post_training_quantization.py 63.3 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14
#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import math
15 16
import os
import re
17 18
import logging
import numpy as np
19
import shutil
20
from inspect import isgeneratorfunction
21 22 23
from .... import io
from .... import core
from .... import framework
24
from .... import unique_name
25
from ....executor import global_scope, Executor
26 27 28 29 30
from ....framework import IrGraph
from ....log_helper import get_logger
from .quantization_pass import QuantizationTransformPass
from .quantization_pass import QuantizationFreezePass
from .quantization_pass import AddQuantDequantPass
31 32 33
from .quantization_pass import _out_scale_op_list
from .quantization_pass import _get_op_input_var_names
from .quantization_pass import _get_op_output_var_names
34
from .quantization_pass import _get_output_name_index
35
from .quantization_pass import _get_input_name_index
36
from .quantization_pass import _channelwise_quant_axis1_ops
37
from .cal_kl_threshold import cal_kl_threshold
38

39
__all__ = ['PostTrainingQuantization', 'WeightQuantization']
40 41 42 43 44

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


45 46 47 48
def _load_variable_data(scope, var_name):
    '''
    Load variable value from scope
    '''
49 50 51 52
    var_node = scope.find_var(var_name)
    assert var_node is not None, \
        "Cannot find " + var_name + " in scope."
    return np.array(var_node.get_tensor())
53 54 55 56 57 58 59


def _set_variable_data(scope, place, var_name, np_value):
    '''
    Set the value of var node by name, if the node exits,
    '''
    assert isinstance(np_value, np.ndarray), \
X
XGZhang 已提交
60
       'The type of value should be numpy array.'
61 62 63 64 65 66
    var_node = scope.find_var(var_name)
    if var_node != None:
        tensor = var_node.get_tensor()
        tensor.set(np_value, place)


67 68 69 70 71 72 73 74
def _all_persistable_var_names(program):
    persistable_var_names = []
    for var in program.list_vars():
        if var.persistable:
            persistable_var_names.append(var.name)
    return persistable_var_names


75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124
def _remove_unused_var_nodes(graph):
    all_used_vars = set()
    ops = graph.all_op_nodes()
    for op_node in ops:
        for input_node in op_node.inputs:
            all_used_vars.add(input_node)
        for output_node in op_node.outputs:
            all_used_vars.add(output_node)

    all_used_vars = {n.node for n in all_used_vars}
    all_unused_vars = {
        n
        for n in filter(lambda node: node.node not in all_used_vars,
                        graph.all_var_nodes())
    }
    graph.safe_remove_nodes(all_unused_vars)
    return graph


def _remove_ctrl_vars(graph):
    remove_ctr_vars = set()
    for node in graph.all_var_nodes():
        if node.is_ctrl_var():
            remove_ctr_vars.add(node)
    graph.safe_remove_nodes(remove_ctr_vars)
    return graph


def _apply_pass(scope,
                graph,
                pass_name,
                attrs=None,
                attr_values=None,
                debug=False):
    ir_pass = core.get_pass(pass_name)
    cpp_graph = graph.graph
    if not cpp_graph.has('__param_scope__'):
        cpp_graph.set_not_owned('__param_scope__', scope)
    if attrs:
        assert attr_values and len(attrs) == len(
            attr_values), "Different number of pass attributes and their values."
        for attr, value in zip(attrs, attr_values):
            ir_pass.set(attr, value)
    ir_pass.apply(cpp_graph)
    if debug:
        graph.draw('.', 'qat_fp32_{}'.format(pass_name), graph.all_op_nodes())
    _remove_unused_var_nodes(graph)
    return graph


125
class PostTrainingQuantization(object):
126 127 128 129 130 131
    """
    Utilizing post training quantization methon to quantize the FP32 model,
    and it uses calibrate data to get the quantization information for all 
    quantized variables.
    """

132
    def __init__(self,
133 134 135
                 executor=None,
                 scope=None,
                 model_dir=None,
136 137
                 model_filename=None,
                 params_filename=None,
138
                 batch_generator=None,
139
                 sample_generator=None,
140
                 data_loader=None,
141 142 143
                 batch_size=10,
                 batch_nums=None,
                 algo="KL",
X
XGZhang 已提交
144
                 hist_percent=0.99999,
145
                 quantizable_op_type=["conv2d", "depthwise_conv2d", "mul"],
146
                 is_full_quantize=False,
X
XGZhang 已提交
147
                 bias_correction=False,
148
                 activation_bits=8,
149 150 151
                 weight_bits=8,
                 activation_quantize_type='range_abs_max',
                 weight_quantize_type='channel_wise_abs_max',
152
                 optimize_model=False,
153
                 is_use_cache_file=False,
154
                 cache_dir=None):
155
        '''
156
        Constructor.
157 158

        Args:
159
            executor(fluid.Executor): The executor to load, run and save the
160
                quantized model.
161 162
            scope(fluid.Scope, optional): The scope of the program, use it to load 
                and save variables. If scope=None, get scope by global_scope(). 
163 164 165 166 167 168 169 170 171
            model_dir(str): The path of the fp32 model that will be quantized, 
                and the model and params files are under the path.
            model_filename(str, optional): The name of file to load the inference 
                program. If it is None, the default filename '__model__' will 
                be used. Default is 'None'.
            params_filename(str, optional): The name of file to load all parameters.
                When all parameters were saved in a single binary file, set it 
                as the real filename. If parameters were saved in separate files, 
                set it as 'None'. Default is 'None'.
172 173 174 175 176 177 178 179
            batch_generator(Python Generator): The batch generator provides 
                calibrate data for DataLoader, and it returns a batch every
                time. Note that, sample_generator and batch_generator, only one
                should be set. Beisdes, batch_generator supports lod tensor.
            sample_generator(Python Generator): The sample generator provides
                calibrate data for DataLoader, and it only returns a sample every
                time. Note that, sample_generator and batch_generator, only one
                should be set. Beisdes, sample_generator dose not support lod tensor.
180 181 182
            data_loader(Python Generator, Paddle.io.DataLoader, optional): The
                Generator or Dataloader provides calibrate data, and it could
                return a batch every time.
183 184 185 186
            batch_size(int, optional): The batch size of DataLoader. Default is 10.
            batch_nums(int, optional): If batch_nums is not None, the number of 
                calibrate data is batch_size*batch_nums. If batch_nums is None, use 
                all data provided by sample_generator as calibrate data.
187 188 189 190
            algo(str, optional): If algo='KL', use KL-divergenc method to
                get the KL threshold for quantized activations and get the abs_max
                value for quantized weights. If algo='abs_max', get the abs max 
                value for activations and weights. If algo= 'min_max', get the min 
X
XGZhang 已提交
191 192 193 194 195 196 197
                and max value for quantized activations and weights. If algo='avg',
                get the average value among the max values for activations. If 
                algo= 'hist', get the value of 'hist_percent' quantile as the threshold.
                If algo='mse', get the value which makes the quantization mse loss 
                minimal. Default is KL.
            hist_percent(float, optional): The threshold of algo 'hist' for activations.
                Default is 0.99999.
198 199
            quantizable_op_type(list[str], optional): List the type of ops 
                that will be quantized. Default is ["conv2d", "depthwise_conv2d", 
200 201
                "mul"].
            is_full_quantized(bool, optional): If set is_full_quantized as True, 
202
                apply quantization to all supported quantizable op type. If set
203 204
                is_full_quantized as False, only apply quantization to the op type 
                according to the input quantizable_op_type.
X
XGZhang 已提交
205 206
            bias_correction(bool, optional): If set as True, use the bias correction
                method of https://arxiv.org/abs/1810.05723. Default is False.
207
            activation_bits(int): quantization bit number for activation.
208 209 210 211 212 213 214 215 216 217 218 219
            weight_bits(int, optional): quantization bit number for weights.
            activation_quantize_type(str): quantization type for activation,
                now support 'range_abs_max', 'moving_average_abs_max' and 'abs_max'.
                This param only specifies the fake ops in saving quantized model.
                If it is 'range_abs_max' or 'moving_average_abs_max', we save the scale
                obtained by post training quantization in fake ops. Note that, if it
                is 'abs_max', the scale will not be saved in fake ops.
            weight_quantize_type(str): quantization type for weights,
                support 'abs_max' and 'channel_wise_abs_max'. This param only specifies
                the fake ops in saving quantized model, and we save the scale obtained
                by post training quantization in fake ops. Compared to 'abs_max',
                the model accuracy is usually higher when it is 'channel_wise_abs_max'.
220 221 222 223 224 225 226 227
            optimize_model(bool, optional): If set optimize_model as True, it applies
                some passes to the model before quantization, and it supports
                `conv2d/depthwise_conv2d + bn` pass so far. Some targets require the
                weights are quantized by tensor-wise method, which means the weights
                scale for all channel are the same. However, if fuse
                `conv2d/depthwise_conv2d + bn`, the weights scale for all channel will
                be different. In address this problem, fuse the pattern before
                quantization. Default False.
228 229
            is_use_cache_file(bool, optional): This param is deprecated.
            cache_dir(str, optional): This param is deprecated.
230 231 232
        Returns:
            None

233 234 235 236 237 238
        Examples:
        .. code-block:: python
            import paddle.fluid as fluid
            from paddle.fluid.contrib.slim.quantization import PostTrainingQuantization
            
            exe = fluid.Executor(fluid.CPUPlace())
239 240 241 242 243 244 245 246 247
            model_dir = path/to/fp32_model_params
            # set model_filename as None when the filename is __model__, 
            # otherwise set it as the real filename
            model_filename = None 
            # set params_filename as None when all parameters were saved in 
            # separate files, otherwise set it as the real filename
            params_filename = None
            save_model_path = path/to/save_model_path
            # prepare the sample generator according to the model, and the 
248
            # sample generator must return a sample every time. The reference
249 250 251
            # document: https://www.paddlepaddle.org.cn/documentation/docs/zh
            # /user_guides/howto/prepare_data/use_py_reader.html
            sample_generator = your_sample_generator
252 253 254
            batch_size = 10
            batch_nums = 10
            algo = "KL"
255
            quantizable_op_type = ["conv2d", "depthwise_conv2d", "mul"]
256 257
            ptq = PostTrainingQuantization(
                        executor=exe,
258 259 260 261
                        sample_generator=sample_generator,
                        model_dir=model_dir,
                        model_filename=model_filename,
                        params_filename=params_filename,
262 263 264 265 266 267 268
                        batch_size=batch_size,
                        batch_nums=batch_nums,
                        algo=algo,
                        quantizable_op_type=quantizable_op_type)
            ptq.quantize()
            ptq.save_quantized_model(save_model_path)
        '''
269

270 271 272 273
        self._support_activation_quantize_type = [
            'range_abs_max', 'moving_average_abs_max', 'abs_max'
        ]
        self._support_weight_quantize_type = ['abs_max', 'channel_wise_abs_max']
X
XGZhang 已提交
274
        self._support_algo_type = [
275
            'KL', 'hist', 'avg', 'mse', 'emd', 'abs_max', 'min_max'
X
XGZhang 已提交
276
        ]
277
        self._dynamic_quantize_op_type = ['lstm']
278 279
        self._support_quantize_op_type = \
            list(set(QuantizationTransformPass._supported_quantizable_op_type +
280 281
                AddQuantDequantPass._supported_quantizable_op_type +
                self._dynamic_quantize_op_type))
282 283

        # Check inputs
284 285
        assert executor is not None, "The executor cannot be None."
        assert model_dir is not None, "The model_dir cannot be None."
286
        assert any([gen is not None] for gen in [sample_generator,
287 288 289 290 291
            batch_generator, data_loader]), "The sample_generator, batch_generator " \
            "and data_loader cannot be None in the same time."
        if data_loader is not None:
            assert isinstance(data_loader, (io.DataLoader, type(isgeneratorfunction))), \
                "data_loader only accepts `paddle.io.DataLoader` or Generator instance."
292 293
        assert batch_size > 0, "The batch_size should be greater than 0."
        assert algo in self._support_algo_type, \
X
XGZhang 已提交
294
            "The algo should be KL, hist, mse, avg, abs_max or min_max."
295 296 297 298 299 300 301 302
        assert activation_quantize_type in self._support_activation_quantize_type, \
            "The activation_quantize_type ({}) should in ({}).".format(
            activation_quantize_type, self._support_activation_quantize_type)
        assert weight_quantize_type in self._support_weight_quantize_type, \
            "The weight_quantize_type ({}) shoud in ({}).".format(
            weight_quantize_type, self._support_weight_quantize_type)

        # Save input params
X
XGZhang 已提交
303
        self._bias_correction = bias_correction
304
        self._executor = executor
305
        self._scope = global_scope() if scope == None else scope
306 307 308
        self._model_dir = model_dir
        self._model_filename = model_filename
        self._params_filename = params_filename
309
        self._sample_generator = sample_generator
310
        self._batch_generator = batch_generator
311 312 313
        self._batch_size = batch_size
        self._batch_nums = batch_nums
        self._algo = algo
X
XGZhang 已提交
314
        self._hist_percent = hist_percent
315 316 317 318 319
        self._activation_bits = activation_bits
        self._weight_bits = weight_bits
        self._activation_quantize_type = activation_quantize_type
        self._weight_quantize_type = weight_quantize_type
        self._is_full_quantize = is_full_quantize
320
        if is_full_quantize:
321
            self._quantizable_op_type = self._support_quantize_op_type
322 323 324
        else:
            self._quantizable_op_type = quantizable_op_type
            for op_type in self._quantizable_op_type:
325
                assert op_type in self._support_quantize_op_type, \
326
                    op_type + " is not supported for quantization."
327
        self._optimize_model = optimize_model
328

329
        # Define variables
330 331 332 333
        self._place = self._executor.place
        self._program = None
        self._feed_list = None
        self._fetch_list = None
334
        self._data_loader = data_loader
335

336
        self._out_scale_op_list = _out_scale_op_list
337 338
        self._quantized_weight_var_name = set()
        self._quantized_act_var_name = set()
339
        self._weight_op_pairs = {}
X
XGZhang 已提交
340
        # The vars for alog = KL or hist
341 342
        self._sampling_act_abs_min_max = {}
        self._sampling_act_histogram = {}
343
        self._sampling_data = {}
X
XGZhang 已提交
344
        self._quantized_var_threshold = {}
345 346
        self._histogram_bins = 2048
        # The vars for algo = min_max
347 348
        self._quantized_var_min = {}
        self._quantized_var_max = {}
X
XGZhang 已提交
349 350 351
        # The vars for algo = avg
        self._quantized_var_avg = {}
        # The best loss of algo = mse
352
        self._best_calibration_loss = {}
X
XGZhang 已提交
353 354
        # The threshold for algo = abs_max, mse or avg
        self._quantized_threshold = {}
355 356 357

    def quantize(self):
        '''
358 359 360
        Load the FP32 model, and use the calibrate data to calculate the forward-stage.
        Based on the sample data, we can get the quantization information, and obtain
        the final quantized model.
361 362 363 364

        Args:
            None
        Returns:
365 366
            the program of quantized model.
        '''
367
        self._load_model_data()
368
        self._collect_target_varnames()
369
        self._set_activation_persistable()
370

X
XGZhang 已提交
371
        if self._algo in ["KL", "hist"]:
372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389
            _logger.info("Preparation stage ...")
            batch_id = 0
            for data in self._data_loader():
                self._executor.run(program=self._program,
                                   feed=data,
                                   fetch_list=self._fetch_list,
                                   return_numpy=False,
                                   scope=self._scope)
                self._collect_activation_abs_min_max()
                if batch_id % 5 == 0:
                    _logger.info("Run batch: " + str(batch_id))
                batch_id += 1
                if self._batch_nums and batch_id >= self._batch_nums:
                    break
            _logger.info("Finish preparation stage, all batch:" + str(batch_id))
            self._init_sampling_act_histogram()

        _logger.info("Sampling stage ...")
390 391 392 393
        batch_id = 0
        for data in self._data_loader():
            self._executor.run(program=self._program,
                               feed=data,
394
                               fetch_list=self._fetch_list,
395 396
                               return_numpy=False,
                               scope=self._scope)
397
            self._sampling()
398
            if batch_id % 5 == 0:
399
                _logger.info("Run batch: " + str(batch_id))
400 401 402
            batch_id += 1
            if self._batch_nums and batch_id >= self._batch_nums:
                break
403
        _logger.info("Finish sampling stage, all batch: " + str(batch_id))
404
        self._reset_activation_persistable()
X
XGZhang 已提交
405 406 407 408 409 410
        if self._algo == 'avg':
            for var_name in self._quantized_act_var_name:
                self._quantized_threshold[var_name] = \
                np.array(self._quantized_var_avg[var_name]).mean()
        if self._algo in ["KL", "hist"]:
            self._calculate_kl_hist_threshold()
411
        if self._algo in ["KL", "abs_max", "hist", "avg", "mse", "emd"]:
412 413 414 415 416
            self._update_program()
        else:
            self._save_input_threhold()

        self._save_output_threshold()
417 418 419 420
        if any(op_type in self._quantizable_op_type
               for op_type in self._dynamic_quantize_op_type):
            self._collect_dynamic_quantize_op_threshold(
                self._dynamic_quantize_op_type)
421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437

        # Move sub blocks persistable var to global block
        global_block = self._program.global_block()
        for _op in global_block.ops:
            if _op.type == "while":
                _block_id = _op.attr("sub_block").id
                _block = self._program.block(_block_id)
                persistables = []
                for _name, _var in _block.vars.items():
                    if _var.persistable:
                        global_block._clone_variable(_var)
                        persistables.append(_name)
                for _name in persistables:
                    _block._remove_var(_name)
                persistables.extend(_op.input('X'))
                _op.desc.set_input("X", persistables)

438 439
        return self._program

440 441 442 443
    def save_quantized_model(self,
                             save_model_path,
                             model_filename=None,
                             params_filename=None):
444 445 446 447
        '''
        Save the quantized model to the disk.

        Args:
448 449 450 451 452 453 454
            save_model_path(str): The path to save the quantized model.
            model_filename(str, optional): If the model_filename is None,
                save the model to '__model__'. Otherwise, save the model
                to the specified filename. Default: None.
            params_filename(str, optional): If the params_filename is None,
                save params to separted files. Otherwise, save all params
                to the specified filename.
455
        Returns:
456 457 458 459
            None
        '''
        io.save_inference_model(
            dirname=save_model_path,
460 461
            model_filename=model_filename,
            params_filename=params_filename,
462 463 464 465
            feeded_var_names=self._feed_list,
            target_vars=self._fetch_list,
            executor=self._executor,
            main_program=self._program)
466
        _logger.info("The quantized model is saved in " + save_model_path)
467

468
    def _load_model_data(self):
469
        '''
470
        Load model and set data loader.
471
        '''
472
        _logger.info("Load model and set data loader ...")
473
        [self._program, self._feed_list, self._fetch_list] = \
474 475 476 477
            io.load_inference_model(dirname=self._model_dir,
                                    executor=self._executor,
                                    model_filename=self._model_filename,
                                    params_filename=self._params_filename)
478 479 480 481

        if self._optimize_model:
            self._optimize_fp32_model()

482 483
        feed_vars = [framework._get_var(str(var_name), self._program) \
            for var_name in self._feed_list]
484 485 486

        if self._data_loader is not None:
            return
487 488
        self._data_loader = io.DataLoader.from_generator(
            feed_list=feed_vars, capacity=3 * self._batch_size, iterable=True)
489 490 491 492 493 494 495 496 497 498
        if self._sample_generator is not None:
            self._data_loader.set_sample_generator(
                self._sample_generator,
                batch_size=self._batch_size,
                drop_last=True,
                places=self._place)
        elif self._batch_generator is not None:
            self._data_loader.set_batch_generator(
                self._batch_generator, places=self._place)

499 500 501 502 503 504 505 506
    def _optimize_fp32_model(self):
        '''
        Fuse the `conv2d/depthwise_conv2d + bn` in FP32 model.
        '''
        _logger.info("Optimize FP32 model ...")
        graph = IrGraph(core.Graph(self._program.desc), for_test=True)
        graph = _remove_ctrl_vars(graph)
        graph = _apply_pass(self._scope, graph, 'conv_bn_fuse_pass')
507 508
        graph = _apply_pass(self._scope, graph, 'depthwise_conv_bn_fuse_pass')
        graph = _apply_pass(self._scope, graph, 'conv_transpose_bn_fuse_pass')
509 510 511 512
        graph = _apply_pass(self._scope, graph, 'conv_eltwiseadd_bn_fuse_pass')
        graph = _apply_pass(self._scope, graph,
                            'depthwise_conv_eltwiseadd_bn_fuse_pass')

513 514
        self._program = graph.to_program()

515
    def _collect_target_varnames(self):
516 517 518 519
        '''
        Collect the variable names for sampling, and set activation
        variables to be persistable.
        '''
520
        # TODO(juncaipeng), consider the name_scope of skip_quant
521
        _logger.info("Collect quantized variable names ...")
522

523
        def collect_var_name(var_name_list, persistable_var_names, op_type):
524 525 526
            for var_name in var_name_list:
                if var_name in persistable_var_names:
                    self._quantized_weight_var_name.add(var_name)
527
                    self._weight_op_pairs[var_name] = op_type
528 529 530
                else:
                    self._quantized_act_var_name.add(var_name)

531
        persistable_var_names = _all_persistable_var_names(self._program)
532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551
        for block_id in range(len(self._program.blocks)):
            for op in self._program.blocks[block_id].ops:
                op_type = op.type
                if self._is_full_quantize and \
                    op_type not in self._quantizable_op_type:
                    _logger.warning(op_type +
                                    " is not supported for quantization.")
                # For quantized ops, sample inputs and outputs
                if op_type in self._quantizable_op_type:
                    collect_var_name(
                        _get_op_input_var_names(op), persistable_var_names,
                        op_type)
                    collect_var_name(
                        _get_op_output_var_names(op), persistable_var_names,
                        op_type)
                # For other op, only sample output scale
                elif op_type in self._out_scale_op_list:
                    collect_var_name(
                        _get_op_output_var_names(op), persistable_var_names,
                        op_type)
552 553 554 555 556 557

    def _set_activation_persistable(self):
        '''
        Set activation variables to be persistable, so can obtain 
        the tensor data in sample_data
        '''
558 559 560 561
        for var in self._program.list_vars():
            if var.name in self._quantized_act_var_name:
                var.persistable = True

562 563 564 565
    def _reset_activation_persistable(self):
        '''
        Reset activations to be not persistable.
        '''
566
        to_erase = []
567 568 569
        for var in self._program.list_vars():
            if var.name in self._quantized_act_var_name:
                var.persistable = False
570 571
                to_erase.append(var.name)
        self._scope.erase(to_erase)
572

573
    def _sampling(self):
574
        '''
575
        Sample the min/max, abs_max or histogram in every iterations.
576 577
        '''
        if self._algo == "abs_max":
578
            self._sample_abs_max()
X
XGZhang 已提交
579 580
        elif self._algo == "avg":
            self._sample_avg()
581
        elif self._algo == "min_max":
582
            self._sample_min_max()
X
XGZhang 已提交
583 584
        elif self._algo == "mse":
            self._sample_mse()
585 586
        elif self._algo == "emd":
            self._sample_emd()
X
XGZhang 已提交
587
        elif self._algo in ["KL", "hist"]:
588
            self._sample_histogram()
589

X
XGZhang 已提交
590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612
    def _sample_mse(self):
        if self._quantized_threshold == {}:
            for var_name in self._quantized_weight_var_name:
                var_tensor = _load_variable_data(self._scope, var_name)
                if self._weight_quantize_type == "abs_max":
                    abs_max_value = float(np.max(np.abs(var_tensor)))
                elif self._weight_quantize_type == "channel_wise_abs_max":
                    abs_max_value = []
                    if self._weight_op_pairs[
                            var_name] in _channelwise_quant_axis1_ops:
                        for i in range(var_tensor.shape[1]):
                            abs_max_value.append(
                                float(np.max(np.abs(var_tensor[:, i]))))
                    else:
                        for i in range(var_tensor.shape[0]):
                            abs_max_value.append(
                                float(np.max(np.abs(var_tensor[i]))))
                self._quantized_threshold[var_name] = abs_max_value
        _logger.info("MSE searching stage ...")
        for var_name in self._quantized_act_var_name:
            var_tensor = _load_variable_data(self._scope, var_name)
            var_tensor = var_tensor.flatten()
            abs_max_value = float(np.max(np.abs(var_tensor)))
X
XGZhang 已提交
613
            abs_max_value = 1e-8 if abs_max_value == 0.0 else abs_max_value
X
XGZhang 已提交
614
            s = 0.3
615 616
            if var_name not in self._best_calibration_loss:
                self._best_calibration_loss[var_name] = float('inf')
X
XGZhang 已提交
617 618 619 620 621 622 623 624
            while s <= 1.0:
                scale = s * abs_max_value
                s += 0.02
                bins = 2**(self._activation_bits - 1) - 1
                quant_dequant_var = np.round(
                    np.clip(var_tensor, 0.0, scale) / scale *
                    bins) / bins * scale
                mse_loss = ((var_tensor - quant_dequant_var)**2).mean()
625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667
                if mse_loss <= self._best_calibration_loss[var_name]:
                    self._best_calibration_loss[var_name] = mse_loss
                    self._quantized_threshold[var_name] = scale

    def _sample_emd(self):
        if self._quantized_threshold == {}:
            for var_name in self._quantized_weight_var_name:
                var_tensor = _load_variable_data(self._scope, var_name)
                if self._weight_quantize_type == "abs_max":
                    abs_max_value = float(np.max(np.abs(var_tensor)))
                elif self._weight_quantize_type == "channel_wise_abs_max":
                    abs_max_value = []
                    if self._weight_op_pairs[
                            var_name] in _channelwise_quant_axis1_ops:
                        for i in range(var_tensor.shape[1]):
                            abs_max_value.append(
                                float(np.max(np.abs(var_tensor[:, i]))))
                    else:
                        for i in range(var_tensor.shape[0]):
                            abs_max_value.append(
                                float(np.max(np.abs(var_tensor[i]))))
                self._quantized_threshold[var_name] = abs_max_value
        _logger.info("EMD searching stage ...")
        for var_name in self._quantized_act_var_name:
            var_tensor = _load_variable_data(self._scope, var_name)
            var_tensor = var_tensor.flatten()
            abs_max_value = float(np.max(np.abs(var_tensor)))
            abs_max_value = 1e-8 if abs_max_value == 0.0 else abs_max_value
            s = 0.3
            if var_name not in self._best_calibration_loss:
                self._best_calibration_loss[var_name] = float('inf')
            while s <= 1.0:
                scale = s * abs_max_value
                s += 0.02
                bins = 2**(self._activation_bits - 1) - 1
                quant_dequant_var = np.round(
                    np.clip(var_tensor, 0.0, scale) / scale *
                    bins) / bins * scale
                emd_loss = np.abs(
                    np.mean(var_tensor) - np.mean(quant_dequant_var)) + np.abs(
                        np.std(var_tensor) - np.std(quant_dequant_var))
                if emd_loss <= self._best_calibration_loss[var_name]:
                    self._best_calibration_loss[var_name] = emd_loss
X
XGZhang 已提交
668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698
                    self._quantized_threshold[var_name] = scale

    def _sample_avg(self):
        if self._quantized_threshold == {}:
            for var_name in self._quantized_weight_var_name:
                var_tensor = _load_variable_data(self._scope, var_name)
                if self._weight_quantize_type == "abs_max":
                    abs_max_value = float(np.max(np.abs(var_tensor)))
                elif self._weight_quantize_type == "channel_wise_abs_max":
                    abs_max_value = []
                    if self._weight_op_pairs[
                            var_name] in _channelwise_quant_axis1_ops:
                        for i in range(var_tensor.shape[1]):
                            abs_max_value.append(
                                float(np.max(np.abs(var_tensor[:, i]))))
                    else:
                        for i in range(var_tensor.shape[0]):
                            abs_max_value.append(
                                float(np.max(np.abs(var_tensor[i]))))
                self._quantized_threshold[var_name] = abs_max_value

        for var_name in self._quantized_act_var_name:
            var_tensor = _load_variable_data(self._scope, var_name)
            abs_max_value = float(np.max(np.abs(var_tensor)))
            if (var_name not in self._quantized_var_avg):
                self._quantized_var_avg[var_name] = []
            abs_avg_value = float(np.mean(np.max(  \
            np.abs(var_tensor.reshape(var_tensor.shape[0], -1)), axis=(1))))
            self._quantized_var_avg[var_name].append(abs_avg_value)
            continue

699
    def _sample_abs_max(self):
X
XGZhang 已提交
700
        if self._quantized_threshold == {}:
701 702 703 704 705 706
            for var_name in self._quantized_weight_var_name:
                var_tensor = _load_variable_data(self._scope, var_name)
                if self._weight_quantize_type == "abs_max":
                    abs_max_value = float(np.max(np.abs(var_tensor)))
                elif self._weight_quantize_type == "channel_wise_abs_max":
                    abs_max_value = []
707
                    if self._weight_op_pairs[
708 709 710 711 712 713 714 715
                            var_name] in _channelwise_quant_axis1_ops:
                        for i in range(var_tensor.shape[1]):
                            abs_max_value.append(
                                float(np.max(np.abs(var_tensor[:, i]))))
                    else:
                        for i in range(var_tensor.shape[0]):
                            abs_max_value.append(
                                float(np.max(np.abs(var_tensor[i]))))
X
XGZhang 已提交
716
                self._quantized_threshold[var_name] = abs_max_value
717 718 719 720

        for var_name in self._quantized_act_var_name:
            var_tensor = _load_variable_data(self._scope, var_name)
            abs_max_value = float(np.max(np.abs(var_tensor)))
X
XGZhang 已提交
721 722 723
            if (var_name not in self._quantized_threshold) or \
                (abs_max_value > self._quantized_threshold[var_name]):
                self._quantized_threshold[var_name] = abs_max_value
724

725
    def _sample_min_max(self):
726 727
        if self._quantized_var_min == {} and self._quantized_var_max == {}:
            for var_name in self._quantized_weight_var_name:
728
                var_tensor = _load_variable_data(self._scope, var_name)
729 730 731 732 733 734
                if self._weight_quantize_type == "abs_max":
                    min_value = float(np.min(var_tensor))
                    max_value = float(np.max(var_tensor))
                elif self._weight_quantize_type == "channel_wise_abs_max":
                    min_value = []
                    max_value = []
735
                    if self._weight_op_pairs[
736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756
                            var_name] in _channelwise_quant_axis1_ops:
                        for i in range(var_tensor.shape[1]):
                            min_value.append(float(np.min(var_tensor[:, i])))
                            max_value.append(float(np.max(var_tensor[:, i])))
                    else:
                        for i in range(var_tensor.shape[0]):
                            min_value.append(float(np.min(var_tensor[i])))
                            max_value.append(float(np.max(var_tensor[i])))
                self._quantized_var_min[var_name] = min_value
                self._quantized_var_max[var_name] = max_value

        for var_name in self._quantized_act_var_name:
            var_tensor = _load_variable_data(self._scope, var_name)
            min_value = float(np.min(var_tensor))
            max_value = float(np.max(var_tensor))
            if (var_name not in self._quantized_var_min) or \
                (min_value < self._quantized_var_min[var_name]):
                self._quantized_var_min[var_name] = min_value
            if (var_name not in self._quantized_var_max) or \
                (max_value > self._quantized_var_max[var_name]):
                self._quantized_var_max[var_name] = max_value
757

758 759 760 761 762 763 764 765
    def _sample_histogram(self):
        for var_name in self._quantized_act_var_name:
            var_tensor = _load_variable_data(self._scope, var_name)
            var_tensor_abs = np.abs(var_tensor)
            bins = self._sampling_act_histogram[var_name][1]
            hist, _ = np.histogram(var_tensor_abs, bins=bins)
            self._sampling_act_histogram[var_name][0] += hist

766 767 768 769 770 771
    def _save_input_threhold(self):
        '''
        Save input threshold to the quantized op.
        '''
        assert self._algo == "min_max", \
            "The algo should be min_max to save input threshold."
772 773 774 775 776 777 778 779 780 781 782
        for block_id in range(len(self._program.blocks)):
            for op in self._program.blocks[block_id].ops:
                if op.type in self._quantizable_op_type:
                    for var_name in _get_op_input_var_names(op):
                        assert var_name in self._quantized_var_min
                        assert var_name in self._quantized_var_max
                        op._set_attr(var_name + ".min",
                                     self._quantized_var_min[var_name])
                        op._set_attr(var_name + ".max",
                                     self._quantized_var_max[var_name])
                        op._set_attr("with_quant_attr", True)
783

784
    def _collect_activation_abs_min_max(self):
785
        '''
786 787
        Collect the abs_min and abs_max for all activation. When algo = KL,
        get the min and max value, and then calculate the threshold.
788
        '''
789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813
        for var_name in self._quantized_act_var_name:
            var_tensor = _load_variable_data(self._scope, var_name)
            var_tensor = np.abs(var_tensor)
            min_value = float(np.min(var_tensor))
            max_value = float(np.max(var_tensor))
            if var_name not in self._sampling_act_abs_min_max:
                self._sampling_act_abs_min_max[
                    var_name] = [min_value, max_value]
            else:
                if min_value < self._sampling_act_abs_min_max[var_name][0]:
                    self._sampling_act_abs_min_max[var_name][0] = min_value
                if max_value > self._sampling_act_abs_min_max[var_name][1]:
                    self._sampling_act_abs_min_max[var_name][1] = max_value

    def _init_sampling_act_histogram(self):
        '''
        Based on the min/max value, init the sampling_act_histogram.
        '''
        for var_name in self._quantized_act_var_name:
            if var_name not in self._sampling_act_histogram:
                min_val = self._sampling_act_abs_min_max[var_name][0]
                max_val = self._sampling_act_abs_min_max[var_name][1]
                hist, hist_edeges = np.histogram(
                    [], bins=self._histogram_bins, range=(min_val, max_val))
                self._sampling_act_histogram[var_name] = [hist, hist_edeges]
814

X
XGZhang 已提交
815
    def _calculate_kl_hist_threshold(self):
816
        '''
X
XGZhang 已提交
817
        Calculate the KL or hist threshold of quantized variables.
818
        '''
X
XGZhang 已提交
819 820
        _logger.info("Calculate {} threshold ...".format(self._algo))
        assert self._algo in ["KL", "hist"], "The algo should be KL or hist."
821 822

        # Abs_max threshold for weights
823
        for var_name in self._quantized_weight_var_name:
824
            weight_data = _load_variable_data(self._scope, var_name)
825
            if self._weight_quantize_type == "abs_max":
826
                weight_threshold = float(np.max(np.abs(weight_data)))
827 828
            elif self._weight_quantize_type == "channel_wise_abs_max":
                weight_threshold = []
829
                if self._weight_op_pairs[
830 831 832 833 834 835 836 837
                        var_name] in _channelwise_quant_axis1_ops:
                    for i in range(weight_data.shape[1]):
                        weight_threshold.append(
                            float(np.max(np.abs(weight_data[:, i]))))
                else:
                    for i in range(weight_data.shape[0]):
                        weight_threshold.append(
                            float(np.max(np.abs(weight_data[i]))))
X
XGZhang 已提交
838
            self._quantized_var_threshold[var_name] = weight_threshold
839

840 841
        for var_name in self._quantized_act_var_name:
            hist, hist_edeges = self._sampling_act_histogram[var_name]
X
XGZhang 已提交
842
            if self._algo == "KL":
843
                bin_width = hist_edeges[1] - hist_edeges[0]
X
XGZhang 已提交
844
                self._quantized_var_threshold[var_name] = \
845
                    cal_kl_threshold(hist, bin_width, self._activation_bits)
X
XGZhang 已提交
846 847 848
            elif self._algo == "hist":
                self._quantized_var_threshold[var_name] = \
                    self._get_hist_scaling_factor(hist, hist_edeges)
849 850 851

    def _update_program(self):
        '''
852 853
        Use QuantizationTransformPass and AddQuantDequantPass to insert 
        fake_quantize, fake_dequantize and fake_quant_dequant op. 
X
XGZhang 已提交
854
        Besides, save all threshold to the scale var node.
855
        '''
856
        _logger.info("Update the program ...")
857 858
        graph = IrGraph(core.Graph(self._program.desc), for_test=True)

859
        # use QuantizationTransformPass to insert fake_quant/fake_dequantize op
860 861
        major_quantizable_op_types = []
        for op_type in QuantizationTransformPass._supported_quantizable_op_type:
862
            if op_type in self._quantizable_op_type:
863
                major_quantizable_op_types.append(op_type)
864 865 866
        transform_pass = QuantizationTransformPass(
            scope=self._scope,
            place=self._place,
867 868 869 870
            weight_bits=self._weight_bits,
            activation_bits=self._activation_bits,
            activation_quantize_type=self._activation_quantize_type,
            weight_quantize_type=self._weight_quantize_type,
871
            quantizable_op_type=major_quantizable_op_types)
872 873 874 875 876 877

        for sub_graph in graph.all_sub_graphs():
            # Insert fake_quant/fake_dequantize op must in test graph, so
            # set per graph's _for_test is True.
            sub_graph._for_test = True
            transform_pass.apply(sub_graph)
878 879

        # use AddQuantDequantPass to insert fake_quant_dequant op
880 881
        minor_quantizable_op_types = []
        for op_type in AddQuantDequantPass._supported_quantizable_op_type:
882
            if op_type in self._quantizable_op_type:
883
                minor_quantizable_op_types.append(op_type)
884 885 886
        add_quant_dequant_pass = AddQuantDequantPass(
            scope=self._scope,
            place=self._place,
887
            quantizable_op_type=minor_quantizable_op_types)
888 889 890 891

        for sub_graph in graph.all_sub_graphs():
            sub_graph._for_test = True
            add_quant_dequant_pass.apply(sub_graph)
892

X
XGZhang 已提交
893 894 895
        # save threshold to scale var node
        if self._algo in ["KL", "hist"]:
            scale_dict = self._quantized_var_threshold
896
        else:
X
XGZhang 已提交
897
            scale_dict = self._quantized_threshold
898
        for key, val in scale_dict.items():
899 900 901 902 903
            _set_variable_data(
                self._scope,
                self._place,
                key + ".scale",
                np.array(
904
                    [val], dtype=np.float32))
905 906 907 908 909
            _set_variable_data(
                self._scope,
                self._place,
                key + ".quant_dequant.scale",
                np.array(
910 911 912 913 914 915
                    [val], dtype=np.float32))

        # apply QuantizationFreezePass, and obtain the final quant model
        freeze_pass = QuantizationFreezePass(
            scope=self._scope,
            place=self._place,
X
XGZhang 已提交
916
            bias_correction=self._bias_correction,
917 918 919
            weight_bits=self._weight_bits,
            activation_bits=self._activation_bits,
            weight_quantize_type=self._weight_quantize_type,
920
            quantizable_op_type=major_quantizable_op_types)
921 922 923 924 925

        for sub_graph in graph.all_sub_graphs():
            sub_graph._for_test = True
            freeze_pass.apply(sub_graph)

926 927
        self._program = graph.to_program()

928
    def _save_output_threshold(self):
929
        '''
930
        Save output threshold to the quantized op.
931
        '''
932 933 934 935 936 937 938

        def save_info(op_node, out_var_name, threshold_map, out_info_name,
                      quantized_type):
            assert out_var_name in threshold_map, \
                "The output ({}) of {} node does not have threshold.".format(
                out_var_name, op_node.type)
            op_node._set_attr(out_info_name, threshold_map[var_name])
939
            op_node._set_attr("with_quant_attr", True)
940 941 942 943
            if op_node.type in self._quantizable_op_type:
                op._set_attr("quantization_type", quantized_type)

        def analysis_and_save_info(op_node, out_var_name):
944 945 946
            argname_index = _get_output_name_index(op_node, out_var_name)
            assert argname_index is not None, \
                out_var_name + " is not the output of the op"
947
            if self._algo == "KL":
948
                # For compatibility, we save output threshold by two methods.
X
XGZhang 已提交
949 950
                save_info(op_node, out_var_name, self._quantized_var_threshold,
                          "out_threshold", "post_kl")
951
                save_info(
X
XGZhang 已提交
952
                    op_node, out_var_name, self._quantized_var_threshold,
953 954
                    argname_index[0] + str(argname_index[1]) + "_threshold",
                    "post_kl")
X
XGZhang 已提交
955 956 957 958
            elif self._algo == "hist":
                # For compatibility, we save output threshold by two methods.
                save_info(op_node, out_var_name, self._quantized_var_threshold,
                          "out_threshold", "post_hist")
959
                save_info(
X
XGZhang 已提交
960
                    op_node, out_var_name, self._quantized_var_threshold,
961
                    argname_index[0] + str(argname_index[1]) + "_threshold",
X
XGZhang 已提交
962 963 964 965 966 967 968 969 970
                    "post_hist")

            elif self._algo in ["avg", "abs_max", "mse"]:
                save_info(op_node, out_var_name, self._quantized_threshold,
                          "out_threshold", "post_" + str(self._algo))
                save_info(
                    op_node, out_var_name, self._quantized_threshold,
                    argname_index[0] + str(argname_index[1]) + "_threshold",
                    "post_" + str(self._algo))
971 972 973 974 975 976
            elif self._algo == "min_max":
                save_info(op_node, out_var_name, self._quantized_var_min,
                          "out_min", "post_min_max")
                save_info(op_node, out_var_name, self._quantized_var_max,
                          "out_max", "post_min_max")

977 978 979 980 981 982 983 984 985
        for block_id in range(len(self._program.blocks)):
            for op in self._program.blocks[block_id].ops:
                if op.type in (
                        self._quantizable_op_type + self._out_scale_op_list):
                    out_var_names = _get_op_output_var_names(op)
                    assert len(out_var_names) == 1, "Post training " + \
                        "quantization only support one output for " + op.type
                    for var_name in out_var_names:
                        analysis_and_save_info(op, var_name)
986

987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013
    def _collect_dynamic_quantize_op_threshold(self, target_ops_type):
        """
        Collect and save the weight threshold for dynamic quantize ops,
        such as lstm and gru.
        Args:
            target_ops_type(list): the op type of target ops
        Returns:
            None
        """

        target_ops = []
        for index in range(self._program.num_blocks):
            for op in self._program.block(index).ops:
                if op.type in target_ops_type:
                    target_ops.append(op)

        quantization_type = str("post_" + self._algo).lower()
        persistable_var_names = _all_persistable_var_names(self._program)
        for op in target_ops:
            for var_name in _get_op_input_var_names(op):
                if var_name in persistable_var_names:
                    var_data = _load_variable_data(self._scope, var_name)
                    threshold = float(np.max(np.abs(var_data)))
                    argname, index = _get_input_name_index(op, var_name)
                    op._set_attr(argname + str(index) + "_threshold", threshold)
                    op._set_attr("quantization_type", quantization_type)
                    op._set_attr("bit_length", self._weight_bits)
1014
                    op._set_attr("with_quant_attr", True)
1015

X
XGZhang 已提交
1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031
    def _get_hist_scaling_factor(self, hist, hist_edges):
        '''
        Using the hist method to get the scaling factor.
        '''
        threshold_rate = self._hist_percent
        hist = hist / float(sum(hist))
        hist_sum = 0
        hist_index = 0
        for i in range(len(hist)):
            hist_sum += hist[i]
            if hist_sum >= threshold_rate:
                hist_index = i + 1
                break
        bin_width = hist_edges[1] - hist_edges[0]
        return (hist_index - 0.5) * bin_width

1032 1033 1034

class WeightQuantization(object):
    _supported_quantizable_op_type = ['conv2d', 'depthwise_conv2d', 'mul']
1035
    _supported_weight_quantize_type = ['channel_wise_abs_max', 'abs_max']
1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061

    def __init__(self, model_dir, model_filename=None, params_filename=None):
        '''
        This class quantizes the weight of some ops to reduce the size of model
        or improve the perforemace.

        Args:
            model_dir(str): The path of the fp32 model that will be quantized,
                and the model and params files are under the path.
            model_filename(str, optional): The name of file to load the inference
                program. If it is None, the default filename '__model__' will
                be used. Default is 'None'.
            params_filename(str, optional): The name of file to load all parameters.
                When all parameters were saved in a single binary file, set it
                as the real filename. If parameters were saved in separate files,
                set it as 'None'. Default is 'None'.
        '''
        self._model_dir = model_dir
        self._model_filename = model_filename
        self._params_filename = params_filename

    def quantize_weight_to_int(self,
                               save_model_dir,
                               save_model_filename=None,
                               save_params_filename=None,
                               quantizable_op_type=["conv2d", "mul"],
1062
                               weight_bits=8,
1063 1064
                               weight_quantize_type="channel_wise_abs_max",
                               generate_test_model=False,
1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083
                               threshold_rate=0.0):
        '''
        In order to reduce the size of model, this api quantizes the weight
        of some ops from float32 to int8/16. In the inference stage, the 
        quantized weight will be dequantized to float32 again.
        
        Args:
            save_model_dir(str): The path to save the quantized model.
            save_model_filename(str, optional): The name of file to 
                save the inference program. If it is None, the default 
                filename '__model__' will be used. Default is 'None'.
            save_params_filename(str, optional): The name of file to 
                save all parameters. If it is None, parameters were 
                saved in separate files. If it is not None, all 
                parameters were saved in a single binary file.
            quantizable_op_type(list[str], optional): The list of ops 
                that will be quantized, and the quantized ops should be
                contained in ["conv2d", "depthwise_conv2d", "mul"]. 
                Default is ["conv2d","mul"].
1084 1085
            weight_bits(int, optional): The bits for the quantized weight, 
                and it should be 8 or 16. Default is 8.
1086 1087 1088 1089 1090 1091 1092
            weight_quantize_type(str, optional): quantization type for weights,
                support 'channel_wise_abs_max' and 'abs_max'. Set it as
                'channel_wise_abs_max', the accuracy performs better.
            generate_test_model(bool, optional): If set generate_test_model 
                as True, it saves a fake quantized model, in which the weights 
                are quantized and dequantized. We can use PaddlePaddle to load 
                the fake quantized model and test the accuracy on GPU or CPU.
1093 1094 1095 1096 1097 1098 1099 1100 1101
            threshold_rate(float, optional): This api uses abs_max methd to 
                quantize the weight from float32 to int8/16, and the abs max 
                value is important for quantization diff. When the abs_max 
                value is far away from the center of the numerical distribution, 
                we can set threshold_rate between 1e-6 and 1e-8, so the abs max 
                value will be optimized. Default is 0.0.
        '''
        for op_type in quantizable_op_type:
            assert op_type in self._supported_quantizable_op_type, \
1102
                "Input error:" + op_type + \
1103
                " is not supported for weight quantization."
1104
        assert weight_bits in [8, 16], \
1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122
            "Input error: weight_bits should be 8 or 16."
        assert weight_quantize_type in self._supported_weight_quantize_type, \
            "Input error: weight_quantize_type should in {}".format(
                self._supported_weight_quantize_type)

        quantized_model_dir = os.path.join(save_model_dir, "quantized_model")
        self._quantize_weight_to_int(quantized_model_dir, save_model_filename,
                                     save_params_filename, quantizable_op_type,
                                     weight_bits, weight_quantize_type, False,
                                     threshold_rate)

        if generate_test_model:
            test_model_dir = os.path.join(save_model_dir, "test_model")
            self._quantize_weight_to_int(
                test_model_dir, save_model_filename, save_params_filename,
                quantizable_op_type, weight_bits, weight_quantize_type, True,
                threshold_rate)

1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198
    def convert_weight_to_fp16(self, save_model_dir):
        """
        Convert all presistable vars from fp32 to fp16.
        Note that, this api only changes the data type of variables in
        __params__ file, and the __model__ file remains unchanged. 

        Args:
            save_model_dir(str): The path to save the fp16 model.
        """

        # Load model
        place = core.CPUPlace()
        exe = Executor(place)
        scope = global_scope()
        [infer_program, feed_list, fetch_list] = \
            io.load_inference_model(dirname=self._model_dir,
                                    executor=exe,
                                    model_filename=self._model_filename,
                                    params_filename=self._params_filename)

        # Clone and save fp16 weights
        save_program = framework.Program()
        save_block = save_program.global_block()
        save_var_map = {}

        for var in infer_program.list_vars():
            if (var.type == core.VarDesc.VarType.RAW) or \
                (not var.persistable) or (var.name in ['feed', 'fetch']) \
                or (var.dtype != core.VarDesc.VarType.FP32):
                continue

            #new_var = _clone_var_to_block_(var, save_block)
            new_var = save_block._clone_variable(var)
            if self._params_filename is not None:
                save_var_map[new_var.name] = new_var
            else:
                save_file_path = os.path.join(
                    os.path.normpath(save_model_dir), new_var.name)
                save_block.append_op(
                    type='save',
                    inputs={'X': [new_var]},
                    outputs={},
                    attrs={
                        'file_path': os.path.normpath(save_file_path),
                        'save_as_fp16': True
                    })

        if self._params_filename is not None:
            save_var_list = []
            for name in sorted(save_var_map.keys()):
                save_var_list.append(save_var_map[name])

            saved_params_var = save_block.create_var(
                type=core.VarDesc.VarType.RAW,
                name=unique_name.generate("saved_params"))
            saved_params_var.desc.set_persistable(True)

            save_path = os.path.join(
                os.path.normpath(save_model_dir), self._params_filename)
            save_block.append_op(
                type='save_combine',
                inputs={'X': save_var_list},
                outputs={'Y': saved_params_var},
                attrs={'file_path': save_path,
                       'save_as_fp16': True})

        save_program._sync_with_cpp()
        exe.run(save_program)

        # Copy model
        model_filename = "__model__" if self._model_filename is None \
                    else self._model_filename
        src_model = os.path.join(self._model_dir, model_filename)
        dest_model = os.path.join(save_model_dir, model_filename)
        shutil.copyfile(src_model, dest_model)

1199 1200 1201 1202 1203 1204 1205 1206
    def _quantize_weight_to_int(self, save_model_dir, save_model_filename,
                                save_params_filename, quantizable_op_type,
                                weight_bits, weight_quantize_type, for_test,
                                threshold_rate):
        """
        Generate quantized model or fake quantized model.
        """
        # Load model
1207 1208 1209 1210 1211 1212 1213 1214 1215
        place = core.CPUPlace()
        exe = Executor(place)
        scope = global_scope()
        [program, feed_list, fetch_list] = \
            io.load_inference_model(dirname=self._model_dir,
                                    executor=exe,
                                    model_filename=self._model_filename,
                                    params_filename=self._params_filename)

1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234
        quantized_ops = []
        for index in range(program.num_blocks):
            block = program.block(index)
            for op in block.ops:
                if op.type in quantizable_op_type:
                    quantized_ops.append(op)

        # Quantize weights
        persistable_var_names = _all_persistable_var_names(program)
        for op in quantized_ops:
            for var_name in op.input_arg_names:
                if var_name in persistable_var_names:
                    if weight_quantize_type == "abs_max":
                        self._weight_abs_max_quantization(
                            scope, place, weight_bits, threshold_rate, op,
                            var_name, for_test)
                    elif weight_quantize_type == "channel_wise_abs_max":
                        self._weight_channel_wise_abs_max_quantization(
                            scope, place, weight_bits, op, var_name, for_test)
1235 1236 1237 1238 1239 1240 1241 1242 1243 1244

        io.save_inference_model(
            dirname=save_model_dir,
            feeded_var_names=feed_list,
            target_vars=fetch_list,
            executor=exe,
            main_program=program,
            model_filename=save_model_filename,
            params_filename=save_params_filename)

1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277
    def _weight_abs_max_quantization(self, scope, place, weight_bits,
                                     threshold_rate, op, var_name, for_test):
        '''
        Use abs_max method to quantize weight.
        '''
        quantize_range = (1 << (weight_bits - 1)) - 1
        save_weight_dtype = np.int8 if weight_bits == 8 else np.int16

        # Get quantized scale and weight data
        weight_data = _load_variable_data(scope, var_name)
        if abs(threshold_rate) < 1e-10:
            threshold_value = np.max(np.abs(weight_data))
        else:
            threshold_value = self._calculate_threshold(\
                weight_data, threshold_rate)
            weight_data[weight_data > threshold_value] = threshold_value
            weight_data[weight_data < -threshold_value] = -threshold_value
        scale = threshold_value / quantize_range
        quantized_weight_data = \
            np.around(weight_data / scale).astype(save_weight_dtype)

        # Set weight data
        if not for_test:
            _set_variable_data(scope, place, var_name, quantized_weight_data)
        else:
            dequantized_weight_data = \
                (quantized_weight_data * scale).astype(np.float32)
            _set_variable_data(scope, place, var_name, dequantized_weight_data)

        # Save info
        op._set_attr('quantization_type', 'post_weight_abs_max')
        op._set_attr('quantize_weight_bits', weight_bits)
        op._set_attr(var_name + "_quant_scale", [scale])  # Save as list
1278
        op._set_attr("with_quant_attr", True)
1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319

    def _weight_channel_wise_abs_max_quantization(
            self, scope, place, weight_bits, op, var_name, for_test):
        ''' 
        Use channel_wise_abs_max method to quantize weight.
        '''
        quantize_range = (1 << (weight_bits - 1)) - 1
        save_weight_dtype = np.int8 if weight_bits == 8 else np.int16

        # Get quantized scale and weight data
        weight_data = _load_variable_data(scope, var_name)
        if op.type == "mul":
            scales, quantized_weight_data = \
                self._mul_channel_wise_quantization(weight_data,
                    quantize_range, save_weight_dtype)
        elif op.type in ["conv2d", "depthwise_conv2d"]:
            scales, quantized_weight_data = \
                self._conv_channel_wise_quantization(weight_data,
                    quantize_range, save_weight_dtype)
        else:
            _logger.error(op.type + " is not supported by weight quantization")

        # Set weight data
        if not for_test:
            _set_variable_data(scope, place, var_name, quantized_weight_data)
        else:
            if op.type == "mul":
                dequantized_weight_data = \
                    self._mul_channel_wise_dequantization(quantized_weight_data, scales)
            elif op.type in ["conv2d", "depthwise_conv2d"]:
                dequantized_weight_data = \
                    self._conv_channel_wise_dequantization(quantized_weight_data, scales)
            else:
                _logger.error(op.type +
                              " is not supported by weight quantization")
            _set_variable_data(scope, place, var_name, dequantized_weight_data)

        # Save info
        op._set_attr('quantization_type', 'post_weight_channel_wise_abs_max')
        op._set_attr('quantize_weight_bits', weight_bits)
        op._set_attr(var_name + "_quant_scale", scales)
1320
        op._set_attr("with_quant_attr", True)
1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377

    def _conv_channel_wise_quantization(self, weight_data, quantize_range,
                                        save_weight_dtype):
        '''
        Get channel wise scale for the weights of conv2d and depthwise_conv2d,
        and quantize the weights.
        '''
        scales = []
        quantized_weight_data = np.zeros_like(
            weight_data, dtype=save_weight_dtype)
        channel_num = weight_data.shape[0]
        for i in range(channel_num):
            scale = np.max(np.abs(weight_data[i])) / quantize_range
            scales.append(scale)
            quantized_weight_data[i] = \
                np.around(weight_data[i] / scale).astype(save_weight_dtype)
        return scales, quantized_weight_data

    def _conv_channel_wise_dequantization(self, quantized_weight_data, scales):
        '''
        For conv2d and depthwise_conv2d, dequantize the weights to fp32.
        '''
        dequantized_weight_data = np.zeros_like(
            quantized_weight_data, dtype=np.float32)
        for i in range(len(scales)):
            dequantized_weight_data[i] = \
                (quantized_weight_data[i] * scales[i]).astype(np.float32)
        return dequantized_weight_data

    def _mul_channel_wise_quantization(self, weight_data, quantize_range,
                                       save_weight_dtype):
        '''
        Get channel wise scale for the weights of conv2d and depthwise_conv2d,
        and quantize the weights.
        '''
        scales = []
        quantized_weight_data = np.zeros_like(
            weight_data, dtype=save_weight_dtype)
        channel_num = weight_data.shape[-1]
        for i in range(channel_num):
            scale = np.max(np.abs(weight_data[:, i])) / quantize_range
            scales.append(scale)
            quantized_weight_data[:, i] = \
                np.around(weight_data[:, i] / scale).astype(save_weight_dtype)
        return scales, quantized_weight_data

    def _mul_channel_wise_dequantization(self, quantized_weight_data, scales):
        '''
        For mul, dequantize the weights to fp32.
        '''
        dequantized_weight_data = np.zeros_like(
            quantized_weight_data, dtype=np.float32)
        for i in range(len(scales)):
            dequantized_weight_data[:, i] = \
                (quantized_weight_data[:, i] * scales[i]).astype(np.float32)
        return dequantized_weight_data

1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391
    def _calculate_threshold(self, input, threshold_rate, histogram_bins=5000):
        input_abs = np.abs(input)
        hist, hist_edeges = np.histogram(
            input_abs, bins=histogram_bins, range=(0, np.max(input_abs)))
        hist = hist / float(sum(hist))
        hist_sum = 0
        hist_index = 0
        for i in range(len(hist)):
            hist_sum += hist[i]
            if hist_sum >= 1.0 - threshold_rate:
                hist_index = i + 1
                break
        bin_width = hist_edeges[1] - hist_edeges[0]
        return hist_index * bin_width