,there is a Bug on 'MAC python3'
    if sys.platform == 'darwin' and sys.version_info.major == 3:
        pickle_bytes = pickle.dumps(param_dict, protocol=protocol)
1883 1884 1885 1886 1887 1888
        with open(model_path + ".pdparams", 'wb') as f:
            max_bytes = 2**30
            for i in range(0, len(pickle_bytes), max_bytes):
                f.write(pickle_bytes[i:i + max_bytes])
    else:
        with open(model_path + ".pdparams", 'wb') as f:
1889
            pickle.dump(param_dict, f, protocol=protocol)
H
hong 已提交
1890 1891 1892 1893

    optimizer_var_list = list(
        filter(is_belong_to_optimizer, program.list_vars()))

Y
Yang Zhang 已提交
1894 1895
    opt_dict = {p.name: get_tensor(p) for p in optimizer_var_list}
    with open(model_path + ".pdopt", 'wb') as f:
1896
        pickle.dump(opt_dict, f, protocol=protocol)
H
hong 已提交
1897 1898 1899 1900

    main_program = program.clone()
    program.desc.flush()
    main_program.desc._set_version()
1901
    paddle.fluid.core.save_op_version_info(program.desc)
H
hong 已提交
1902 1903 1904 1905 1906

    with open(model_path + ".pdmodel", "wb") as f:
        f.write(program.desc.serialize_to_string())


1907 1908 1909 1910 1911 1912
def _pickle_loads_mac(path, f):
    pickle_bytes = bytearray(0)
    file_size = os.path.getsize(path)
    max_bytes = 2**30
    for _ in range(0, file_size, max_bytes):
        pickle_bytes += f.read(max_bytes)
T
tianshuo78520a 已提交
1913
    load_result = pickle.loads(pickle_bytes, encoding='latin1')
1914 1915 1916
    return load_result


1917
@static_only
H
hong 已提交
1918
def load(program, model_path, executor=None, var_list=None):
H
hong 已提交
1919
    """
1920 1921
    :api_attr: Static Graph

H
hong 已提交
1922
    This function get parameters and optimizer information from program, and then get corresponding value from file.
1923
    An exception will throw if shape or dtype of the parameters is not match.
H
hong 已提交
1924

1925 1926
    This function can also load model file saved with [ save_params, save_persistables, save_vars ].
    var_list can not be None  when load single model file
H
hong 已提交
1927 1928
    ( filename is not None When save_params, save_persistables or save_vars is called ).

1929
    Args:
1930 1931
        program(Program): The program will be loaded
        model_path(str): The file prefix store the program
1932
        executor(Executor, optional): The executor used for initialize the parameter
1933
                                      When startup program is not run.
1934
        var_list(list|tuple, optional): The Tensor list/tuple to load single model file saved with
1935
                                  [ save_params, save_persistables, save_vars ].
H
hong 已提交
1936
                                  Default: None
H
hong 已提交
1937 1938 1939

    Returns:
        None
1940

H
hong 已提交
1941 1942 1943
     Examples:
        .. code-block:: python

1944
            import paddle
1945
            import paddle.static as static
H
hong 已提交
1946

1947
            paddle.enable_static()
H
hong 已提交
1948

1949 1950 1951
            x = static.data(name="x", shape=[10, 10], dtype='float32')
            y = static.nn.fc(x, 10)
            z = static.nn.fc(y, 10)
H
hong 已提交
1952

1953 1954 1955 1956 1957 1958 1959
            place = paddle.CPUPlace()
            exe = static.Executor(place)
            exe.run(static.default_startup_program())
            prog = static.default_main_program()

            static.save(prog, "./temp")
            static.load(prog, "./temp")
H
hong 已提交
1960 1961
    """

1962 1963
    assert executor is None or isinstance(executor, Executor)

H
hong 已提交
1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976
    model_prefix = model_path
    if model_prefix.endswith(".pdparams"):
        model_prefix = model_prefix[:-9]
    elif model_prefix.endswith(".pdopt"):
        model_prefix = model_prefix[:-6]
    elif model_prefix.endswith(".pdmodel"):
        model_prefix = model_prefix[:-8]

    parameter_file_name = model_prefix + ".pdparams"

    if not os.path.exists(parameter_file_name):
        # model file save by fluid.save not found, try to load model file saved with
        # [save_vars, save_params, save_persistables]
1977
        _logger.debug(
1978 1979
            "{} not found, try to load model file saved with [ save_params, save_persistables, save_vars ]"
            .format(parameter_file_name))
H
hong 已提交
1980 1981 1982 1983
        if executor is None:
            raise ValueError(
                "executor is required when loading model file saved with [ save_params, save_persistables, save_vars ]"
            )
1984 1985 1986 1987 1988 1989

        if var_list is not None:
            var_list_names = [var.name for var in var_list]
        else:
            var_list_names = None

H
hong 已提交
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
        if os.path.isdir(model_path):
            binary_file_set = set()
            for root, dirs, files in os.walk(model_path, topdown=False):
                for f in files:
                    binary_file_set.add(
                        os.path.join(root, f).replace("\\", "/"))
            program_var_list = list(program.list_vars())
            loaded_var_list = []
            for var in program_var_list:
                var_path = os.path.join(model_path, var.name).replace("\\", "/")
2000 2001
                load_condition = var_list_names is None or var.name in var_list_names
                if var_path in binary_file_set and load_condition:
H
hong 已提交
2002 2003 2004 2005 2006 2007 2008
                    loaded_var_list.append(var)
                    binary_file_set.remove(var_path)
            if len(binary_file_set) > 0:
                unused_var_list = " ".join(list(binary_file_set))
                _logger.warning("variable file [ %s ] not used" %
                                (" ".join(list(binary_file_set))))
            try:
2009 2010 2011
                load_vars(executor=executor,
                          dirname=model_path,
                          vars=loaded_var_list)
H
hong 已提交
2012 2013 2014 2015 2016
            except RuntimeError as e:
                _logger.error(e)
                raise e
            except:
                raise RuntimeError(
2017
                    "Failed to load model file, please make sure model file is saved with the "
H
hong 已提交
2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032
                    "following APIs: save_params, save_persistables, save_vars")

            return
        elif os.path.isfile(model_path):
            if var_list == None:
                raise ValueError(
                    "var_list is required when loading model file saved with [ save_params, save_persistables, save_vars ]"
                )
            program_var_list = program.list_vars()
            program_var_name_set = set([var.name for var in program_var_list])

            # check all the variable inlcuded in program
            for var in var_list:
                if var.name not in program_var_name_set:
                    raise LookupError(
2033
                        "loaded var [{}] is not in program variable list")
H
hong 已提交
2034 2035 2036

            dir_name, file_name = os.path.split(model_path)
            try:
2037 2038 2039 2040
                load_vars(executor=executor,
                          dirname=dir_name,
                          vars=var_list,
                          filename=file_name)
H
hong 已提交
2041 2042 2043 2044
            except RuntimeError as e:
                _logger.error(e)
                raise e
            except:
2045 2046 2047
                raise RuntimeError("Failed to load model file , please make sure model file is saved with the " \
                                   "the following APIs: [ save_params, save_persistables, save_vars ]. " \
                                   "When these API called, filename CANNOT be None")
H
hong 已提交
2048 2049

            return
Y
Yang Zhang 已提交
2050 2051 2052 2053 2054 2055 2056 2057

    def set_var(var, ndarray):
        t = global_scope().find_var(var.name).get_tensor()
        p = t._place()
        if p.is_cpu_place():
            place = paddle.fluid.CPUPlace()
        elif p.is_cuda_pinned_place():
            place = paddle.fluid.CUDAPinnedPlace()
2058 2059 2060 2061
        elif p.is_xpu_place():
            p = paddle.fluid.core.Place()
            p.set_place(t._place())
            place = paddle.fluid.XPUPlace(p.xpu_device_id())
2062 2063 2064 2065
        elif p.is_npu_place():
            p = paddle.fluid.core.Place()
            p.set_place(t._place())
            place = paddle.fluid.NPUPlace(p.npu_device_id())
2066 2067 2068 2069
        elif p.is_mlu_place():
            p = paddle.fluid.core.Place()
            p.set_place(t._place())
            place = paddle.fluid.MLUPlace(p.mlu_device_id())
Y
Yang Zhang 已提交
2070 2071 2072 2073 2074 2075
        else:
            p = paddle.fluid.core.Place()
            p.set_place(t._place())
            place = paddle.fluid.CUDAPlace(p.gpu_device_id())

        t.set(ndarray, place)
H
hong 已提交
2076 2077

    parameter_list = list(filter(is_parameter, program.list_vars()))
2078 2079 2080 2081 2082

    if executor:
        paddle.fluid.core._create_loaded_parameter(parameter_list,
                                                   global_scope(),
                                                   executor._default_executor)
Y
Yang Zhang 已提交
2083
    with open(parameter_file_name, 'rb') as f:
2084 2085 2086 2087 2088

        # When value of dict is lager than 4GB ,there is a Bug on 'MAC python3'
        if sys.platform == 'darwin' and sys.version_info.major == 3:
            load_dict = _pickle_loads_mac(parameter_file_name, f)
        else:
T
tianshuo78520a 已提交
2089
            load_dict = pickle.load(f, encoding='latin1')
2090
        load_dict = _pack_loaded_dict(load_dict)
Y
Yang Zhang 已提交
2091 2092 2093 2094 2095
    for v in parameter_list:
        assert v.name in load_dict, \
            "Can not find [{}] in model file [{}]".format(
                v.name, parameter_file_name)
        set_var(v, load_dict[v.name])
H
hong 已提交
2096 2097 2098 2099 2100

    optimizer_var_list = list(
        filter(is_belong_to_optimizer, program.list_vars()))

    if len(optimizer_var_list) > 0:
H
hong 已提交
2101
        opt_file_name = model_prefix + ".pdopt"
H
hong 已提交
2102
        assert os.path.exists(opt_file_name), \
T
tangwei12 已提交
2103
            "Optimizer file [{}] not exits".format(opt_file_name)
2104 2105 2106 2107

        if executor:
            paddle.fluid.core._create_loaded_parameter(
                optimizer_var_list, global_scope(), executor._default_executor)
Y
Yang Zhang 已提交
2108 2109

        with open(opt_file_name, 'rb') as f:
T
tianshuo78520a 已提交
2110
            load_dict = pickle.load(f, encoding='latin1')
Y
Yang Zhang 已提交
2111 2112 2113 2114 2115
        for v in optimizer_var_list:
            assert v.name in load_dict, \
                "Can not find [{}] in model file [{}]".format(
                    v.name, opt_file_name)
            set_var(v, load_dict[v.name])
2116 2117


H
hong 已提交
2118
def load_program_state(model_path, var_list=None):
2119
    """
2120

2121
    Load program state from local file
2122

2123 2124
    Args:
        model_path(str): The file prefix store the program
2125
        var_list(list|tuple, optional): The Tensor list/tuple to load saved with
2126
                                  [ save_params, save_persistables, save_vars ].
H
hong 已提交
2127
                                  Default: None.
2128
                                  The var_list is only used to get name,
H
hong 已提交
2129
                                  will not be modified.
2130 2131 2132 2133
    Returns:
        state_dict(dict): the dict store Parameter and optimizer information

    Examples:
2134
    
2135 2136
        .. code-block:: python

2137
            import paddle
2138
            import paddle.static as static
2139 2140

            paddle.enable_static()
2141

2142 2143 2144
            x = static.data(name="x", shape=[10, 10], dtype='float32')
            y = static.nn.fc(x, 10)
            z = static.nn.fc(y, 10)
2145

2146 2147 2148 2149
            place = paddle.CPUPlace()
            exe = static.Executor(place)
            exe.run(static.default_startup_program())
            prog = static.default_main_program()
2150

2151 2152
            static.save(prog, "./temp")
            program_state = static.load_program_state("./temp")
2153
    """
H
hong 已提交
2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165
    model_prefix = model_path
    if model_prefix.endswith(".pdparams"):
        model_prefix = model_prefix[:-9]
    elif model_prefix.endswith(".pdopt"):
        model_prefix = model_prefix[:-6]
    elif model_prefix.endswith(".pdmodel"):
        model_prefix = model_prefix[:-8]

    parameter_file_name = model_prefix + ".pdparams"
    if not os.path.exists(parameter_file_name):
        # model file saved with fluid.save is not found, try to load model file saved with
        # [save_vars, save_params, save_persistables]
2166
        _logger.debug(
2167 2168
            "{} not found, try to load model file saved with [ save_params, save_persistables, save_vars ]"
            .format(parameter_file_name))
H
hong 已提交
2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193

        var_name_list = []
        if var_list is None and os.path.isfile(model_path):
            raise ValueError(
                "var_list can not be None when model_path is a file type")

        for root, dirs, files in os.walk(model_path, topdown=False):
            for f in files:
                file_path = os.path.join(root, f)
                var_temp_name = os.path.relpath(file_path, model_path)
                var_temp_name = var_temp_name.replace("\\", "/")
                var_name_list.append(var_temp_name)

        with _load_program_scope():
            load_prog = Program()
            load_block = load_prog.global_block()

            def clone_var_to_block(block, var):
                if not isinstance(var, Variable):
                    raise TypeError("value in var_list must be variable")
                return block.create_var(
                    name=var.name,
                    shape=var.shape,
                    dtype=var.dtype,
                    type=var.type,
2194 2195
                    lod_level=var.lod_level if var.desc.type()
                    == core.VarDesc.VarType.LOD_TENSOR else None,
H
hong 已提交
2196 2197
                    persistable=True)

2198 2199 2200 2201 2202 2203
            def _load_vars_with_try_catch(exe,
                                          dirname,
                                          vars,
                                          filename,
                                          raise_error=True):
                try:
2204 2205 2206 2207
                    load_vars(executor=exe,
                              dirname=dirname,
                              vars=vars,
                              filename=filename)
2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223
                    return True
                except:
                    error_str = "Failed to load model/variables `%s`, please make sure " \
                                "model/variables file is saved with the following APIs: " \
                                "save_params, save_persistables, save_vars."
                    filenames = [var.name for var in vars
                                 ] if filename is None else filename
                    if raise_error:
                        raise RuntimeError(error_str % filenames)
                    else:
                        warnings.warn(error_str % filenames, RuntimeWarning)
                return False

            place = paddle.fluid.CPUPlace()
            exe = paddle.fluid.Executor(place)

H
hong 已提交
2224 2225
            loaded_var_list = []

2226 2227 2228
            if os.path.isfile(model_path):
                # when model_path is file, var_list cannot be None
                dir_name, file_name = os.path.split(model_path)
H
hong 已提交
2229 2230
                for var in var_list:
                    loaded_var_list.append(clone_var_to_block(load_block, var))
2231 2232
                _load_vars_with_try_catch(exe, dir_name, loaded_var_list,
                                          file_name)
H
hong 已提交
2233
            else:
2234 2235 2236 2237 2238 2239 2240
                # var_list can be None or not None
                if var_list is not None:
                    for var in var_list:
                        loaded_var_list.append(
                            clone_var_to_block(load_block, var))
                    _load_vars_with_try_catch(exe, model_path, loaded_var_list,
                                              None)
H
hong 已提交
2241
                else:
2242
                    for var_name in var_name_list:
2243 2244 2245 2246
                        # NOTE(chenweihang): If identify which files the user wants
                        # to load from the disk, we load these variables one by one.
                        # If a file does not exist, we only warn the user that the
                        # file may be an irrelevant file, but does not throw an error
2247
                        # to ensure that other legal variables can be loaded.
2248 2249
                        temp_var = load_block.create_var(name=var_name,
                                                         persistable=True)
2250 2251 2252 2253
                        if _load_vars_with_try_catch(exe, model_path,
                                                     [temp_var], None, False):
                            loaded_var_list.append(temp_var)

H
hong 已提交
2254 2255
            res_dict = {}
            for var in loaded_var_list:
2256 2257
                res_dict[var.name] = np.asarray(
                    paddle.fluid.global_scope().find_var(var.name).get_tensor())
H
hong 已提交
2258 2259 2260

            return res_dict

2261
    assert os.path.exists(parameter_file_name), \
T
tangwei12 已提交
2262
        "Parameter file [{}] not exits".format(parameter_file_name)
2263 2264

    with open(parameter_file_name, 'rb') as f:
2265 2266 2267 2268
        # When value of dict is lager than 4GB ,there is a Bug on 'MAC python3'
        if sys.platform == 'darwin' and sys.version_info.major == 3:
            para_dict = _pickle_loads_mac(parameter_file_name, f)
        else:
T
tianshuo78520a 已提交
2269
            para_dict = pickle.load(f, encoding='latin1')
2270
    para_dict = _pack_loaded_dict(para_dict)
2271

H
hong 已提交
2272
    opt_file_name = model_prefix + ".pdopt"
2273 2274
    if os.path.exists(opt_file_name):
        with open(opt_file_name, 'rb') as f:
T
tianshuo78520a 已提交
2275
            opti_dict = pickle.load(f, encoding='latin1')
2276 2277 2278 2279 2280 2281

        para_dict.update(opti_dict)

    return para_dict


2282
@static_only
2283 2284 2285 2286
def set_program_state(program, state_dict):
    """
    Set program parameter from state_dict

2287
    An exception will throw if shape or dtype of the parameters is not match.
2288 2289 2290 2291 2292 2293

    NOTICE: This function MUST called after run start_up_program

    Args:
        program(Program): The program to be set
        state_dict(dict): the dict store Parameter and optimizer information
2294
    Returns:
2295
        None
2296

2297 2298
    Examples:
        .. code-block:: python
2299

2300
            import paddle
2301
            import paddle.static as static
2302 2303

            paddle.enable_static()
2304

2305 2306 2307
            x = static.data(name="x", shape=[10, 10], dtype='float32')
            y = static.nn.fc(x, 10)
            z = static.nn.fc(y, 10)
2308

2309 2310 2311 2312
            place = paddle.CPUPlace()
            exe = static.Executor(place)
            exe.run(static.default_startup_program())
            prog = static.default_main_program()
2313

2314 2315
            static.save(prog, "./temp")
            program_state = static.load_program_state("./temp")
H
hong 已提交
2316

2317
            static.set_program_state(prog, program_state)
2318
    """
2319
    state_dict = _pack_loaded_dict(state_dict)
2320 2321 2322 2323 2324 2325
    parameter_list = list(filter(is_persistable, program.list_vars()))

    used_para_list = {}
    for para in parameter_list:
        var_temp = paddle.fluid.global_scope().find_var(para.name)
        assert var_temp != None, \
T
tangwei12 已提交
2326
            "Variable [ {} ] Not found, Please make sure run startup program".format(para.name)
2327 2328 2329 2330
        if para.name in state_dict:
            # set value from state dict
            orig_para_np = np.array(var_temp.get_tensor())
            new_para_np = state_dict[para.name]
T
tangwei12 已提交
2331
            assert orig_para_np.shape == new_para_np.shape, \
2332
                "Parameter's shape does not match, the Program requires a parameter with the shape of ({}), " \
T
tangwei12 已提交
2333
                "while the loaded parameter (namely [ {} ]) has a shape of  ({})." \
2334
                    .format(orig_para_np.shape, para.name, new_para_np.shape)
T
tangwei12 已提交
2335
            assert orig_para_np.dtype == new_para_np.dtype, \
2336
                "Parameter's data type does not match, the Program requires a parameter with a dtype of ({}), " \
T
tangwei12 已提交
2337
                "while the loaded parameter (namely [ {} ]) has a dtype of  ({})." \
2338 2339 2340 2341 2342
                    .format(orig_para_np.dtype, para.name, new_para_np.dtype)

            ten = var_temp.get_tensor()
            ten_place = ten._place()

Q
QingshuChen 已提交
2343 2344
            #assert ten_place.is_gpu_place() or ten_place.is_cpu_place(), \
            #    "Place not support, only support CPUPlace and GPUPlace, now is {}".format(str(ten_place))
2345 2346 2347 2348 2349 2350 2351
            py_place = paddle.fluid.CPUPlace()
            if ten_place.is_cuda_pinned_place():
                place = paddle.fluid.CUDAPinnedPlace()
            elif ten_place.is_gpu_place():
                p = paddle.fluid.core.Place()
                p.set_place(ten_place)
                py_place = paddle.fluid.CUDAPlace(p.gpu_device_id())
Q
QingshuChen 已提交
2352 2353 2354 2355
            elif ten_place.is_xpu_place():
                p = paddle.fluid.core.Place()
                p.set_place(ten_place)
                py_place = paddle.fluid.XPUPlace(p.xpu_device_id())
2356 2357 2358 2359
            elif ten_place.is_npu_place():
                p = paddle.fluid.core.Place()
                p.set_place(ten_place)
                py_place = paddle.fluid.NPUPlace(p.npu_device_id())
2360 2361 2362 2363
            elif ten_place.is_mlu_place():
                p = paddle.fluid.core.Place()
                p.set_place(ten_place)
                py_place = paddle.fluid.MLUPlace(p.mlu_device_id())
2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374

            ten.set(new_para_np, py_place)

            used_para_list[para.name] = 1

    unused_para_list = []
    for k, v in state_dict.items():
        if k not in used_para_list:
            unused_para_list.append(k)
    if len(unused_para_list) > 0:
        warnings.warn(
2375 2376
            "This list is not set, Because of Paramerter not found in program. There are: {}"
            .format(" ".join(unused_para_list)))