jit.py 56.3 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14
# Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

15 16
from __future__ import print_function

17 18
import os
import pickle
19
import warnings
20
import functools
21
from collections import OrderedDict
22
import inspect
23 24

import six
25
import paddle
26
from paddle.fluid import core
27 28
from paddle.fluid.compiler import BuildStrategy, CompiledProgram, ExecutionStrategy
from paddle.fluid.data_feeder import check_type
29
from paddle.fluid.layers.utils import flatten, pack_sequence_as
30
from paddle.fluid.dygraph.base import program_desc_tracing_guard, switch_to_static_graph
31
from paddle.fluid.dygraph.dygraph_to_static import logging_utils
32
from paddle.fluid.dygraph.dygraph_to_static.convert_call_func import ConversionOptions, CONVERSION_OPTIONS
33
from paddle.fluid.dygraph.dygraph_to_static.logging_utils import set_code_level, set_verbosity
34
from paddle.fluid.dygraph.dygraph_to_static.program_translator import ProgramTranslator, StaticFunction, unwrap_decorators
35
from paddle.fluid.dygraph.io import TranslatedLayer, INFER_MODEL_SUFFIX, INFER_PARAMS_SUFFIX, INFER_PARAMS_INFO_SUFFIX
36 37
from paddle.fluid.dygraph.layers import Layer
from paddle.fluid.executor import Executor, scope_guard
38
from paddle.fluid.framework import Block, ParamBase, Program, Variable, Parameter
39 40
from paddle.fluid.framework import _current_expected_place, _dygraph_guard, _dygraph_tracer
from paddle.fluid.framework import dygraph_only, in_dygraph_mode
41
from paddle.fluid.wrapped_decorator import wrap_decorator
42

43 44
__all__ = [
    'TracedLayer', 'declarative', 'dygraph_to_static_func', 'set_code_level',
45
    'set_verbosity', 'save', 'load', 'not_to_static'
46
]
47 48 49 50 51 52 53 54 55 56


def create_program_from_desc(program_desc):
    program = Program()
    program.desc = program_desc
    program.blocks = [Block(program, 0)]
    program._sync_with_cpp()
    return program


57
def _extract_vars(inputs, result_list, err_tag='inputs'):
58
    if isinstance(inputs, Variable):
59
        result_list.append(inputs)
60
    elif isinstance(inputs, (list, tuple)):
61
        for var in inputs:
62
            _extract_vars(var, result_list, err_tag)
63 64
    else:
        raise TypeError(
65 66
            "The type of 'each element of {}' in fluid.dygraph.jit.TracedLayer.trace must be fluid.Variable, but received {}.".
            format(err_tag, type(inputs)))
67 68


69
def extract_vars(inputs, err_tag='inputs'):
70
    result_list = []
71
    _extract_vars(inputs, result_list, err_tag)
72 73 74
    return result_list


75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123
def _dygraph_to_static_func_(dygraph_func):
    """
    Converts imperative dygraph APIs into declarative function APIs. Decorator
    @dygraph_to_static_func only converts imperative dygraph APIs into
    declarative net-building APIs, which means it doesn't return immediate
    digital result as imperative mode. Users should handle Program and Executor
    by themselves.

    Note:
    This decorator is NOT our recommended way to transform imperative function
    to declarative function. We will remove this decorator after we finalize
    cleaning up code.

    Args:
        dygraph_func (callable): callable imperative function.

    Returns:
        Callable: converting imperative dygraph APIs into declarative
        net-building APIs.

    Examples:
        .. code-block:: python

          import paddle.fluid as fluid
          import numpy as np
          from paddle.fluid.dygraph.jit import dygraph_to_static_func

          @dygraph_to_static_func
          def func(x):
              if fluid.layers.mean(x) < 0:
                  x_v = x - 1
              else:
                  x_v = x + 1

               return x_v

          x = fluid.layers.fill_constant(shape=[3, 3], value=0, dtype='float64')

          x_v = func(x)
          exe = fluid.Executor(fluid.CPUPlace())
          out = exe.run(fetch_list=[x_v])
          print(out[0])
          # [[1. 1. 1.]
          #  [1. 1. 1.]
          #  [1. 1. 1.]]

    """

    # TODO: remove this decorator after we finalize training API
124 125
    def __impl__(*args, **kwargs):
        program_translator = ProgramTranslator()
126
        if in_dygraph_mode() or not program_translator.enable_to_static:
127
            logging_utils.warn(
128
                "The decorator 'dygraph_to_static_func' doesn't work in "
129
                "dygraph mode or set ProgramTranslator.enable to False. "
130 131 132 133
                "We will just return dygraph output.")
            return dygraph_func(*args, **kwargs)
        static_func = program_translator.get_func(dygraph_func)
        return static_func(*args, **kwargs)
134 135 136 137

    return __impl__


138
dygraph_to_static_func = wrap_decorator(_dygraph_to_static_func_)
139

140

141 142 143 144 145 146
def copy_decorator_attrs(original_func, decorated_obj):
    """
    Copies some necessary attributes from original function into decorated function.

    Args:
        original_func(callable): the original decorated function.
147
        decorated_obj(StaticFunction): the target decorated StaticFunction object.
148 149 150 151 152 153 154 155 156 157 158 159 160
    """
    decorator_name = "declarative"

    decorated_obj.__name__ = original_func.__name__
    decorated_obj._decorator_name = decorator_name
    decorated_obj.__wrapped__ = original_func
    decorated_obj.__doc__ = original_func.__doc__
    if hasattr(original_func, "__module__"):
        decorated_obj.__module__ = original_func.__module__

    return decorated_obj


161
def declarative(function=None, input_spec=None, build_strategy=None):
162 163 164
    """
    Converts imperative dygraph APIs into declarative function APIs. Decorator
    @declarative handles the Program and Executor of static mode and returns
165 166 167 168
    the result as dygraph Tensor(s). Users could use the returned dygraph
    Tensor(s) to do imperative training, inference, or other operations. If the
    decorated function calls other imperative function, the called one will be
    converted into declarative function as well.
169

170
    Args:
171
        function (callable): callable imperative function.
172
        input_spec(list[InputSpec]|tuple[InputSpec]): list/tuple of InputSpec to specific the shape/dtype/name
173
            information of each input Tensor.
174 175 176 177 178 179
        build_strategy(BuildStrategy|None): This argument is used to compile the
            converted program with the specified options, such as operators' fusion
            in the computational graph and memory optimization during the execution
            of the computational graph. For more information about build_strategy,
            please refer to :code:`paddle.static.BuildStrategy`. The default is None.

180

181
    Returns:
182
        Tensor(s): containing the numerical result.
183

184 185
    Examples:
        .. code-block:: python
186

187 188 189 190 191 192 193 194 195 196 197 198 199 200
            import paddle
            from paddle.jit import to_static

            @to_static
            def func(x):
                if paddle.mean(x) < 0:
                    x_v = x - 1
                else:
                    x_v = x + 1
                return x_v

            x = paddle.ones([1, 2], dtype='float32')
            x_v = func(x)
            print(x_v) # [[2. 2.]]
201

202
    """
203

204 205
    def decorated(python_func):
        """
206
        Decorates a python function into a StaticFunction object.
207 208 209
        """
        # Step 1. unwrap the function if it is already decorated.
        _, python_func = unwrap_decorators(python_func)
210

211 212 213
        # Step 2. copy some attributes from original python function.
        static_layer = copy_decorator_attrs(
            original_func=python_func,
214
            decorated_obj=StaticFunction(
215 216 217
                function=python_func,
                input_spec=input_spec,
                build_strategy=build_strategy))
218 219

        return static_layer
220

221 222 223 224 225 226
    build_strategy = build_strategy or BuildStrategy()
    if not isinstance(build_strategy, BuildStrategy):
        raise TypeError(
            "Required type(build_strategy) shall be `paddle.static.BuildStrategy`, but received {}".
            format(type(build_strategy).__name__))

227 228
    # for usage: `declarative(foo, ...)`
    if function is not None:
229
        if isinstance(function, Layer):
230
            if isinstance(function.forward, StaticFunction):
231
                class_name = function.__class__.__name__
232
                logging_utils.warn(
233 234 235 236 237 238
                    "`{}.forward` has already been decorated somewhere. It will be redecorated to replace previous one.".
                    format(class_name))
            function.forward = decorated(function.forward)
            return function
        else:
            return decorated(function)
239

240 241
    # for usage: `@declarative`
    return decorated
242 243


244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283
def not_to_static(func=None):
    """
    A Decorator to suppresses the convertion of a function.

    Args:
        func(callable): The function to decorate.

    Returns:
        callable: A function which won't be converted in Dynamic-to-Static.

    Examples:
        .. code-block:: python

            import paddle

            @paddle.jit.not_to_static
            def func_not_to_static(x):
                res = x - 1
                return res

            @paddle.jit.to_static
            def func(x):
                if paddle.mean(x) < 0:
                    out = func_not_to_static(x)
                else:
                    out = x + 1
                return out

            x = paddle.ones([1, 2], dtype='float32')
            out = func(x)
            print(out) # [[2. 2.]]
    """
    if func is None:
        return not_to_static

    options = ConversionOptions(not_convert=True)
    setattr(func, CONVERSION_OPTIONS, options)
    return func


284
class _SaveLoadConfig(object):
285 286 287 288 289
    def __init__(self):
        self._output_spec = None
        self._model_filename = None
        self._params_filename = None
        self._separate_params = False
290 291
        # used for `paddle.load`
        self._keep_name_table = False
292 293 294 295

        # NOTE: Users rarely use following configs, so these configs are not open to users,
        # reducing user learning costs, but we retain the configuration capabilities

296 297
        # If True, programs are modified to only support direct inference deployment.
        # Otherwise,more information will be stored for flexible optimization and re-training.
298 299 300 301 302 303 304 305 306 307 308 309
        # Currently, only True is supported
        self._export_for_deployment = True

        # If True, It will save inference program only, and do not save params of Program
        self._program_only = False

    @property
    def output_spec(self):
        return self._output_spec

    @output_spec.setter
    def output_spec(self, spec):
310 311
        if spec is None:
            return
312 313
        if not isinstance(spec, list):
            raise TypeError(
314
                "The config `output_spec` should be 'list', but received input type is %s."
315 316 317 318
                % type(input))
            for var in spec:
                if not isinstance(var, core.VarBase):
                    raise TypeError(
319
                        "The element in config `output_spec` list should be 'Variable', but received element's type is %s."
320 321 322 323 324 325 326 327 328
                        % type(var))
        self._output_spec = spec

    @property
    def model_filename(self):
        return self._model_filename

    @model_filename.setter
    def model_filename(self, filename):
329 330
        if filename is None:
            return
331 332
        if not isinstance(filename, six.string_types):
            raise TypeError(
333
                "The config `model_filename` should be str, but received input's type is %s."
334 335
                % type(filename))
        if len(filename) == 0:
336
            raise ValueError("The config `model_filename` is empty string.")
337 338 339 340 341 342 343 344
        self._model_filename = filename

    @property
    def params_filename(self):
        return self._params_filename

    @params_filename.setter
    def params_filename(self, filename):
345 346
        if filename is None:
            return
347 348
        if not isinstance(filename, six.string_types):
            raise TypeError(
349
                "The config `params_filename` should be str, but received input's type is %s."
350 351
                % type(filename))
        if len(filename) == 0:
352
            raise ValueError("The config `params_filename` is empty string.")
353 354
        self._params_filename = filename

355 356 357 358 359 360
    @property
    def keep_name_table(self):
        return self._keep_name_table

    @keep_name_table.setter
    def keep_name_table(self, value):
361 362
        if value is None:
            return
363 364
        if not isinstance(value, bool):
            raise TypeError(
365
                "The config `keep_name_table` should be bool value, but received input's type is %s."
366 367 368
                % type(value))
        self._keep_name_table = value

369

370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404
def _parse_save_configs(configs):
    supported_configs = ['output_spec']

    # input check
    for key in configs:
        if key not in supported_configs:
            raise ValueError(
                "The additional config (%s) of `paddle.jit.save` is not supported."
                % (key))

    # construct inner config
    inner_config = _SaveLoadConfig()
    inner_config.output_spec = configs.get('output_spec', None)

    return inner_config


def _parse_load_config(configs):
    supported_configs = ['model_filename', 'params_filename']

    # input check
    for key in configs:
        if key not in supported_configs:
            raise ValueError(
                "The additional config (%s) of `paddle.jit.load` is not supported."
                % (key))

    # construct inner config
    inner_config = _SaveLoadConfig()
    inner_config.model_filename = configs.get('model_filename', None)
    inner_config.params_filename = configs.get('params_filename', None)

    return inner_config


405 406 407 408 409 410 411 412 413 414
def _get_input_var_names(inputs, input_spec):
    name_none_error = "The %s's name is None. " \
        "When using jit.save, please set InputSepc's name in " \
        "to_static(input_spec=[]) and jit.save(input_spec=[]) " \
        "and make sure they are consistent."
    name_no_exists_error = "The tensor `%s` does not exists. " \
        "Please make sure the name of InputSpec or example Tensor " \
        "in input_spec is the same as the name of InputSpec in " \
        "`to_static` decorated on the Layer.forward method."
    result_list = []
415 416 417
    input_var_names = [
        var.name for var in flatten(inputs) if isinstance(var, Variable)
    ]
418 419
    if input_spec is None:
        # no prune
420 421 422 423 424 425 426 427 428
        return input_var_names
    else:
        # fileter out non-tensor type spec infos.
        input_spec = [
            spec for spec in input_spec
            if isinstance(spec, paddle.static.InputSpec)
        ]

    if len(input_spec) == len(input_var_names):
429 430
        # no prune
        result_list = input_var_names
431
        # if input spec name not in input_var_names, only raise warning
432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461
        for spec in input_spec:
            if spec.name is None:
                warnings.warn(name_none_error % spec)
            elif spec.name not in input_var_names:
                warnings.warn(name_no_exists_error % spec.name)
            else:
                # do nothing
                pass
    else:
        # prune
        for spec in input_spec:
            if spec.name is None:
                # name is None, the input_spec only can be InputSpec
                raise ValueError(name_none_error % spec)
            elif spec.name not in input_var_names:
                # the input_spec can be `InputSpec` or `VarBase`
                raise ValueError(name_no_exists_error % spec.name)
            else:
                result_list.append(spec.name)

    return result_list


def _get_output_vars(outputs, output_spec):
    name_no_exists_error = "The tensor `%s` does not exists. " \
        "Please make sure the name of example Tensor " \
        "in configs.output_spec is the output tensor of " \
        "Layer.forward method."
    result_list = []
    output_vars_dict = OrderedDict()
462
    for var in flatten(outputs):
463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480
        if isinstance(var, Variable):
            output_vars_dict[var.name] = var
    if output_spec is None:
        result_list = output_vars_dict.values()
    elif output_spec is not None and len(output_spec) == len(output_vars_dict):
        result_list = output_vars_dict.values()
        for var in output_spec:
            if var.name not in output_vars_dict:
                warnings.warn(name_no_exists_error % var.name)
    else:
        for var in output_spec:
            if var.name not in output_vars_dict:
                raise ValueError(name_no_exists_error % var.name)
            else:
                result_list.append(output_vars_dict[var.name])
    return result_list


481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523
# NOTE(chenweihang): [ Handling of use cases of API paddle.jit.load ]
# `paddle.jit.load` may be used to load saved results of:
# 1. Expected cases:
#   - paddle.jit.save
#   - paddle.static.save_inference_model
#   - paddle.fluid.io.save_inference_model
# 2. Error cases:
#   - paddle.save: no .pdmodel for prefix
#   - paddle.static.save: no .pdiparams but .pdparams exists
#   - paddle.fluid.io.save_params/save_persistables: no __model__
# TODO(chenweihang): polish error message in above error cases
def _build_load_path_and_config(path, config):
    # NOTE(chenweihang): If both [prefix save format] and [directory save format] exist,
    # raise error, avoid confusing behavior
    prefix_format_path = path + INFER_MODEL_SUFFIX
    prefix_format_exist = os.path.exists(prefix_format_path)
    directory_format_exist = os.path.isdir(path)
    if prefix_format_exist and directory_format_exist:
        raise ValueError(
            "The %s.pdmodel and %s directory exist at the same time, "
            "don't know which one to load, please make sure that the specified target "
            "of ``path`` is unique." % (path, path))
    elif not prefix_format_exist and not directory_format_exist:
        raise ValueError("The ``path`` (%s) to load model not exists." % path)
    else:
        if prefix_format_exist:
            file_prefix = os.path.basename(path)
            model_path = os.path.dirname(path)
            if config.model_filename is not None:
                warnings.warn(
                    "When loading the result saved with the "
                    "specified file prefix, the ``model_filename`` config does "
                    "not take effect.")
            config.model_filename = file_prefix + INFER_MODEL_SUFFIX
            if config.params_filename is not None:
                warnings.warn(
                    "When loading the result saved with the "
                    "specified file prefix, the ``params_filename`` config does "
                    "not take effect.")
            config.params_filename = file_prefix + INFER_PARAMS_SUFFIX
        else:
            # Compatible with the old save_inference_model format
            model_path = path
524

525
    return model_path, config
526 527


528
@switch_to_static_graph
529
def save(layer, path, input_spec=None, **configs):
530
    """
531
    Saves input Layer or function as ``paddle.jit.TranslatedLayer``
532 533
    format model, which can be used for inference or fine-tuning after loading.

534
    It will save the translated program and all related persistable
535
    variables of input Layer to given ``path`` .
536 537

    ``path`` is the prefix of saved objects, and the saved translated program file
538
    suffix is ``.pdmodel`` , the saved persistable variables file suffix is ``.pdiparams`` ,
539
    and here also saved some additional variable description information to a file,
540
    its suffix is ``.pdiparams.info``, these additional information is used in fine-tuning.
541 542

    The saved model can be loaded by follow APIs:
543 544
      - ``paddle.jit.load``
      - ``paddle.static.load_inference_model``
545 546
      - Other C++ inference APIs

547 548 549 550
    .. note::
        When using ``paddle.jit.save`` to save a function, parameters will not be saved. If you have to 
        save the parameter, please pass the Layer containing function and parameter to ``paddle.jit.save``.

551
    Args:
552
        layer (Layer|function): The Layer or function to be saved.
553
        path (str): The path prefix to save model. The format is ``dirname/file_prefix`` or ``file_prefix``.
554 555 556
        input_spec (list or tuple[InputSpec|Tensor|Python built-in variable], optional): Describes the input of the saved model's forward
            method, which can be described by InputSpec or example Tensor. Moreover, we support to specify non-tensor type argument,
            such as int, float, string, or list/dict of them.If None, all input variables of
557
            the original Layer's forward method would be the inputs of the saved model. Default None.
558 559
        **configs (dict, optional): Other save configuration options for compatibility. We do not
            recommend using these configurations, they may be removed in the future. If not necessary,
560 561 562
            DO NOT use them. Default None.
            The following options are currently supported:
            (1) output_spec (list[Tensor]): Selects the output targets of the saved model.
563 564 565
            By default, all return variables of original Layer's forward method are kept as the
            output of the saved model. If the provided ``output_spec`` list is not all output variables,
            the saved model will be pruned according to the given ``output_spec`` list.
566

567 568 569 570 571 572
    Returns:
        None

    Examples:
        .. code-block:: python

573
            # example 1: save layer
574
            import numpy as np
575 576 577
            import paddle
            import paddle.nn as nn
            import paddle.optimizer as opt
578

579 580 581
            BATCH_SIZE = 16
            BATCH_NUM = 4
            EPOCH_NUM = 4
582

583 584 585 586 587 588 589
            IMAGE_SIZE = 784
            CLASS_NUM = 10

            # define a random dataset
            class RandomDataset(paddle.io.Dataset):
                def __init__(self, num_samples):
                    self.num_samples = num_samples
590

591 592 593 594
                def __getitem__(self, idx):
                    image = np.random.random([IMAGE_SIZE]).astype('float32')
                    label = np.random.randint(0, CLASS_NUM - 1, (1, )).astype('int64')
                    return image, label
595

596 597
                def __len__(self):
                    return self.num_samples
598

599 600
            class LinearNet(nn.Layer):
                def __init__(self):
601
                    super(LinearNet, self).__init__()
602
                    self._linear = nn.Linear(IMAGE_SIZE, CLASS_NUM)
603

604
                @paddle.jit.to_static
605 606 607
                def forward(self, x):
                    return self._linear(x)

608 609 610 611 612 613 614 615 616 617 618 619
            def train(layer, loader, loss_fn, opt):
                for epoch_id in range(EPOCH_NUM):
                    for batch_id, (image, label) in enumerate(loader()):
                        out = layer(image)
                        loss = loss_fn(out, label)
                        loss.backward()
                        opt.step()
                        opt.clear_grad()
                        print("Epoch {} batch {}: loss = {}".format(
                            epoch_id, batch_id, np.mean(loss.numpy())))

            # 1. train & save model.
620

621 622 623 624
            # create network
            layer = LinearNet()
            loss_fn = nn.CrossEntropyLoss()
            adam = opt.Adam(learning_rate=0.001, parameters=layer.parameters())
625

626 627 628 629 630 631 632
            # create data loader
            dataset = RandomDataset(BATCH_NUM * BATCH_SIZE)
            loader = paddle.io.DataLoader(dataset,
                batch_size=BATCH_SIZE,
                shuffle=True,
                drop_last=True,
                num_workers=2)
633

634 635
            # train
            train(layer, loader, loss_fn, adam)
636

637
            # save
638 639
            path = "example_model/linear"
            paddle.jit.save(layer, path)
640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661

            # example 2: save function
            import paddle
            from paddle.static import InputSpec


            def save_function():
                @paddle.jit.to_static
                def fun(inputs):
                    return paddle.tanh(inputs)

                path = 'test_jit_save_load_function_1/func'
                inps = paddle.rand([3, 6])
                origin = fun(inps)

                paddle.jit.save(fun, path)
                load_func = paddle.jit.load(path)

                load_result = load_func(inps)
                print((load_result - origin).abs().max() < 1e-10)
                
            save_function()
662 663
    """

664
    # 1. input build & check
665
    prog_translator = ProgramTranslator()
666
    if not prog_translator.enable_to_static:
667
        raise RuntimeError(
668
            "The paddle.jit.save doesn't work when setting ProgramTranslator.enable to False."
669
        )
670 671 672

    if not (isinstance(layer, Layer) or inspect.isfunction(layer) or isinstance(
            layer, StaticFunction)):
673
        raise TypeError(
674
            "The input of paddle.jit.save should be 'Layer' or 'Function', but received input type is %s."
675
            % type(layer))
676 677 678 679
    elif inspect.isfunction(layer) or isinstance(layer, StaticFunction):
        warnings.warn(
            'What you save is a function, and `jit.save` will generate the name of the model file according to `path` you specify. When loading these files with `jit.load`, you get a `TranslatedLayer` whose inference result is the same as the inference result of the function you saved.'
        )
680

681 682
    # NOTE(chenweihang): If the input layer be wrapped by DataParallel,
    # the args and kwargs of forward method will can't be parsed by
683
    # function_spec, so here we save DataParallel._layers instead
684 685 686 687 688 689 690
    # DataParallel it self
    # NOTE(chenweihang): using inner_layer, do not change input layer
    if isinstance(layer, paddle.DataParallel):
        inner_layer = layer._layers
    else:
        inner_layer = layer

691 692 693 694 695 696 697 698 699 700 701
    # path check
    file_prefix = os.path.basename(path)
    if file_prefix == "":
        raise ValueError(
            "The input path MUST be format of dirname/file_prefix "
            "[dirname\\file_prefix in Windows system], but received "
            "file_prefix is empty string.")

    dirname = os.path.dirname(path)
    if dirname and not os.path.exists(dirname):
        os.makedirs(dirname)
702

703 704
    # avoid change user given input_spec
    inner_input_spec = None
705
    if input_spec is not None:
706 707 708 709 710 711 712 713 714
        if isinstance(layer, Layer):
            for attr_func in dir(inner_layer):
                static_func = getattr(inner_layer, attr_func, None)
                if isinstance(static_func,
                              StaticFunction) and 'forward' != attr_func:
                    raise ValueError(
                        "If there are static functions other than 'forward' that need to be saved, the input 'input_spec' should be None, but received the type of 'input_spec' is %s."
                        % type(input_spec))

715
        if not isinstance(input_spec, (list, tuple)):
716 717 718
            raise TypeError(
                "The input input_spec should be 'list', but received input_spec's type is %s."
                % type(input_spec))
719
        inner_input_spec = []
720
        for var in flatten(input_spec):
721 722 723 724 725 726
            if isinstance(var, paddle.static.InputSpec):
                inner_input_spec.append(var)
            elif isinstance(var, (core.VarBase, Variable)):
                inner_input_spec.append(
                    paddle.static.InputSpec.from_tensor(var))
            else:
727 728
                # NOTE(Aurelius84): Support non-Tensor type in `input_spec`.
                inner_input_spec.append(var)
729

730 731
    # parse configs
    configs = _parse_save_configs(configs)
732 733
    scope = core.Scope()
    extra_var_info = dict()
734 735 736 737 738 739 740 741 742 743 744 745 746
    if isinstance(layer, Layer):
        functions = dir(inner_layer)
    else:
        # layer is function
        functions = [layer, ]
    for attr_func in functions:
        if isinstance(layer, Layer):
            static_func = getattr(inner_layer, attr_func, None)
            if isinstance(static_func, StaticFunction):
                concrete_program = static_func.concrete_program_specify_input_spec(
                    inner_input_spec)
            elif 'forward' == attr_func:
                # transform in jit.save, if input_spec is incomplete, declarative will throw error
747
                # inner_input_spec is list[InputSpec], it should be packed with same structure
748 749 750 751 752 753 754 755 756 757 758 759 760 761
                # as original input_spec here.
                if inner_input_spec:
                    inner_input_spec = pack_sequence_as(input_spec,
                                                        inner_input_spec)
                static_forward = declarative(
                    inner_layer.forward, input_spec=inner_input_spec)
                concrete_program = static_forward.concrete_program
                # the input_spec has been used in declarative, which is equal to
                # @declarative with input_spec and jit.save without input_spec,
                # avoid needless warning
                inner_input_spec = None
            else:
                continue

762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787
        else:
            # When layer is a function
            if isinstance(attr_func, StaticFunction):
                concrete_program = attr_func.concrete_program_specify_input_spec(
                    inner_input_spec)
            else:
                if inner_input_spec:
                    inner_input_spec = pack_sequence_as(input_spec,
                                                        inner_input_spec)
                static_function = declarative(
                    attr_func, input_spec=inner_input_spec)
                concrete_program = static_function.concrete_program

                if static_function._class_instance is None:
                    warnings.warn(
                        '`jit.save` will only save the `Program`, not the parameters. If you have to save the parameters, please make sure that {} is a member function of `paddle.nn.Layer` and the saved parameters are in `state_dict`'.
                        format(layer))

        dygraph_state_dict = None
        if isinstance(inner_layer, Layer):
            dygraph_state_dict = inner_layer.state_dict()
        elif isinstance(attr_func, StaticFunction):
            if attr_func._class_instance:
                dygraph_state_dict = attr_func._class_instance.state_dict()

        if dygraph_state_dict:
788 789 790 791 792
            # NOTE(chenweihang): we maintain the mapping of variable name to
            # structured name, the buffer variable (non-persistable)
            # saved to inference program may not need by dygraph Layer,
            # we only record the state_dict variable's structured name
            state_names_dict = dict()
793
            for structured_name, var in six.iteritems(dygraph_state_dict):
794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815
                state_names_dict[var.name] = structured_name

            # 3. share parameters from Layer to scope & record var info
            for param_or_buffer in concrete_program.parameters:
                # share to scope
                param_or_buffer_tensor = scope.var(
                    param_or_buffer.name).get_tensor()
                src_tensor = param_or_buffer.value().get_tensor()
                param_or_buffer_tensor._share_data_with(src_tensor)
                # record var info
                if param_or_buffer.name not in extra_var_info:
                    extra_info_dict = dict()
                    if param_or_buffer.name in state_names_dict:
                        extra_info_dict['structured_name'] = state_names_dict[
                            param_or_buffer.name]
                    extra_info_dict[
                        'stop_gradient'] = param_or_buffer.stop_gradient
                    if isinstance(param_or_buffer, ParamBase):
                        extra_info_dict['trainable'] = param_or_buffer.trainable
                    extra_var_info[param_or_buffer.name] = extra_info_dict

        # 4. build input & output of save_infernece_model
816 817 818 819 820 821 822 823 824 825 826 827
        # NOTE(chenweihang): [ Get input variables name ]
        # There are two cases, whether to prune the inputs or not
        # - not prune inputs (recommend):
        #   - the len(input_spec) == len((concrete_program.inputs) - 1
        #   - here can use concrete_program.inputs directly
        # - prune inputs:
        #   - the input_spec length < len((concrete_program.inputs) - 1
        #   - the input_spec's name should be in concrete_program.inputs
        input_var_names = _get_input_var_names(concrete_program.inputs,
                                               inner_input_spec)

        # NOTE(chenweihang): [ Get output variables ]
828 829
        # the rule is like [ Get input variables name ]. For output var,
        # we only support VarBase spec, and actually, we only need the
830 831 832 833 834 835 836 837 838 839
        # var name of output, and we don't recommended to use output_spec
        output_vars = _get_output_vars(concrete_program.outputs,
                                       configs.output_spec)

        # 5. save inference model
        from paddle.fluid.io import save_inference_model

        # construct new save_inference_model arguments
        model_path = dirname
        # NOTE(chenweihang): because prefix contains model and params filename,
840
        # so we don't support set model_filename & params_filename
841
        if 'forward' == attr_func or not isinstance(layer, Layer):
842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857
            model_filename = file_prefix + INFER_MODEL_SUFFIX
            params_filename = file_prefix + INFER_PARAMS_SUFFIX
        else:
            model_filename = file_prefix + '.' + attr_func + INFER_MODEL_SUFFIX
            params_filename = file_prefix + '.' + attr_func + INFER_PARAMS_SUFFIX

        with scope_guard(scope):
            save_inference_model(
                dirname=model_path,
                feeded_var_names=input_var_names,
                target_vars=output_vars,
                executor=Executor(_current_expected_place()),
                main_program=concrete_program.main_program.clone(),
                model_filename=model_filename,
                params_filename=params_filename,
                export_for_deployment=configs._export_for_deployment,
858 859
                program_only=configs._program_only,
                clip_extra=False)
860 861 862 863 864 865 866 867

    # NOTE(chenweihang): [ Save extra variable info ]
    # save_inference_model will lose some important variable information, including:
    #   - Variable name and correspondence (when saved variables as one file)
    #   - Variable.stop_gradient information
    #   - Which persistent variable are parameter and which are not
    #   - Parameter.trainable information
    #
868 869
    # The lost information cannot be recovered when it is loaded again,
    # so if we want to perform fine-tune after loading, we may need to
870 871
    # configure redundant information to proceed.
    #
872 873
    # Due to compatibility issues, we cannot change the original storage structure,
    # but we can save these information in `jit.save` without changing the original
874 875
    # storage to improve user experience. So we save extra information into
    # file `***.pdiparams.info`
876 877 878 879 880 881 882 883

    # "layer" can only be Layer or function or StaticFunction.

    contain_parameter = False
    for var in concrete_program.main_program.list_vars():
        contain_parameter |= isinstance(var, Parameter)

    if (isinstance(layer, Layer) or contain_parameter) and extra_var_info:
884 885 886 887
        with scope_guard(scope):
            extra_var_info_path = path + INFER_PARAMS_INFO_SUFFIX
            with open(extra_var_info_path, 'wb') as f:
                pickle.dump(extra_var_info, f, protocol=2)
888 889 890


@dygraph_only
891
def load(path, **configs):
892 893 894
    """
    :api_attr: imperative

895 896
    Load model saved by ``paddle.jit.save`` or ``paddle.static.save_inference_model`` or
    paddle 1.x API ``paddle.fluid.io.save_inference_model`` as ``paddle.jit.TranslatedLayer``,
897
    then performing inference or fine-tune training.
898 899

    .. note::
900
        If you load model saved by ``paddle.static.save_inference_model`` ,
901 902
        there will be the following limitations when using it in fine-tuning:
        1. Imperative mode do not support LoDTensor. All original model's feed targets or parametars that depend on LoD are temporarily unavailable.
903
        2. All saved model's feed targets need to be passed into TranslatedLayer's forward function.
904 905 906 907
        3. The variable's ``stop_gradient`` information is lost and can not be recovered.
        4. The parameter's ``trainable`` information is lost and can not be recovered.

    Args:
908
        path (str): The path prefix to load model. The format is ``dirname/file_prefix`` or ``file_prefix`` .
909 910
        **configs (dict, optional): Other load configuration options for compatibility. We do not
            recommend using these configurations, they may be removed in the future. If not necessary,
911 912
            DO NOT use them. Default None.
            The following options are currently supported:
913 914 915 916
            (1) model_filename (str): The inference model file name of the paddle 1.x
            ``save_inference_model`` save format. Default file name is :code:`__model__` .
            (2) params_filename (str): The persistable variables file name of the paddle 1.x
            ``save_inference_model`` save format. No default file name, save variables separately
917 918
            by default.

919 920 921 922 923

    Returns:
        TranslatedLayer: A Layer object can run saved translated model.

    Examples:
924
        1. Load model saved by ``paddle.jit.save`` then performing inference and fine-tune training.
925 926 927 928

        .. code-block:: python

            import numpy as np
929 930 931
            import paddle
            import paddle.nn as nn
            import paddle.optimizer as opt
932

933 934 935
            BATCH_SIZE = 16
            BATCH_NUM = 4
            EPOCH_NUM = 4
936

937 938
            IMAGE_SIZE = 784
            CLASS_NUM = 10
939

940 941 942 943
            # define a random dataset
            class RandomDataset(paddle.io.Dataset):
                def __init__(self, num_samples):
                    self.num_samples = num_samples
944

945 946 947 948
                def __getitem__(self, idx):
                    image = np.random.random([IMAGE_SIZE]).astype('float32')
                    label = np.random.randint(0, CLASS_NUM - 1, (1, )).astype('int64')
                    return image, label
949

950 951 952 953 954
                def __len__(self):
                    return self.num_samples

            class LinearNet(nn.Layer):
                def __init__(self):
955
                    super(LinearNet, self).__init__()
956
                    self._linear = nn.Linear(IMAGE_SIZE, CLASS_NUM)
957

958
                @paddle.jit.to_static
959 960 961
                def forward(self, x):
                    return self._linear(x)

962 963 964 965 966 967 968 969 970 971 972
            def train(layer, loader, loss_fn, opt):
                for epoch_id in range(EPOCH_NUM):
                    for batch_id, (image, label) in enumerate(loader()):
                        out = layer(image)
                        loss = loss_fn(out, label)
                        loss.backward()
                        opt.step()
                        opt.clear_grad()
                        print("Epoch {} batch {}: loss = {}".format(
                            epoch_id, batch_id, np.mean(loss.numpy())))

973
            # 1. train & save model.
974

975
            # create network
976 977 978 979
            layer = LinearNet()
            loss_fn = nn.CrossEntropyLoss()
            adam = opt.Adam(learning_rate=0.001, parameters=layer.parameters())

980
            # create data loader
981 982 983 984 985 986
            dataset = RandomDataset(BATCH_NUM * BATCH_SIZE)
            loader = paddle.io.DataLoader(dataset,
                batch_size=BATCH_SIZE,
                shuffle=True,
                drop_last=True,
                num_workers=2)
987

988 989
            # train
            train(layer, loader, loss_fn, adam)
990

991
            # save
992 993
            path = "example_model/linear"
            paddle.jit.save(layer, path)
994

995
            # 2. load model
996

997
            # load
998
            loaded_layer = paddle.jit.load(path)
999 1000

            # inference
1001 1002 1003
            loaded_layer.eval()
            x = paddle.randn([1, IMAGE_SIZE], 'float32')
            pred = loaded_layer(x)
1004 1005

            # fine-tune
1006 1007 1008
            loaded_layer.train()
            adam = opt.Adam(learning_rate=0.001, parameters=loaded_layer.parameters())
            train(loaded_layer, loader, loss_fn, adam)
1009 1010


1011
        2. Load model saved by ``paddle.fluid.io.save_inference_model`` then performing and fine-tune training.
1012 1013 1014 1015

        .. code-block:: python

            import numpy as np
1016
            import paddle
1017
            import paddle.static as static
1018 1019
            import paddle.nn as nn
            import paddle.optimizer as opt
1020
            import paddle.nn.functional as F
1021

1022 1023 1024
            BATCH_SIZE = 16
            BATCH_NUM = 4
            EPOCH_NUM = 4
1025

1026 1027 1028 1029 1030 1031 1032
            IMAGE_SIZE = 784
            CLASS_NUM = 10

            # define a random dataset
            class RandomDataset(paddle.io.Dataset):
                def __init__(self, num_samples):
                    self.num_samples = num_samples
1033

1034 1035 1036 1037
                def __getitem__(self, idx):
                    image = np.random.random([IMAGE_SIZE]).astype('float32')
                    label = np.random.randint(0, CLASS_NUM - 1, (1, )).astype('int64')
                    return image, label
1038

1039 1040
                def __len__(self):
                    return self.num_samples
1041

1042 1043
            paddle.enable_static()

1044 1045
            image = static.data(name='image', shape=[None, 784], dtype='float32')
            label = static.data(name='label', shape=[None, 1], dtype='int64')
1046
            pred = static.nn.fc(x=image, size=10, activation='softmax')
1047 1048
            loss = F.cross_entropy(input=pred, label=label)
            avg_loss = paddle.mean(loss)
1049

1050
            optimizer = paddle.optimizer.SGD(learning_rate=0.001)
1051 1052
            optimizer.minimize(avg_loss)

1053 1054 1055
            place = paddle.CPUPlace()
            exe = static.Executor(place)
            exe.run(static.default_startup_program())
1056

1057 1058 1059 1060 1061
            # create data loader
            dataset = RandomDataset(BATCH_NUM * BATCH_SIZE)
            loader = paddle.io.DataLoader(dataset,
                feed_list=[image, label],
                places=place,
1062
                batch_size=BATCH_SIZE,
1063 1064
                shuffle=True,
                drop_last=True,
W
WeiXin 已提交
1065
                return_list=False,
1066
                num_workers=2)
1067 1068 1069 1070

            # 1. train and save inference model
            for data in loader():
                exe.run(
1071
                    static.default_main_program(),
1072
                    feed=data,
1073 1074 1075
                    fetch_list=[avg_loss])

            model_path = "fc.example.model"
1076
            paddle.fluid.io.save_inference_model(
1077 1078 1079
                model_path, ["image"], [pred], exe)

            # 2. load model
1080 1081

            # enable dygraph mode
1082 1083 1084 1085
            paddle.disable_static(place)

            # load
            fc = paddle.jit.load(model_path)
1086

1087 1088 1089
            # inference
            fc.eval()
            x = paddle.randn([1, IMAGE_SIZE], 'float32')
1090 1091
            pred = fc(x)

1092
            # fine-tune
1093
            fc.train()
1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110
            loss_fn = nn.CrossEntropyLoss()
            adam = opt.Adam(learning_rate=0.001, parameters=fc.parameters())
            loader = paddle.io.DataLoader(dataset,
                places=place,
                batch_size=BATCH_SIZE,
                shuffle=True,
                drop_last=True,
                num_workers=2)
            for epoch_id in range(EPOCH_NUM):
                for batch_id, (image, label) in enumerate(loader()):
                    out = fc(image)
                    loss = loss_fn(out, label)
                    loss.backward()
                    adam.step()
                    adam.clear_grad()
                    print("Epoch {} batch {}: loss = {}".format(
                        epoch_id, batch_id, np.mean(loss.numpy())))
1111
    """
1112 1113 1114 1115
    # 1. construct correct config
    config = _parse_load_config(configs)
    model_path, config = _build_load_path_and_config(path, config)

1116
    return TranslatedLayer._construct(model_path, config)
1117 1118


1119
@dygraph_only
Z
Zeng Jinle 已提交
1120 1121 1122 1123 1124
def _trace(layer,
           inputs,
           feed_prefix='feed_',
           fetch_prefix='fetch_',
           tmp_prefix='t_'):
1125
    assert isinstance(layer, Layer)
1126 1127 1128 1129 1130 1131 1132 1133 1134

    if not isinstance(inputs, (list, tuple)):
        inputs = [inputs]

    tracer = _dygraph_tracer()._get_program_desc_tracer()

    var_list = extract_vars(inputs)

    with program_desc_tracing_guard(True):
1135
        original_outputs = layer(*inputs)
1136 1137 1138 1139
        if not isinstance(original_outputs, (list, tuple)):
            outputs = [original_outputs]
        else:
            outputs = original_outputs
1140
        out_vars = extract_vars(outputs, err_tag='outputs')
1141

1142
        program_desc, feed_names, fetch_names, parameters = tracer.create_program_desc(
Z
Zeng Jinle 已提交
1143
            var_list, feed_prefix, out_vars, fetch_prefix, tmp_prefix)
1144 1145 1146 1147 1148
        tracer.reset()

    with _dygraph_guard(None):
        program = create_program_from_desc(program_desc)

1149
    return original_outputs, program, feed_names, fetch_names, parameters
1150 1151 1152 1153


class TracedLayer(object):
    """
1154
    :api_attr: imperative
1155

1156 1157 1158 1159 1160
    TracedLayer is used to convert a forward dygraph model to a static
    graph model. This is mainly used to save the dygraph model for online
    inference using C++. Besides, users can also do inference in Python
    using the converted static graph model, which usually has better
    performance than the original dygraph model.
1161 1162 1163 1164

    TracedLayer would run the static graph model using :code:`Executor`
    and :code:`CompiledProgram` . The static graph model would share
    parameters with the dygraph model.
1165 1166

    All TracedLayer objects should not be created by constructor and should
1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177
    be created by static method :code:`TracedLayer.trace(layer, inputs)` .

    The TracedLayer can only be used to convert the data-independent dygraph
    model into the static graph model, which means the dygraph model should
    be independent with the tensor data and shape.
    """

    def __init__(self, program, parameters, feed_names, fetch_names):
        self._program = program
        self._feed_names = feed_names
        self._fetch_names = fetch_names
1178
        self._params = parameters
1179 1180 1181 1182 1183

        self._place = _current_expected_place()

        self._scope = core.Scope()
        for p in parameters:
1184
            src_tensor = p.value().get_tensor()
1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207
            dst_tensor = self._scope.var(p.name).get_tensor()
            dst_tensor._share_data_with(src_tensor)

        self._exe = Executor(self._place)
        self._compiled_program = None
        self._build_strategy = None
        self._exec_strategy = None

    @property
    def program(self):
        return self._program

    def _switch(self, is_test=True):
        for block_id in range(self._program.num_blocks):
            block = self._program.block(block_id)
            for op in block.ops:
                if op.has_attr("is_test"):
                    op._set_attr("is_test", is_test)

    @staticmethod
    @dygraph_only
    def trace(layer, inputs):
        """
1208
        This method is the only allowed method to create TracedLayer object.
1209 1210 1211 1212
        It would call the :code:`layer(*inputs)` method to run the dygraph
        model and convert it into a static graph model.

        Args:
1213
            layer (paddle.nn.Layer): the layer object to be traced.
1214 1215
            inputs (list(Tensor)|tuple(Tensor)|Tensor): the input tensors of
                the layer object.
1216 1217

        Returns:
1218
            tuple: A tuple of 2 items, whose the first item is the output of
1219 1220
                :code:`layer(*inputs)` , and the second item is the created
                TracedLayer object.
1221

1222
        Examples:
1223 1224
            .. code-block:: python:

1225
                import paddle
1226

1227
                class ExampleLayer(paddle.nn.Layer):
1228 1229
                    def __init__(self):
                        super(ExampleLayer, self).__init__()
1230
                        self._fc = paddle.nn.Linear(3, 10)
1231 1232 1233 1234

                    def forward(self, input):
                        return self._fc(input)

1235

1236 1237 1238 1239 1240 1241
                layer = ExampleLayer()
                in_var = paddle.uniform(shape=[2, 3], dtype='float32')
                out_dygraph, static_layer = paddle.jit.TracedLayer.trace(layer, inputs=[in_var])

                # run the static graph model using Executor inside
                out_static_graph = static_layer([in_var])
1242

1243 1244
                print(len(out_static_graph)) # 1
                print(out_static_graph[0].shape) # (2, 10)
1245

1246 1247
                # save the static graph model for inference
                static_layer.save_inference_model(dirname='./saved_infer_model')
1248

1249
        """
1250 1251 1252 1253
        assert isinstance(
            layer, Layer
        ), "The type of 'layer' in fluid.dygraph.jit.TracedLayer.trace must be fluid.dygraph.Layer, but received {}.".format(
            type(layer))
1254 1255
        outs, prog, feed, fetch, parameters = _trace(layer, inputs)
        traced = TracedLayer(prog, parameters, feed, fetch)
1256 1257 1258 1259 1260 1261 1262
        return outs, traced

    def set_strategy(self, build_strategy=None, exec_strategy=None):
        """
        Set the strategies when running static graph model.

        Args:
1263
            build_strategy (BuildStrategy, optional): build strategy of
1264 1265 1266 1267 1268 1269 1270 1271 1272 1273
                :code:`CompiledProgram` inside TracedLayer. Default None.
            exec_strategy (ExecutionStrategy, optional): execution strategy of
                :code:`CompiledProgram` inside TracedLayer. Default None.

        Returns:
            None

        Examples:
            .. code-block:: python:

1274
                import paddle
1275

1276
                class ExampleLayer(paddle.nn.Layer):
1277 1278
                    def __init__(self):
                        super(ExampleLayer, self).__init__()
1279
                        self._fc = paddle.nn.Linear(3, 10)
1280 1281 1282 1283

                    def forward(self, input):
                        return self._fc(input)

1284 1285 1286 1287
                layer = ExampleLayer()
                in_var = paddle.uniform(shape=[2, 3], dtype='float32')

                out_dygraph, static_layer = paddle.jit.TracedLayer.trace(layer, inputs=[in_var])
1288

1289 1290
                build_strategy = paddle.static.BuildStrategy()
                build_strategy.enable_inplace = True
1291

1292 1293
                exec_strategy = paddle.static.ExecutionStrategy()
                exec_strategy.num_threads = 2
1294

1295 1296
                static_layer.set_strategy(build_strategy=build_strategy, exec_strategy=exec_strategy)
                out_static_graph = static_layer([in_var])
1297 1298 1299

        """
        assert self._compiled_program is None, "Cannot set strategy after run"
1300 1301 1302 1303 1304 1305 1306 1307
        assert isinstance(
            build_strategy, (type(None), BuildStrategy)
        ), "The type of 'build_strategy' in fluid.dygraph.jit.TracedLayer.set_strategy must be fluid.BuildStrategy, but received {}.".format(
            type(build_strategy))
        assert isinstance(
            exec_strategy, (type(None), ExecutionStrategy)
        ), "The type of 'exec_strategy' in fluid.dygraph.jit.TracedLayer.set_strategy must be fluid.ExecutionStrategy, but received {}.".format(
            type(exec_strategy))
1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325
        self._build_strategy = build_strategy
        self._exec_strategy = exec_strategy

    @switch_to_static_graph
    def _compile(self):
        self._compiled_program = CompiledProgram(
            self._program).with_data_parallel(
                build_strategy=self._build_strategy,
                exec_strategy=self._exec_strategy,
                places=self._place)

    def _build_feed(self, inputs):
        assert isinstance(inputs, (list, tuple)), \
            "Inputs should be a list or tuple of variables"
        assert len(inputs) == len(self._feed_names)
        feed_dict = {}
        if in_dygraph_mode():
            for x, name in zip(inputs, self._feed_names):
1326
                feed_dict[name] = x.value().get_tensor()
1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346
        else:
            for x, name in zip(inputs, self._feed_names):
                feed_dict[name] = x

        return feed_dict

    @switch_to_static_graph
    def _run(self, feed):
        return self._exe.run(self._compiled_program,
                             feed=feed,
                             fetch_list=self._fetch_names)

    def __call__(self, inputs):
        with scope_guard(self._scope):
            if self._compiled_program is None:
                self._compile()

            return self._run(self._build_feed(inputs))

    @switch_to_static_graph
1347
    def save_inference_model(self, path, feed=None, fetch=None, **kwargs):
1348
        """
1349 1350
        Save the TracedLayer to a model for inference. The saved
        inference model can be loaded by C++ inference APIs.
1351

1352 1353 1354
        ``path`` is the prefix of saved objects, and the saved translated program file
        suffix is ``.pdmodel`` , the saved persistable variables file suffix is ``.pdiparams`` .

1355
        Args:
1356
            path(str): The path prefix to save model. The format is ``dirname/file_prefix`` or ``file_prefix``.
1357
            feed (list[int], optional): the input variable indices of the saved
1358
                inference model. If None, all input variables of the
1359 1360 1361 1362 1363 1364
                TracedLayer object would be the inputs of the saved inference
                model. Default None.
            fetch (list[int], optional): the output variable indices of the
                saved inference model. If None, all output variables of the
                TracedLayer object would be the outputs of the saved inference
                model. Default None.
1365
            kwargs: Supported keys including 'clip_extra'.set to True if you want to clip extra information for every operator.
1366 1367

        Returns:
1368
            None
1369 1370 1371 1372 1373

        Examples:
            .. code-block:: python:

                import numpy as np
1374
                import paddle
1375

1376
                class ExampleLayer(paddle.nn.Layer):
1377 1378
                    def __init__(self):
                        super(ExampleLayer, self).__init__()
1379
                        self._fc = paddle.nn.Linear(3, 10)
1380 1381 1382 1383

                    def forward(self, input):
                        return self._fc(input)

1384 1385
                save_dirname = './saved_infer_model'
                in_np = np.random.random([2, 3]).astype('float32')
1386 1387
                in_var = paddle.to_tensor(in_np)
                layer = ExampleLayer()
1388

1389 1390
                out_dygraph, static_layer = paddle.jit.TracedLayer.trace(layer, inputs=[in_var])
                static_layer.save_inference_model(save_dirname, feed=[0], fetch=[0])
1391

1392 1393 1394 1395
                paddle.enable_static()
                place = paddle.CPUPlace()
                exe = paddle.static.Executor(place)
                program, feed_vars, fetch_vars = paddle.static.load_inference_model(save_dirname,
1396
                                                    exe)
1397 1398 1399

                fetch, = exe.run(program, feed={feed_vars[0]: in_np}, fetch_list=fetch_vars)
                print(fetch.shape) # (2, 10)
1400
        """
1401
        check_type(path, "path", str,
1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414
                   "fluid.dygraph.jit.TracedLayer.save_inference_model")
        check_type(feed, "feed", (type(None), list),
                   "fluid.dygraph.jit.TracedLayer.save_inference_model")
        if isinstance(feed, list):
            for f in feed:
                check_type(f, "each element of feed", int,
                           "fluid.dygraph.jit.TracedLayer.save_inference_model")
        check_type(fetch, "fetch", (type(None), list),
                   "fluid.dygraph.jit.TracedLayer.save_inference_model")
        if isinstance(fetch, list):
            for f in fetch:
                check_type(f, "each element of fetch", int,
                           "fluid.dygraph.jit.TracedLayer.save_inference_model")
1415
        clip_extra = kwargs.get('clip_extra', False)
1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427
        # path check
        file_prefix = os.path.basename(path)
        if file_prefix == "":
            raise ValueError(
                "The input path MUST be format of dirname/file_prefix "
                "[dirname\\file_prefix in Windows system], but received "
                "file_prefix is empty string.")

        dirname = os.path.dirname(path)
        if dirname and not os.path.exists(dirname):
            os.makedirs(dirname)

1428
        from paddle.fluid.io import save_inference_model
1429 1430 1431 1432 1433

        def get_feed_fetch(all_vars, partial_vars):
            if partial_vars is None:
                return all_vars

1434
            return [all_vars[idx] for idx in partial_vars]
1435 1436 1437 1438 1439 1440 1441 1442 1443 1444

        with scope_guard(self._scope):
            feeded_var_names = get_feed_fetch(self._feed_names, feed)
            target_var_names = get_feed_fetch(self._fetch_names, fetch)
            target_vars = []
            for name in target_var_names:
                target_var = self._program.global_block().vars.get(name, None)
                assert target_var is not None, "{} cannot be found".format(name)
                target_vars.append(target_var)

1445 1446 1447
            model_filename = file_prefix + INFER_MODEL_SUFFIX
            params_filename = file_prefix + INFER_PARAMS_SUFFIX

1448
            save_inference_model(
1449 1450 1451 1452
                dirname=dirname,
                feeded_var_names=feeded_var_names,
                target_vars=target_vars,
                executor=self._exe,
1453 1454
                main_program=self._program.clone(),
                model_filename=model_filename,
1455 1456
                params_filename=params_filename,
                clip_extra=clip_extra)