Added unit test for QAT FP32 & INT8 comparison (#17814)

* added unit test for QAT FP32 & INT8 comparison

test=develop

* enabled other models and updated filenames

test=develop

* added accuracy check and multiple batch handling

test=develop

* removed quantization_mkldnn_pass.py

test=develop

* cleanup

test=develop

* updated model paths

test=develop

* renamed tests without MKL-DNN

test=develop

* fix reusing mkldnn pool2d primitive

test=develop

* add performance measuring

test=develop

* fix accuracy statistics

test=develop

* removed non-mkldnn tests

test=develop

* added conv2d_depthwise->conv2d mkldnn transformation

test=develop

* format update

test=develop

* fixed creating key for pool2d grad

test=develop

* added pass

* Fix the accuracy issue while using float precision to get the scale.

test=develop

* Fix the format issue when 'X' is not nchw.

test=develop

* removed output comparing and changed number of images

test=develop

* cmake and comment fix

test=develop

* updated acc threshold for QAT comparison tests

test=develop

* added OMP_NUM_THREADS setting

test=develop

* enable all QAT INT8 tests

test=develop

* restored upstream version of a file

test=develop

* modified directory names

test=develop

paddle/fluid/operators/mkldnn/pool_mkldnn_op.cc

@@ -36,7 +36,8 @@ std::string CreateKey(const paddle::framework::ExecutionContext& ctx,
                       const std::vector<int>& ksize,
                       const std::vector<int>& strides,
                       const std::vector<int>& paddings,
-                      const memory::data_type& dt, const std::string& suffix) {
+                      const memory::data_type& dt, const memory::format& fmt,
+                      const std::string& suffix) {
   std::string key;
   key.reserve(platform::MKLDNNHandler::MaxKeyLength);
   platform::MKLDNNHandler::AppendKeyDims(&key, input_dims);
@@ -45,6 +46,7 @@ std::string CreateKey(const paddle::framework::ExecutionContext& ctx,
   platform::MKLDNNHandler::AppendKeyVec(&key, strides);
   platform::MKLDNNHandler::AppendKeyVec(&key, paddings);
   platform::MKLDNNHandler::AppendKey(&key, std::to_string(dt));
+  platform::MKLDNNHandler::AppendKey(&key, std::to_string(fmt));
   platform::MKLDNNHandler::AppendKey(&key, suffix);
   return key;
 }
@@ -115,8 +117,10 @@ class PoolMKLDNNOpKernel : public paddle::framework::OpKernel<T> {
 
     mkldnn::memory::data_type dt =
         paddle::framework::ToMKLDNNDataType(input->type());
-    const std::string key = CreateKey(ctx, src_tz, pooling_type, ksize, strides,
-                                      paddings, dt, ctx.op().Output("Out"));
+    auto fmt = input->format();
+    const std::string key =
+        CreateKey(ctx, src_tz, pooling_type, ksize, strides, paddings, dt, fmt,
+                  ctx.op().Output("Out"));
     const std::string key_pool_p = key + "@pool_p";
     const std::string key_pool_pd = key + "@pool_pd";
     const std::string key_pool_src_mem_p = key + "@pool_src_mem_p";
@@ -294,9 +298,9 @@ class PoolMKLDNNGradOpKernel : public paddle::framework::OpKernel<T> {
 
     // Get an unique name from "argument" name of "Out" variable
     // This name will be used as key when referring info from device context
-    const std::string key =
-        CreateKey(ctx, diff_src_tz, pooling_type, ksize, strides, paddings,
-                  memory::data_type::f32, ctx.op().Input("Out"));
+    const std::string key = CreateKey(ctx, diff_src_tz, pooling_type, ksize,
+                                      strides, paddings, memory::data_type::f32,
+                                      in_x->format(), ctx.op().Input("Out"));
     const std::string key_pool_bwd_p = key + "@pool_bwd_p";
     const std::string key_pool_diff_src_mem_p = key + "@pool_diff_src_mem_p";
     const std::string key_pool_diff_dst_mem_p = key + "@pool_diff_dst_mem_p";

python/paddle/fluid/contrib/slim/tests/CMakeLists.txt

@@ -11,6 +11,18 @@ function(inference_analysis_python_api_int8_test target model_dir data_dir filen
              --batch_size 50)
 endfunction()
 
+function(inference_qat_int8_test target model_dir data_dir test_script use_mkldnn)
+    py_test(${target} SRCS ${test_script}
+            ENVS FLAGS_OMP_NUM_THREADS=${CPU_NUM_THREADS_ON_CI}
+                 OMP_NUM_THREADS=${CPU_NUM_THREADS_ON_CI}
+                 FLAGS_use_mkldnn=${use_mkldnn}
+            ARGS --qat_model ${model_dir}/model
+                 --infer_data ${data_dir}/data.bin
+                 --batch_size 25
+                 --batch_num 2
+                 --acc_diff_threshold 0.1)
+endfunction()
+
 # NOTE: TODOOOOOOOOOOO
 # temporarily disable test_distillation_strategy since it always failed on a specified machine with 4 GPUs
 # Need to figure out the root cause and then add it back
@@ -62,6 +74,74 @@ endif()
 # with MKL-DNN, we remove it here for not repeating test, or not testing on other systems.
 list(REMOVE_ITEM TEST_OPS test_mkldnn_int8_quantization_strategy)
 
+# QAT FP32 & INT8 comparison python api tests
+if(LINUX AND WITH_MKLDNN)
+	set(DATASET_DIR "${INFERENCE_DEMO_INSTALL_DIR}/int8v2")
+	set(QAT_DATA_DIR "${INFERENCE_DEMO_INSTALL_DIR}/int8v2")
+	set(QAT_MODELS_BASE_URL "${INFERENCE_URL}/int8/QAT_models")
+	set(MKLDNN_QAT_TEST_FILE "qat_int8_comparison.py")
+	set(MKLDNN_QAT_TEST_FILE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/${MKLDNN_QAT_TEST_FILE}")
+
+	# ImageNet small dataset
+	# May be already downloaded for INT8v2 unit tests
+	if (NOT EXISTS ${DATASET_DIR})
+		inference_download_and_uncompress(${DATASET_DIR} "${INFERENCE_URL}/int8" "imagenet_val_100_tail.tar.gz")
+	endif()
+
+	# QAT ResNet50
+	set(QAT_RESNET50_MODEL_DIR "${QAT_DATA_DIR}/ResNet50_QAT")
+	if (NOT EXISTS ${QAT_RESNET50_MODEL_DIR})
+		inference_download_and_uncompress(${QAT_RESNET50_MODEL_DIR} "${QAT_MODELS_BASE_URL}" "ResNet50_qat_model.tar.gz" )
+	endif()
+	inference_qat_int8_test(test_qat_int8_resnet50_mkldnn ${QAT_RESNET50_MODEL_DIR} ${DATASET_DIR} ${MKLDNN_QAT_TEST_FILE_PATH} true)
+
+	# QAT ResNet101
+	set(QAT_RESNET101_MODEL_DIR "${QAT_DATA_DIR}/ResNet101_QAT")
+	if (NOT EXISTS ${QAT_RESNET101_MODEL_DIR})
+		inference_download_and_uncompress(${QAT_RESNET101_MODEL_DIR} "${QAT_MODELS_BASE_URL}" "ResNet101_qat_model.tar.gz" )
+	endif()
+	inference_qat_int8_test(test_qat_int8_resnet101_mkldnn ${QAT_RESNET101_MODEL_DIR} ${DATASET_DIR} ${MKLDNN_QAT_TEST_FILE_PATH} true)
+
+	# QAT GoogleNet
+	set(QAT_GOOGLENET_MODEL_DIR "${QAT_DATA_DIR}/GoogleNet_QAT")
+	if (NOT EXISTS ${QAT_GOOGLENET_MODEL_DIR})
+		inference_download_and_uncompress(${QAT_GOOGLENET_MODEL_DIR} "${QAT_MODELS_BASE_URL}" "GoogleNet_qat_model.tar.gz" )
+	endif()
+	inference_qat_int8_test(test_qat_int8_googlenet_mkldnn ${QAT_GOOGLENET_MODEL_DIR} ${DATASET_DIR} ${MKLDNN_QAT_TEST_FILE_PATH} true)
+
+	# QAT MobileNetV1
+	set(QAT_MOBILENETV1_MODEL_DIR "${QAT_DATA_DIR}/MobileNetV1_QAT")
+	if (NOT EXISTS ${QAT_MOBILENETV1_MODEL_DIR})
+		inference_download_and_uncompress(${QAT_MOBILENETV1_MODEL_DIR} "${QAT_MODELS_BASE_URL}" "MobileNetV1_qat_model.tar.gz" )
+	endif()
+	inference_qat_int8_test(test_qat_int8_mobilenetv1_mkldnn ${QAT_MOBILENETV1_MODEL_DIR} ${DATASET_DIR} ${MKLDNN_QAT_TEST_FILE_PATH} true)
+
+	# QAT MobileNetV2
+	set(QAT_MOBILENETV2_MODEL_DIR "${QAT_DATA_DIR}/MobileNetV2_QAT")
+	if (NOT EXISTS ${QAT_MOBILENETV2_MODEL_DIR})
+		inference_download_and_uncompress(${QAT_MOBILENETV2_MODEL_DIR} "${QAT_MODELS_BASE_URL}" "MobileNetV2_qat_model.tar.gz" )
+	endif()
+	inference_qat_int8_test(test_qat_int8_mobilenetv2_mkldnn ${QAT_MOBILENETV2_MODEL_DIR} ${DATASET_DIR} ${MKLDNN_QAT_TEST_FILE_PATH} true)
+
+	# QAT VGG16
+	set(QAT_VGG16_MODEL_DIR "${QAT_DATA_DIR}/VGG16_QAT")
+	if (NOT EXISTS ${QAT_VGG16_MODEL_DIR})
+		inference_download_and_uncompress(${QAT_VGG16_MODEL_DIR} "${QAT_MODELS_BASE_URL}" "VGG16_qat_model.tar.gz" )
+	endif()
+	inference_qat_int8_test(test_qat_int8_vgg16_mkldnn ${QAT_VGG16_MODEL_DIR} ${DATASET_DIR} ${MKLDNN_QAT_TEST_FILE_PATH} true)
+
+	# QAT VGG19
+	set(QAT_VGG19_MODEL_DIR "${QAT_DATA_DIR}/VGG19_QAT")
+	if (NOT EXISTS ${QAT_VGG19_MODEL_DIR})
+		inference_download_and_uncompress(${QAT_VGG19_MODEL_DIR} "${QAT_MODELS_BASE_URL}" "VGG19_qat_model.tar.gz" )
+	endif()
+	inference_qat_int8_test(test_qat_int8_vgg19_mkldnn ${QAT_VGG19_MODEL_DIR} ${DATASET_DIR} ${MKLDNN_QAT_TEST_FILE_PATH} true)
+endif()
+
+# Since the test for QAT FP32 & INT8 comparison supports only testing on Linux 
+# with MKL-DNN, we remove it here to not test it on other systems.
+list(REMOVE_ITEM TEST_OPS qat_int8_comparison.py)
+
 foreach(src ${TEST_OPS})
     py_test(${src} SRCS ${src}.py)
 endforeach()

python/paddle/fluid/contrib/slim/tests/qat_int8_comparison.py

+#   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.
+
+import unittest
+import os
+import sys
+import argparse
+import logging
+import struct
+import six
+import numpy as np
+import time
+import paddle
+import paddle.fluid as fluid
+from paddle.fluid.framework import IrGraph
+from paddle.fluid.contrib.slim.quantization import TransformForMkldnnPass
+from paddle.fluid import core
+
+logging.basicConfig(format='%(asctime)s-%(levelname)s: %(message)s')
+_logger = logging.getLogger(__name__)
+_logger.setLevel(logging.INFO)
+
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--batch_size', type=int, default=1, help='Batch size.')
+    parser.add_argument(
+        '--skip_batch_num',
+        type=int,
+        default=0,
+        help='Number of the first minibatches to skip in performance statistics.'
+    )
+    parser.add_argument(
+        '--qat_model', type=str, default='', help='A path to a QAT model.')
+    parser.add_argument('--infer_data', type=str, default='', help='Data file.')
+    parser.add_argument(
+        '--batch_num',
+        type=int,
+        default=1,
+        help='Number of batches to process. 0 or less means all.')
+    parser.add_argument(
+        '--acc_diff_threshold',
+        type=float,
+        default=0.01,
+        help='Accepted accuracy difference threshold.')
+
+    test_args, args = parser.parse_known_args(namespace=unittest)
+
+    return test_args, sys.argv[:1] + args
+
+
+class TestQatInt8Comparison(unittest.TestCase):
+    """
+    Test for accuracy comparison of QAT FP32 and INT8 inference.
+    """
+
+    def _reader_creator(self, data_file='data.bin'):
+        def reader():
+            with open(data_file, 'rb') as fp:
+                num = fp.read(8)
+                num = struct.unpack('q', num)[0]
+                imgs_offset = 8
+                img_ch = 3
+                img_w = 224
+                img_h = 224
+                img_pixel_size = 4
+                img_size = img_ch * img_h * img_w * img_pixel_size
+                label_size = 8
+                labels_offset = imgs_offset + num * img_size
+
+                step = 0
+                while step < num:
+                    fp.seek(imgs_offset + img_size * step)
+                    img = fp.read(img_size)
+                    img = struct.unpack_from('{}f'.format(img_ch * img_w *
+                                                          img_h), img)
+                    img = np.array(img)
+                    img.shape = (img_ch, img_w, img_h)
+                    fp.seek(labels_offset + label_size * step)
+                    label = fp.read(label_size)
+                    label = struct.unpack('q', label)[0]
+                    yield img, int(label)
+                    step += 1
+
+        return reader
+
+    def _get_batch_accuracy(self, batch_output=None, labels=None):
+        total = 0
+        correct = 0
+        correct_5 = 0
+        for n, result in enumerate(batch_output):
+            index = result.argsort()
+            top_1_index = index[-1]
+            top_5_index = index[-5:]
+            total += 1
+            if top_1_index == labels[n]:
+                correct += 1
+            if labels[n] in top_5_index:
+                correct_5 += 1
+        acc1 = float(correct) / float(total)
+        acc5 = float(correct_5) / float(total)
+        return acc1, acc5
+
+    def _prepare_for_fp32_mkldnn(self, graph):
+        ops = graph.all_op_nodes()
+        for op_node in ops:
+            name = op_node.name()
+            if name in ['depthwise_conv2d']:
+                input_var_node = graph._find_node_by_name(
+                    op_node.inputs, op_node.input("Input")[0])
+                weight_var_node = graph._find_node_by_name(
+                    op_node.inputs, op_node.input("Filter")[0])
+                output_var_node = graph._find_node_by_name(
+                    graph.all_var_nodes(), op_node.output("Output")[0])
+                attrs = {
+                    name: op_node.op().attr(name)
+                    for name in op_node.op().attr_names()
+                }
+
+                conv_op_node = graph.create_op_node(
+                    op_type='conv2d',
+                    attrs=attrs,
+                    inputs={
+                        'Input': input_var_node,
+                        'Filter': weight_var_node
+                    },
+                    outputs={'Output': output_var_node})
+
+                graph.link_to(input_var_node, conv_op_node)
+                graph.link_to(weight_var_node, conv_op_node)
+                graph.link_to(conv_op_node, output_var_node)
+                graph.safe_remove_nodes(op_node)
+
+        return graph
+
+    def _predict(self,
+                 test_reader=None,
+                 model_path=None,
+                 batch_num=1,
+                 skip_batch_num=0,
+                 transform_to_int8=False):
+        place = fluid.CPUPlace()
+        exe = fluid.Executor(place)
+        inference_scope = fluid.executor.global_scope()
+        with fluid.scope_guard(inference_scope):
+            if os.path.exists(os.path.join(model_path, '__model__')):
+                [inference_program, feed_target_names,
+                 fetch_targets] = fluid.io.load_inference_model(model_path, exe)
+            else:
+                [inference_program, feed_target_names,
+                 fetch_targets] = fluid.io.load_inference_model(
+                     model_path, exe, 'model', 'params')
+
+            graph = IrGraph(core.Graph(inference_program.desc), for_test=True)
+            if (transform_to_int8):
+                mkldnn_int8_pass = TransformForMkldnnPass(
+                    scope=inference_scope, place=place)
+                mkldnn_int8_pass.apply(graph)
+            else:
+                graph = self._prepare_for_fp32_mkldnn(graph)
+            inference_program = graph.to_program()
+
+            dshape = [3, 224, 224]
+            outputs = []
+            infer_accs1 = []
+            infer_accs5 = []
+            fpses = []
+            batch_times = []
+            total_samples = 0
+            top1 = 0.0
+            top5 = 0.0
+            iters = 0
+            infer_start_time = time.time()
+            for data in test_reader():
+                if batch_num > 0 and iters >= batch_num:
+                    break
+                if iters == skip_batch_num:
+                    total_samples = 0
+                    infer_start_time = time.time()
+                if six.PY2:
+                    images = map(lambda x: x[0].reshape(dshape), data)
+                if six.PY3:
+                    images = list(map(lambda x: x[0].reshape(dshape), data))
+                images = np.array(images).astype('float32')
+                labels = np.array([x[1] for x in data]).astype('int64')
+
+                start = time.time()
+                out = exe.run(inference_program,
+                              feed={feed_target_names[0]: images},
+                              fetch_list=fetch_targets)
+                batch_time = time.time() - start
+                outputs.append(out[0])
+                batch_acc1, batch_acc5 = self._get_batch_accuracy(out[0],
+                                                                  labels)
+                infer_accs1.append(batch_acc1)
+                infer_accs5.append(batch_acc5)
+                samples = len(data)
+                total_samples += samples
+                batch_times.append(batch_time)
+                fps = samples / batch_time
+                fpses.append(fps)
+                iters += 1
+                appx = ' (warm-up)' if iters <= skip_batch_num else ''
+                _logger.info(
+                    'batch {0}{5}, acc1: {1:.4f}, acc5: {2:.4f}, '
+                    'batch latency: {3:.4f} s, batch fps: {4:.2f}'.format(
+                        iters, batch_acc1, batch_acc5, batch_time, fps, appx))
+
+            # Postprocess benchmark data
+            latencies = batch_times[skip_batch_num:]
+            latency_avg = np.average(latencies)
+            fpses = fpses[skip_batch_num:]
+            fps_avg = np.average(fpses)
+            infer_total_time = time.time() - infer_start_time
+            acc1_avg = np.mean(infer_accs1)
+            acc5_avg = np.mean(infer_accs5)
+            _logger.info('Total inference run time: {:.2f} s'.format(
+                infer_total_time))
+
+            return outputs, acc1_avg, acc5_avg, fps_avg, latency_avg
+
+    def _compare_accuracy(self, fp32_acc1, fp32_acc5, int8_acc1, int8_acc5,
+                          threshold):
+        _logger.info('Accepted acc1 diff threshold: {0}'.format(threshold))
+        _logger.info('FP32: avg acc1: {0:.4f}, avg acc5: {1:.4f}'.format(
+            fp32_acc1, fp32_acc5))
+        _logger.info('INT8: avg acc1: {0:.4f}, avg acc5: {1:.4f}'.format(
+            int8_acc1, int8_acc5))
+        assert fp32_acc1 > 0.0
+        assert int8_acc1 > 0.0
+        assert fp32_acc1 - int8_acc1 <= threshold
+
+    def test_graph_transformation(self):
+        if not fluid.core.is_compiled_with_mkldnn():
+            return
+
+        qat_model_path = test_case_args.qat_model
+        data_path = test_case_args.infer_data
+        batch_size = test_case_args.batch_size
+        batch_num = test_case_args.batch_num
+        skip_batch_num = test_case_args.skip_batch_num
+        acc_diff_threshold = test_case_args.acc_diff_threshold
+
+        _logger.info('QAT FP32 & INT8 prediction run.')
+        _logger.info('QAT model: {0}'.format(qat_model_path))
+        _logger.info('Dataset: {0}'.format(data_path))
+        _logger.info('Batch size: {0}'.format(batch_size))
+        _logger.info('Batch number: {0}'.format(batch_num))
+        _logger.info('Accuracy diff threshold: {0}. '
+                     '(condition: (fp32_acc - int8_acc) <= threshold)'
+                     .format(acc_diff_threshold))
+
+        _logger.info('--- QAT FP32 prediction start ---')
+        val_reader = paddle.batch(
+            self._reader_creator(data_path), batch_size=batch_size)
+        fp32_output, fp32_acc1, fp32_acc5, fp32_fps, fp32_lat = self._predict(
+            val_reader,
+            qat_model_path,
+            batch_num,
+            skip_batch_num,
+            transform_to_int8=False)
+
+        _logger.info('--- QAT INT8 prediction start ---')
+        val_reader = paddle.batch(
+            self._reader_creator(data_path), batch_size=batch_size)
+        int8_output, int8_acc1, int8_acc5, int8_fps, int8_lat = self._predict(
+            val_reader,
+            qat_model_path,
+            batch_num,
+            skip_batch_num,
+            transform_to_int8=True)
+
+        _logger.info('--- Performance summary ---')
+        _logger.info('FP32: avg fps: {0:.2f}, avg latency: {1:.4f} s'.format(
+            fp32_fps, fp32_lat))
+        _logger.info('INT8: avg fps: {0:.2f}, avg latency: {1:.4f} s'.format(
+            int8_fps, int8_lat))
+
+        _logger.info('--- Comparing accuracy ---')
+        self._compare_accuracy(fp32_acc1, fp32_acc5, int8_acc1, int8_acc5,
+                               acc_diff_threshold)
+
+
+if __name__ == '__main__':
+    global test_case_args
+    test_case_args, remaining_args = parse_args()
+    unittest.main(argv=remaining_args)