test_quantization_pass.py 15.8 KB
Newer Older
W
WangZhen 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
#   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 unittest
import random
import numpy as np
import paddle.fluid as fluid
import six
W
WangZhen 已提交
20
import paddle
21
from paddle.fluid.framework import IrGraph
22
from paddle.fluid.contrib.slim.quantization import QuantizationTransformPass
W
WangZhen 已提交
23
from paddle.fluid.contrib.slim.quantization import QuantizationFreezePass
24 25
from paddle.fluid.contrib.slim.quantization import ConvertToInt8Pass
from paddle.fluid.contrib.slim.quantization import TransformForMobilePass
W
WangZhen 已提交
26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70
from paddle.fluid import core


def linear_fc(num):
    data = fluid.layers.data(name='image', shape=[1, 32, 32], dtype='float32')
    label = fluid.layers.data(name='label', shape=[1], dtype='int64')
    hidden = data
    for _ in six.moves.xrange(num):
        hidden = fluid.layers.fc(hidden, size=128, act='relu')
    loss = fluid.layers.cross_entropy(input=hidden, label=label)
    loss = fluid.layers.mean(loss)
    return loss


def residual_block(num):
    def conv_bn_layer(input,
                      ch_out,
                      filter_size,
                      stride,
                      padding,
                      act='relu',
                      bias_attr=False):
        tmp = fluid.layers.conv2d(
            input=input,
            filter_size=filter_size,
            num_filters=ch_out,
            stride=stride,
            padding=padding,
            act=None,
            bias_attr=bias_attr)
        return fluid.layers.batch_norm(input=tmp, act=act)

    data = fluid.layers.data(name='image', shape=[1, 32, 32], dtype='float32')
    label = fluid.layers.data(name='label', shape=[1], dtype='int64')
    hidden = data
    for _ in six.moves.xrange(num):
        conv = conv_bn_layer(hidden, 16, 3, 1, 1, act=None, bias_attr=True)
        short = conv_bn_layer(hidden, 16, 1, 1, 0, act=None)
        hidden = fluid.layers.elementwise_add(x=conv, y=short, act='relu')
    fc = fluid.layers.fc(input=hidden, size=10)
    loss = fluid.layers.cross_entropy(input=fc, label=label)
    loss = fluid.layers.mean(loss)
    return loss


W
WangZhen 已提交
71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92
def conv_net(img, label):
    conv_pool_1 = fluid.nets.simple_img_conv_pool(
        input=img,
        filter_size=5,
        num_filters=20,
        pool_size=2,
        pool_stride=2,
        act="relu")
    conv_pool_1 = fluid.layers.batch_norm(conv_pool_1)
    conv_pool_2 = fluid.nets.simple_img_conv_pool(
        input=conv_pool_1,
        filter_size=5,
        num_filters=50,
        pool_size=2,
        pool_stride=2,
        act="relu")
    prediction = fluid.layers.fc(input=conv_pool_2, size=10, act='softmax')
    loss = fluid.layers.cross_entropy(input=prediction, label=label)
    avg_loss = fluid.layers.mean(loss)
    return avg_loss


93
class TestQuantizationTransformPass(unittest.TestCase):
W
WangZhen 已提交
94 95 96 97 98 99
    def setUp(self):
        self.quantizable_op_and_inputs = {
            'conv2d': ['Input', 'Filter'],
            'depthwise_conv2d': ['Input', 'Filter'],
            'mul': ['X', 'Y']
        }
100
        self.quantizable_grad_op_inputs = {
W
WangZhen 已提交
101 102 103 104 105
            'conv2d_grad': ['Input', 'Filter'],
            'depthwise_conv2d_grad': ['Input', 'Filter'],
            'mul_grad': ['X', 'Y']
        }

106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125
    def check_program(self, transform_pass, program):
        quantized_ops = set()
        for block in program.blocks:
            for op in block.ops:
                # check forward
                if op.type in self.quantizable_op_and_inputs:
                    for arg_name in op.input_arg_names:
                        self.assertTrue(
                            arg_name.endswith('.quantized.dequantized'))
                        quantized_ops.add(arg_name)

            for op in block.ops:
                # check backward
                if op.type in self.quantizable_grad_op_inputs:
                    for pname in self.quantizable_grad_op_inputs[op.type]:
                        arg_name = op.input(pname)[0]
                        self.assertTrue(
                            arg_name.endswith('.quantized.dequantized'))
                        self.assertTrue(arg_name in quantized_ops)

Z
Zhen Wang 已提交
126
    def linear_fc_quant(self, quant_type, for_ci=False):
W
WangZhen 已提交
127 128 129 130 131 132
        main = fluid.Program()
        startup = fluid.Program()
        with fluid.program_guard(main, startup):
            loss = linear_fc(3)
            opt = fluid.optimizer.Adam(learning_rate=0.001)
            opt.minimize(loss)
133 134
        place = fluid.CPUPlace()
        exe = fluid.Executor(place)
135
        graph = IrGraph(core.Graph(main.desc), for_test=False)
136 137
        transform_pass = QuantizationTransformPass(
            scope=fluid.global_scope(),
138
            place=place,
139 140
            activation_quantize_type=quant_type)
        transform_pass.apply(graph)
Z
Zhen Wang 已提交
141
        if not for_ci:
Z
Zhen Wang 已提交
142 143 144 145 146
            marked_nodes = set()
            for op in graph.all_op_nodes():
                if op.name().find('quantize') > -1:
                    marked_nodes.add(op)
            graph.draw('.', 'quantize_fc_' + quant_type, marked_nodes)
147 148
        program = graph.to_program()
        self.check_program(transform_pass, program)
149
        val_graph = IrGraph(core.Graph(program.desc), for_test=False)
Z
Zhen Wang 已提交
150
        if not for_ci:
Z
Zhen Wang 已提交
151 152 153 154 155
            val_marked_nodes = set()
            for op in val_graph.all_op_nodes():
                if op.name().find('quantize') > -1:
                    val_marked_nodes.add(op)
            val_graph.draw('.', 'val_fc_' + quant_type, val_marked_nodes)
W
WangZhen 已提交
156

157
    def test_linear_fc_quant_abs_max(self):
Z
Zhen Wang 已提交
158
        self.linear_fc_quant('abs_max', for_ci=True)
W
WangZhen 已提交
159

160
    def test_linear_fc_quant_range_abs_max(self):
Z
Zhen Wang 已提交
161
        self.linear_fc_quant('range_abs_max', for_ci=True)
W
WangZhen 已提交
162

Z
Zhen Wang 已提交
163
    def residual_block_quant(self, quant_type, for_ci=False):
W
WangZhen 已提交
164 165 166 167 168 169
        main = fluid.Program()
        startup = fluid.Program()
        with fluid.program_guard(main, startup):
            loss = residual_block(2)
            opt = fluid.optimizer.Adam(learning_rate=0.001)
            opt.minimize(loss)
170 171
        place = fluid.CPUPlace()
        exe = fluid.Executor(place)
172
        graph = IrGraph(core.Graph(main.desc), for_test=False)
173 174
        transform_pass = QuantizationTransformPass(
            scope=fluid.global_scope(),
175
            place=place,
176 177
            activation_quantize_type=quant_type)
        transform_pass.apply(graph)
Z
Zhen Wang 已提交
178
        if not for_ci:
Z
Zhen Wang 已提交
179 180 181 182 183
            marked_nodes = set()
            for op in graph.all_op_nodes():
                if op.name().find('quantize') > -1:
                    marked_nodes.add(op)
            graph.draw('.', 'quantize_residual_' + quant_type, marked_nodes)
184 185
        program = graph.to_program()
        self.check_program(transform_pass, program)
186
        val_graph = IrGraph(core.Graph(program.desc), for_test=False)
Z
Zhen Wang 已提交
187
        if not for_ci:
Z
Zhen Wang 已提交
188 189 190 191 192
            val_marked_nodes = set()
            for op in val_graph.all_op_nodes():
                if op.name().find('quantize') > -1:
                    val_marked_nodes.add(op)
            val_graph.draw('.', 'val_residual_' + quant_type, val_marked_nodes)
W
WangZhen 已提交
193

194
    def test_residual_block_abs_max(self):
Z
Zhen Wang 已提交
195
        self.residual_block_quant('abs_max', for_ci=True)
W
WangZhen 已提交
196

197
    def test_residual_block_range_abs_max(self):
Z
Zhen Wang 已提交
198
        self.residual_block_quant('range_abs_max', for_ci=True)
W
WangZhen 已提交
199 200


W
WangZhen 已提交
201
class TestQuantizationFreezePass(unittest.TestCase):
Z
Zhen Wang 已提交
202
    def freeze_graph(self, use_cuda, seed, quant_type, for_ci=False):
W
WangZhen 已提交
203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227
        def build_program(main, startup, is_test):
            main.random_seed = seed
            startup.random_seed = seed
            with fluid.unique_name.guard():
                with fluid.program_guard(main, startup):
                    img = fluid.layers.data(
                        name='image', shape=[1, 28, 28], dtype='float32')
                    label = fluid.layers.data(
                        name='label', shape=[1], dtype='int64')
                    loss = conv_net(img, label)
                    if not is_test:
                        opt = fluid.optimizer.Adam(learning_rate=0.001)
                        opt.minimize(loss)
            return [img, label], loss

        random.seed(0)
        np.random.seed(0)

        main = fluid.Program()
        startup = fluid.Program()
        test_program = fluid.Program()
        feeds, loss = build_program(main, startup, False)
        build_program(test_program, startup, True)
        test_program = test_program.clone(for_test=True)
        main_graph = IrGraph(core.Graph(main.desc), for_test=False)
W
WangZhen 已提交
228
        test_graph = IrGraph(core.Graph(test_program.desc), for_test=True)
W
WangZhen 已提交
229 230 231

        place = fluid.CUDAPlace(0) if use_cuda else fluid.CPUPlace()
        exe = fluid.Executor(place)
W
WangZhen 已提交
232 233 234
        scope = fluid.Scope()
        with fluid.scope_guard(scope):
            exe.run(startup)
W
WangZhen 已提交
235
        transform_pass = QuantizationTransformPass(
236
            scope=scope, place=place, activation_quantize_type=quant_type)
W
WangZhen 已提交
237 238
        transform_pass.apply(main_graph)
        transform_pass.apply(test_graph)
239
        dev_name = '_gpu_' if use_cuda else '_cpu_'
Z
Zhen Wang 已提交
240
        if not for_ci:
Z
Zhen Wang 已提交
241 242 243 244 245 246 247 248 249 250
            marked_nodes = set()
            for op in main_graph.all_op_nodes():
                if op.name().find('quantize') > -1:
                    marked_nodes.add(op)
            main_graph.draw('.', 'main' + dev_name + quant_type, marked_nodes)
            marked_nodes = set()
            for op in test_graph.all_op_nodes():
                if op.name().find('quantize') > -1:
                    marked_nodes.add(op)
            test_graph.draw('.', 'test' + dev_name + quant_type, marked_nodes)
W
WangZhen 已提交
251

252 253
        quantized_main_program = main_graph.to_program()
        quantized_test_program = test_graph.to_program()
254
        iters = 5
255
        batch_size = 8
W
WangZhen 已提交
256 257 258 259 260 261 262 263

        train_reader = paddle.batch(
            paddle.reader.shuffle(
                paddle.dataset.mnist.train(), buf_size=500),
            batch_size=batch_size)
        test_reader = paddle.batch(
            paddle.dataset.mnist.test(), batch_size=batch_size)
        feeder = fluid.DataFeeder(feed_list=feeds, place=place)
W
WangZhen 已提交
264
        with fluid.scope_guard(scope):
W
WangZhen 已提交
265 266
            for _ in range(iters):
                data = next(train_reader())
267 268 269
                loss_v = exe.run(program=quantized_main_program,
                                 feed=feeder.feed(data),
                                 fetch_list=[loss])
Z
Zhen Wang 已提交
270
                if not for_ci:
Z
Zhen Wang 已提交
271 272
                    print('{}: {}'.format('loss' + dev_name + quant_type,
                                          loss_v))
W
WangZhen 已提交
273

274 275 276 277 278 279 280 281 282 283 284 285 286
        test_data = next(test_reader())
        with fluid.program_guard(quantized_test_program):
            w_var = fluid.framework._get_var('conv2d_1.w_0.quantized',
                                             quantized_test_program)
        # Testing
        with fluid.scope_guard(scope):
            test_loss1, w_quant = exe.run(program=quantized_test_program,
                                          feed=feeder.feed(test_data),
                                          fetch_list=[loss, w_var])

        # Freeze graph for inference, but the weight of fc/conv is still float type.
        freeze_pass = QuantizationFreezePass(scope=scope, place=place)
        freeze_pass.apply(test_graph)
Z
Zhen Wang 已提交
287
        if not for_ci:
Z
Zhen Wang 已提交
288 289 290 291 292 293
            marked_nodes = set()
            for op in test_graph.all_op_nodes():
                if op.name().find('quantize') > -1:
                    marked_nodes.add(op)
            test_graph.draw('.', 'test_freeze' + dev_name + quant_type,
                            marked_nodes)
W
WangZhen 已提交
294

295 296 297 298 299 300
        server_program = test_graph.to_program()
        with fluid.scope_guard(scope):
            test_loss2, = exe.run(program=server_program,
                                  feed=feeder.feed(test_data),
                                  fetch_list=[loss])
        self.assertAlmostEqual(test_loss1, test_loss2, delta=5e-3)
Z
Zhen Wang 已提交
301
        if not for_ci:
Z
Zhen Wang 已提交
302 303 304 305
            print('{}: {}'.format('test_loss1' + dev_name + quant_type,
                                  test_loss1))
            print('{}: {}'.format('test_loss2' + dev_name + quant_type,
                                  test_loss2))
306 307
        w_freeze = np.array(scope.find_var('conv2d_1.w_0').get_tensor())
        # Maybe failed, this is due to the calculation precision
308
        # self.assertAlmostEqual(np.sum(w_freeze), np.sum(w_quant))
Z
Zhen Wang 已提交
309
        if not for_ci:
Z
Zhen Wang 已提交
310 311 312 313
            print('{}: {}'.format('w_freeze' + dev_name + quant_type,
                                  np.sum(w_freeze)))
            print('{}: {}'.format('w_quant' + dev_name + quant_type,
                                  np.sum(w_quant)))
314 315 316 317

        # Convert parameter to 8-bit.
        convert_int8_pass = ConvertToInt8Pass(scope=scope, place=place)
        convert_int8_pass.apply(test_graph)
Z
Zhen Wang 已提交
318
        if not for_ci:
Z
Zhen Wang 已提交
319 320 321 322 323 324
            marked_nodes = set()
            for op in test_graph.all_op_nodes():
                if op.name().find('quantize') > -1:
                    marked_nodes.add(op)
            test_graph.draw('.', 'test_int8' + dev_name + quant_type,
                            marked_nodes)
325 326 327 328 329 330 331 332 333 334 335 336 337
        server_program_int8 = test_graph.to_program()
        # Save the 8-bit parameter and model file.
        with fluid.scope_guard(scope):
            fluid.io.save_inference_model('server_int8' + dev_name + quant_type,
                                          ['image', 'label'], [loss], exe,
                                          server_program_int8)
            # Test whether the 8-bit parameter and model file can be loaded successfully.
            [infer, feed, fetch] = fluid.io.load_inference_model(
                'server_int8' + dev_name + quant_type, exe)
        # Check the loaded 8-bit weight.
        w_8bit = np.array(scope.find_var('conv2d_1.w_0.int8').get_tensor())
        self.assertEqual(w_8bit.dtype, np.int8)
        self.assertEqual(np.sum(w_8bit), np.sum(w_freeze))
Z
Zhen Wang 已提交
338
        if not for_ci:
Z
Zhen Wang 已提交
339 340 341 342
            print('{}: {}'.format('w_8bit' + dev_name + quant_type,
                                  np.sum(w_8bit)))
            print('{}: {}'.format('w_freeze' + dev_name + quant_type,
                                  np.sum(w_freeze)))
343 344 345

        mobile_pass = TransformForMobilePass()
        mobile_pass.apply(test_graph)
Z
Zhen Wang 已提交
346
        if not for_ci:
Z
Zhen Wang 已提交
347 348 349 350 351 352
            marked_nodes = set()
            for op in test_graph.all_op_nodes():
                if op.name().find('quantize') > -1:
                    marked_nodes.add(op)
            test_graph.draw('.', 'test_mobile' + dev_name + quant_type,
                            marked_nodes)
353 354 355 356 357 358

        mobile_program = test_graph.to_program()
        with fluid.scope_guard(scope):
            fluid.io.save_inference_model('mobile_int8' + dev_name + quant_type,
                                          ['image', 'label'], [loss], exe,
                                          mobile_program)
W
WangZhen 已提交
359

360
    def test_freeze_graph_cuda_dynamic(self):
W
WangZhen 已提交
361 362
        if fluid.core.is_compiled_with_cuda():
            with fluid.unique_name.guard():
Z
Zhen Wang 已提交
363
                self.freeze_graph(
Z
Zhen Wang 已提交
364
                    True, seed=1, quant_type='abs_max', for_ci=True)
W
WangZhen 已提交
365

366
    def test_freeze_graph_cpu_dynamic(self):
W
WangZhen 已提交
367
        with fluid.unique_name.guard():
Z
Zhen Wang 已提交
368
            self.freeze_graph(False, seed=2, quant_type='abs_max', for_ci=True)
W
WangZhen 已提交
369

370
    def test_freeze_graph_cuda_static(self):
W
WangZhen 已提交
371 372
        if fluid.core.is_compiled_with_cuda():
            with fluid.unique_name.guard():
Z
Zhen Wang 已提交
373
                self.freeze_graph(
Z
Zhen Wang 已提交
374
                    True, seed=1, quant_type='range_abs_max', for_ci=True)
W
WangZhen 已提交
375

376
    def test_freeze_graph_cpu_static(self):
W
WangZhen 已提交
377
        with fluid.unique_name.guard():
Z
Zhen Wang 已提交
378
            self.freeze_graph(
Z
Zhen Wang 已提交
379
                False, seed=2, quant_type='range_abs_max', for_ci=True)
W
WangZhen 已提交
380 381


W
WangZhen 已提交
382 383
if __name__ == '__main__':
    unittest.main()