diff --git a/python/paddle/fluid/contrib/slim/tests/test_imperative_out_scale.py b/python/paddle/fluid/contrib/slim/tests/test_imperative_out_scale.py index 7b9cd7958b2d3d0704a3770156d86f9cae44592a..dd10c6eb7b7c315d423ce7aef7f6e9d929f4ced5 100644 --- a/python/paddle/fluid/contrib/slim/tests/test_imperative_out_scale.py +++ b/python/paddle/fluid/contrib/slim/tests/test_imperative_out_scale.py @@ -20,6 +20,7 @@ import random import unittest import logging import warnings +import tempfile import paddle import paddle.fluid as fluid @@ -122,6 +123,16 @@ class ImperativeLenet(fluid.dygraph.Layer): class TestImperativeOutSclae(unittest.TestCase): + def setUp(self): + self.root_path = tempfile.TemporaryDirectory() + self.param_save_path = os.path.join(self.root_path.name, + "lenet.pdparams") + self.save_path = os.path.join(self.root_path.name, + "lenet_dynamic_outscale_infer_model") + + def tearDown(self): + self.root_path.cleanup() + def func_out_scale_acc(self): seed = 1000 lr = 0.001 @@ -148,46 +159,17 @@ class TestImperativeOutSclae(unittest.TestCase): loss_list = train_lenet(lenet, reader, adam) lenet.eval() - param_save_path = "test_save_quantized_model/lenet.pdparams" save_dict = lenet.state_dict() - paddle.save(save_dict, param_save_path) - - save_path = "./dynamic_outscale_infer_model/lenet" - imperative_out_scale.save_quantized_model( - layer=lenet, - path=save_path, - input_spec=[ - paddle.static.InputSpec( - shape=[None, 1, 28, 28], dtype='float32') - ]) + paddle.save(save_dict, self.param_save_path) for i in range(len(loss_list) - 1): self.assertTrue( loss_list[i] > loss_list[i + 1], msg='Failed to do the imperative qat.') - def test_out_scale_acc(self): - with _test_eager_guard(): - self.func_out_scale_acc() - self.func_out_scale_acc() - - -class TestSaveQuanztizedModelFromCheckPoint(unittest.TestCase): - def func_save_quantized_model(self): - lr = 0.001 - - load_param_path = "test_save_quantized_model/lenet.pdparams" - save_path = "./dynamic_outscale_infer_model_from_checkpoint/lenet" - - weight_quantize_type = 'abs_max' - activation_quantize_type = 'moving_average_abs_max' - imperative_out_scale = ImperativeQuantAware( - weight_quantize_type=weight_quantize_type, - activation_quantize_type=activation_quantize_type) - with fluid.dygraph.guard(): lenet = ImperativeLenet() - load_dict = paddle.load(load_param_path) + load_dict = paddle.load(self.param_save_path) imperative_out_scale.quantize(lenet) lenet.set_dict(load_dict) @@ -200,7 +182,7 @@ class TestSaveQuanztizedModelFromCheckPoint(unittest.TestCase): imperative_out_scale.save_quantized_model( layer=lenet, - path=save_path, + path=self.save_path, input_spec=[ paddle.static.InputSpec( shape=[None, 1, 28, 28], dtype='float32') @@ -211,10 +193,10 @@ class TestSaveQuanztizedModelFromCheckPoint(unittest.TestCase): loss_list[i] > loss_list[i + 1], msg='Failed to do the imperative qat.') - def test_save_quantized_model(self): + def test_out_scale_acc(self): with _test_eager_guard(): - self.func_save_quantized_model() - self.func_save_quantized_model() + self.func_out_scale_acc() + self.func_out_scale_acc() if __name__ == '__main__': diff --git a/python/paddle/fluid/contrib/slim/tests/test_imperative_ptq.py b/python/paddle/fluid/contrib/slim/tests/test_imperative_ptq.py index fad4c8f9d580b389b861c4c9a992af0c48cd892d..0715fcf2a8bcd33997c3778bd4ee5276a162be68 100644 --- a/python/paddle/fluid/contrib/slim/tests/test_imperative_ptq.py +++ b/python/paddle/fluid/contrib/slim/tests/test_imperative_ptq.py @@ -22,6 +22,7 @@ import time import unittest import copy import logging +import tempfile import paddle.nn as nn import paddle @@ -72,10 +73,6 @@ class TestImperativePTQ(unittest.TestCase): @classmethod def setUpClass(cls): - timestamp = time.strftime('%Y-%m-%d-%H-%M-%S', time.localtime()) - cls.root_path = os.path.join(os.getcwd(), "imperative_ptq_" + timestamp) - cls.save_path = os.path.join(cls.root_path, "model") - cls.download_path = 'dygraph_int8/download' cls.cache_folder = os.path.expanduser('~/.cache/paddle/dataset/' + cls.download_path) @@ -88,14 +85,6 @@ class TestImperativePTQ(unittest.TestCase): paddle.static.default_main_program().random_seed = seed paddle.static.default_startup_program().random_seed = seed - @classmethod - def tearDownClass(cls): - try: - pass - # shutil.rmtree(cls.root_path) - except Exception as e: - print("Failed to delete {} due to {}".format(cls.root_path, str(e))) - def cache_unzipping(self, target_folder, zip_path): if not os.path.exists(target_folder): cmd = 'mkdir {0} && tar xf {1} -C {0}'.format(target_folder, @@ -126,8 +115,8 @@ class TestImperativePTQ(unittest.TestCase): 'batch_norm2d_0': [[0.37673383951187134], [0.44249194860458374]], 're_lu_0': [[0.44249194860458374], [0.25804123282432556]], 'max_pool2d_0': [[0.25804123282432556], [0.25804123282432556]], - 'linear_0': - [[1.7058950662612915], [14.405526161193848], [0.4373355209827423]], + 'linear_0': [[1.7058950662612915], [14.405526161193848], + [0.4373355209827423]], 'add_0': [[1.7058950662612915, 0.0], [1.7058950662612915]], } @@ -141,8 +130,8 @@ class TestImperativePTQ(unittest.TestCase): for batch_id, data in enumerate(test_reader()): x_data = np.array([x[0].reshape(1, 28, 28) for x in data]).astype('float32') - y_data = np.array( - [x[1] for x in data]).astype('int64').reshape(-1, 1) + y_data = np.array([x[1] + for x in data]).astype('int64').reshape(-1, 1) img = paddle.to_tensor(x_data) label = paddle.to_tensor(y_data) @@ -165,8 +154,8 @@ class TestImperativePTQ(unittest.TestCase): def program_test(self, program_path, batch_num=-1, batch_size=8): exe = paddle.static.Executor(paddle.CPUPlace()) - [inference_program, feed_target_names, fetch_targets] = ( - paddle.static.load_inference_model(program_path, exe)) + [inference_program, feed_target_names, fetch_targets + ] = (paddle.static.load_inference_model(program_path, exe)) test_reader = paddle.batch( paddle.dataset.mnist.test(), batch_size=batch_size) @@ -217,33 +206,35 @@ class TestImperativePTQ(unittest.TestCase): paddle.static.InputSpec( shape=[None, 1, 28, 28], dtype='float32') ] - self.ptq.save_quantized_model( - model=quant_model, path=self.save_path, input_spec=input_spec) - print('Quantized model saved in {%s}' % self.save_path) - - after_acc_top1 = self.model_test(quant_model, self.batch_num, - self.batch_size) - - paddle.enable_static() - infer_acc_top1 = self.program_test(self.save_path, self.batch_num, - self.batch_size) - paddle.disable_static() - - # Check - print('Before converted acc_top1: %s' % before_acc_top1) - print('After converted acc_top1: %s' % after_acc_top1) - print('Infer acc_top1: %s' % infer_acc_top1) - - self.assertTrue( - after_acc_top1 >= self.eval_acc_top1, - msg="The test acc {%f} is less than {%f}." % - (after_acc_top1, self.eval_acc_top1)) - self.assertTrue( - infer_acc_top1 >= after_acc_top1, - msg='The acc is lower after converting model.') - - end_time = time.time() - print("total time: %ss \n" % (end_time - start_time)) + with tempfile.TemporaryDirectory(prefix="imperative_ptq_") as tmpdir: + save_path = os.path.join(tmpdir, "model") + self.ptq.save_quantized_model( + model=quant_model, path=save_path, input_spec=input_spec) + print('Quantized model saved in {%s}' % save_path) + + after_acc_top1 = self.model_test(quant_model, self.batch_num, + self.batch_size) + + paddle.enable_static() + infer_acc_top1 = self.program_test(save_path, self.batch_num, + self.batch_size) + paddle.disable_static() + + # Check + print('Before converted acc_top1: %s' % before_acc_top1) + print('After converted acc_top1: %s' % after_acc_top1) + print('Infer acc_top1: %s' % infer_acc_top1) + + self.assertTrue( + after_acc_top1 >= self.eval_acc_top1, + msg="The test acc {%f} is less than {%f}." % + (after_acc_top1, self.eval_acc_top1)) + self.assertTrue( + infer_acc_top1 >= after_acc_top1, + msg='The acc is lower after converting model.') + + end_time = time.time() + print("total time: %ss \n" % (end_time - start_time)) def test_ptq(self): with _test_eager_guard(): @@ -279,37 +270,39 @@ class TestImperativePTQfuse(TestImperativePTQ): paddle.static.InputSpec( shape=[None, 1, 28, 28], dtype='float32') ] - self.ptq.save_quantized_model( - model=quant_model, path=self.save_path, input_spec=input_spec) - print('Quantized model saved in {%s}' % self.save_path) - - after_acc_top1 = self.model_test(quant_model, self.batch_num, - self.batch_size) - - paddle.enable_static() - infer_acc_top1 = self.program_test(self.save_path, self.batch_num, - self.batch_size) - paddle.disable_static() - - # Check - print('Before converted acc_top1: %s' % before_acc_top1) - print('After converted acc_top1: %s' % after_acc_top1) - print('Infer acc_top1: %s' % infer_acc_top1) - - #Check whether the quant_model is correct after converting. - #The acc of quantized model should be higher than 0.95. - self.assertTrue( - after_acc_top1 >= self.eval_acc_top1, - msg="The test acc {%f} is less than {%f}." % - (after_acc_top1, self.eval_acc_top1)) - #Check the saved infer_model.The acc of infer model - #should not be lower than the one of dygraph model. - self.assertTrue( - infer_acc_top1 >= after_acc_top1, - msg='The acc is lower after converting model.') - - end_time = time.time() - print("total time: %ss \n" % (end_time - start_time)) + with tempfile.TemporaryDirectory(prefix="imperative_ptq_") as tmpdir: + save_path = os.path.join(tmpdir, "model") + self.ptq.save_quantized_model( + model=quant_model, path=save_path, input_spec=input_spec) + print('Quantized model saved in {%s}' % save_path) + + after_acc_top1 = self.model_test(quant_model, self.batch_num, + self.batch_size) + + paddle.enable_static() + infer_acc_top1 = self.program_test(save_path, self.batch_num, + self.batch_size) + paddle.disable_static() + + # Check + print('Before converted acc_top1: %s' % before_acc_top1) + print('After converted acc_top1: %s' % after_acc_top1) + print('Infer acc_top1: %s' % infer_acc_top1) + + #Check whether the quant_model is correct after converting. + #The acc of quantized model should be higher than 0.95. + self.assertTrue( + after_acc_top1 >= self.eval_acc_top1, + msg="The test acc {%f} is less than {%f}." % + (after_acc_top1, self.eval_acc_top1)) + #Check the saved infer_model.The acc of infer model + #should not be lower than the one of dygraph model. + self.assertTrue( + infer_acc_top1 >= after_acc_top1, + msg='The acc is lower after converting model.') + + end_time = time.time() + print("total time: %ss \n" % (end_time - start_time)) def test_ptq(self): with _test_eager_guard(): @@ -327,13 +320,13 @@ class TestImperativePTQHist(TestImperativePTQ): self.eval_acc_top1 = 0.98 self.gt_thresholds = { - 'conv2d_0': - [[0.99853515625], [0.35732391771364225], [0.10933732241392136]], + 'conv2d_0': [[0.99853515625], [0.35732391771364225], + [0.10933732241392136]], 'batch_norm2d_0': [[0.35732391771364225], [0.4291427868761275]], 're_lu_0': [[0.4291427868761275], [0.2359918110742001]], 'max_pool2d_0': [[0.2359918110742001], [0.25665526917146053]], - 'linear_0': - [[1.7037603475152991], [14.395224522473026], [0.4373355209827423]], + 'linear_0': [[1.7037603475152991], [14.395224522473026], + [0.4373355209827423]], 'add_0': [[1.7037603475152991, 0.0], [1.7037603475152991]], } diff --git a/python/paddle/fluid/contrib/slim/tests/test_imperative_qat_amp.py b/python/paddle/fluid/contrib/slim/tests/test_imperative_qat_amp.py index 76a6e11d98dffe821501b02683b7565808f9c7c1..e163c745d6d8e9bba5b7a7c3675bee500bce1dca 100644 --- a/python/paddle/fluid/contrib/slim/tests/test_imperative_qat_amp.py +++ b/python/paddle/fluid/contrib/slim/tests/test_imperative_qat_amp.py @@ -21,6 +21,7 @@ import shutil import time import unittest import logging +import tempfile import paddle import paddle.fluid as fluid @@ -45,10 +46,9 @@ class TestImperativeQatAmp(unittest.TestCase): @classmethod def setUpClass(cls): - timestamp = time.strftime('%Y-%m-%d-%H-%M-%S', time.localtime()) - cls.root_path = os.path.join(os.getcwd(), - "imperative_qat_amp_" + timestamp) - cls.save_path = os.path.join(cls.root_path, "model") + cls.root_path = tempfile.TemporaryDirectory( + prefix="imperative_qat_amp_") + cls.save_path = os.path.join(cls.root_path.name, "model") cls.download_path = 'dygraph_int8/download' cls.cache_folder = os.path.expanduser('~/.cache/paddle/dataset/' + @@ -64,10 +64,7 @@ class TestImperativeQatAmp(unittest.TestCase): @classmethod def tearDownClass(cls): - try: - shutil.rmtree(cls.root_path) - except Exception as e: - print("Failed to delete {} due to {}".format(cls.root_path, str(e))) + cls.root_path.cleanup() def cache_unzipping(self, target_folder, zip_path): if not os.path.exists(target_folder): @@ -106,8 +103,8 @@ class TestImperativeQatAmp(unittest.TestCase): for batch_id, data in enumerate(train_reader()): x_data = np.array([x[0].reshape(1, 28, 28) for x in data]).astype('float32') - y_data = np.array( - [x[1] for x in data]).astype('int64').reshape(-1, 1) + y_data = np.array([x[1] + for x in data]).astype('int64').reshape(-1, 1) img = paddle.to_tensor(x_data) label = paddle.to_tensor(y_data) @@ -150,8 +147,8 @@ class TestImperativeQatAmp(unittest.TestCase): for batch_id, data in enumerate(test_reader()): x_data = np.array([x[0].reshape(1, 28, 28) for x in data]).astype('float32') - y_data = np.array( - [x[1] for x in data]).astype('int64').reshape(-1, 1) + y_data = np.array([x[1] + for x in data]).astype('int64').reshape(-1, 1) img = paddle.to_tensor(x_data) label = paddle.to_tensor(y_data) diff --git a/python/paddle/fluid/contrib/slim/tests/test_post_training_quantization_lstm_model.py b/python/paddle/fluid/contrib/slim/tests/test_post_training_quantization_lstm_model.py index 89e0e099f44c2c74ae77df4004389afc1dc9a43f..4ea51233e403a015690f7bad4c2e42294a042746 100644 --- a/python/paddle/fluid/contrib/slim/tests/test_post_training_quantization_lstm_model.py +++ b/python/paddle/fluid/contrib/slim/tests/test_post_training_quantization_lstm_model.py @@ -20,6 +20,7 @@ import math import functools import contextlib import struct +import tempfile import numpy as np import paddle import paddle.fluid as fluid @@ -37,9 +38,9 @@ class TestPostTrainingQuantization(unittest.TestCase): self.download_path = 'int8/download' self.cache_folder = os.path.expanduser('~/.cache/paddle/dataset/' + self.download_path) - self.timestamp = time.strftime('%Y-%m-%d-%H-%M-%S', time.localtime()) - self.int8_model_path = os.path.join(os.getcwd(), - "post_training_" + self.timestamp) + self.root_path = tempfile.TemporaryDirectory() + self.int8_model_path = os.path.join(self.root_path.name, + "post_training_quantization") try: os.system("mkdir -p " + self.int8_model_path) except Exception as e: @@ -48,11 +49,7 @@ class TestPostTrainingQuantization(unittest.TestCase): sys.exit(-1) def tearDown(self): - try: - os.system("rm -rf {}".format(self.int8_model_path)) - except Exception as e: - print("Failed to delete {} due to {}".format(self.int8_model_path, - str(e))) + self.root_path.cleanup() def cache_unzipping(self, target_folder, zip_path): if not os.path.exists(target_folder): diff --git a/python/paddle/fluid/contrib/slim/tests/test_post_training_quantization_mnist.py b/python/paddle/fluid/contrib/slim/tests/test_post_training_quantization_mnist.py index d231aa2a1242c68cf613641cff0ffdeb9487dbcc..cd71ccb95a8e490395335fe2cb4831ce229e34a4 100644 --- a/python/paddle/fluid/contrib/slim/tests/test_post_training_quantization_mnist.py +++ b/python/paddle/fluid/contrib/slim/tests/test_post_training_quantization_mnist.py @@ -18,6 +18,7 @@ import sys import random import math import functools +import tempfile import contextlib import numpy as np import paddle @@ -33,12 +34,12 @@ np.random.seed(0) class TestPostTrainingQuantization(unittest.TestCase): def setUp(self): + self.root_path = tempfile.TemporaryDirectory() + self.int8_model_path = os.path.join(self.root_path.name, + "post_training_quantization") self.download_path = 'int8/download' self.cache_folder = os.path.expanduser('~/.cache/paddle/dataset/' + self.download_path) - self.timestamp = time.strftime('%Y-%m-%d-%H-%M-%S', time.localtime()) - self.int8_model_path = os.path.join(os.getcwd(), - "post_training_" + self.timestamp) try: os.system("mkdir -p " + self.int8_model_path) except Exception as e: @@ -47,11 +48,7 @@ class TestPostTrainingQuantization(unittest.TestCase): sys.exit(-1) def tearDown(self): - try: - os.system("rm -rf {}".format(self.int8_model_path)) - except Exception as e: - print("Failed to delete {} due to {}".format(self.int8_model_path, - str(e))) + self.root_path.cleanup() def cache_unzipping(self, target_folder, zip_path): if not os.path.exists(target_folder): @@ -82,8 +79,8 @@ class TestPostTrainingQuantization(unittest.TestCase): cnt = 0 periods = [] for batch_id, data in enumerate(val_reader()): - image = np.array( - [x[0].reshape(img_shape) for x in data]).astype("float32") + image = np.array([x[0].reshape(img_shape) + for x in data]).astype("float32") input_label = np.array([x[1] for x in data]).astype("int64") t1 = time.time() @@ -121,7 +118,6 @@ class TestPostTrainingQuantization(unittest.TestCase): place = fluid.CPUPlace() exe = fluid.Executor(place) - scope = fluid.global_scope() val_reader = paddle.dataset.mnist.train() ptq = PostTrainingQuantization( @@ -178,13 +174,13 @@ class TestPostTrainingQuantization(unittest.TestCase): print("---Post training quantization of {} method---".format(algo)) print( - "FP32 {0}: batch_size {1}, throughput {2} img/s, latency {3} s, acc1 {4}.". - format(model_name, batch_size, fp32_throughput, fp32_latency, - fp32_acc1)) + "FP32 {0}: batch_size {1}, throughput {2} img/s, latency {3} s, acc1 {4}." + .format(model_name, batch_size, fp32_throughput, fp32_latency, + fp32_acc1)) print( - "INT8 {0}: batch_size {1}, throughput {2} img/s, latency {3} s, acc1 {4}.\n". - format(model_name, batch_size, int8_throughput, int8_latency, - int8_acc1)) + "INT8 {0}: batch_size {1}, throughput {2} img/s, latency {3} s, acc1 {4}.\n" + .format(model_name, batch_size, int8_throughput, int8_latency, + int8_acc1)) sys.stdout.flush() delta_value = fp32_acc1 - int8_acc1 diff --git a/python/paddle/fluid/contrib/slim/tests/test_post_training_quantization_mobilenetv1.py b/python/paddle/fluid/contrib/slim/tests/test_post_training_quantization_mobilenetv1.py index 629529ff1b96555cc8ac9f54badf99d92974a27e..7e2a4b9c5cfa3baa8edc89300a0aaa290d01e13d 100644 --- a/python/paddle/fluid/contrib/slim/tests/test_post_training_quantization_mobilenetv1.py +++ b/python/paddle/fluid/contrib/slim/tests/test_post_training_quantization_mobilenetv1.py @@ -19,6 +19,7 @@ import random import math import functools import contextlib +import tempfile import numpy as np from PIL import Image, ImageEnhance import paddle @@ -146,16 +147,12 @@ class TestPostTrainingQuantization(unittest.TestCase): self.infer_iterations = 50000 if os.environ.get( 'DATASET') == 'full' else 2 - self.timestamp = time.strftime('%Y-%m-%d-%H-%M-%S', time.localtime()) - self.int8_model = os.path.join(os.getcwd(), - "post_training_" + self.timestamp) + self.root_path = tempfile.TemporaryDirectory() + self.int8_model = os.path.join(self.root_path.name, + "post_training_quantization") def tearDown(self): - try: - os.system("rm -rf {}".format(self.int8_model)) - except Exception as e: - print("Failed to delete {} due to {}".format(self.int8_model, - str(e))) + self.root_path.cleanup() def cache_unzipping(self, target_folder, zip_path): if not os.path.exists(target_folder): @@ -207,8 +204,8 @@ class TestPostTrainingQuantization(unittest.TestCase): cnt = 0 periods = [] for batch_id, data in enumerate(val_reader()): - image = np.array( - [x[0].reshape(image_shape) for x in data]).astype("float32") + image = np.array([x[0].reshape(image_shape) + for x in data]).astype("float32") label = np.array([x[1] for x in data]).astype("int64") label = label.reshape([-1, 1]) @@ -308,11 +305,13 @@ class TestPostTrainingQuantization(unittest.TestCase): print("---Post training quantization of {} method---".format(algo)) print( - "FP32 {0}: batch_size {1}, throughput {2} images/second, latency {3} second, accuracy {4}.". - format(model, batch_size, fp32_throughput, fp32_latency, fp32_acc1)) + "FP32 {0}: batch_size {1}, throughput {2} images/second, latency {3} second, accuracy {4}." + .format(model, batch_size, fp32_throughput, fp32_latency, + fp32_acc1)) print( - "INT8 {0}: batch_size {1}, throughput {2} images/second, latency {3} second, accuracy {4}.\n". - format(model, batch_size, int8_throughput, int8_latency, int8_acc1)) + "INT8 {0}: batch_size {1}, throughput {2} images/second, latency {3} second, accuracy {4}.\n" + .format(model, batch_size, int8_throughput, int8_latency, + int8_acc1)) sys.stdout.flush() delta_value = fp32_acc1 - int8_acc1 @@ -405,7 +404,7 @@ class TestPostTrainingAbsMaxForMobilenetv1(TestPostTrainingQuantization): is_full_quantize = False is_use_cache_file = False is_optimize_model = False - # The accuracy diff of post-traing quantization (abs_max) maybe bigger + # The accuracy diff of post-training quantization (abs_max) maybe bigger diff_threshold = 0.05 self.run_test(model, algo, round_type, data_urls, data_md5s, quantizable_op_type, is_full_quantize, is_use_cache_file, diff --git a/python/paddle/fluid/contrib/slim/tests/test_quantization_scale_pass.py b/python/paddle/fluid/contrib/slim/tests/test_quantization_scale_pass.py index ec2c7a91f96ab7b22945732096a89ed38db46682..3d3892ba803d55d9784f68515abade3c31ee6de9 100644 --- a/python/paddle/fluid/contrib/slim/tests/test_quantization_scale_pass.py +++ b/python/paddle/fluid/contrib/slim/tests/test_quantization_scale_pass.py @@ -17,6 +17,7 @@ import unittest import random import numpy as np import six +import tempfile import paddle.fluid as fluid import paddle from paddle.fluid.framework import IrGraph @@ -165,15 +166,20 @@ class TestQuantizationScalePass(unittest.TestCase): marked_nodes.add(op) test_graph.draw('.', 'quant_scale' + dev_name, marked_nodes) - with open('quant_scale_model' + dev_name + '.txt', 'w') as f: + tempdir = tempfile.TemporaryDirectory() + mapping_table_path = os.path.join( + tempdir.name, 'quant_scale_model' + dev_name + '.txt') + save_path = os.path.join(tempdir.name, 'quant_scale_model' + dev_name) + with open(mapping_table_path, 'w') as f: f.write(str(server_program)) with fluid.scope_guard(scope): fluid.io.save_inference_model( - 'quant_scale_model' + dev_name, ['image', 'label'], [loss], + save_path, ['image', 'label'], [loss], exe, server_program, clip_extra=True) + tempdir.cleanup() def test_quant_scale_cuda(self): if fluid.core.is_compiled_with_cuda(): diff --git a/python/paddle/fluid/contrib/slim/tests/test_user_defined_quantization.py b/python/paddle/fluid/contrib/slim/tests/test_user_defined_quantization.py index f03d0faa3981b5767eef1c5fde0f583f08686c13..20691c69b05cb5dcd2a06f24169f2ab57bbe094b 100644 --- a/python/paddle/fluid/contrib/slim/tests/test_user_defined_quantization.py +++ b/python/paddle/fluid/contrib/slim/tests/test_user_defined_quantization.py @@ -18,6 +18,7 @@ import json import random import numpy as np import six +import tempfile import paddle.fluid as fluid import paddle from paddle.fluid.framework import IrGraph @@ -108,18 +109,20 @@ class TestUserDefinedQuantization(unittest.TestCase): def get_optimizer(): return fluid.optimizer.MomentumOptimizer(0.0001, 0.9) - def load_dict(): - with open('mapping_table_for_saving_inference_model', 'r') as file: + def load_dict(mapping_table_path): + with open(mapping_table_path, 'r') as file: data = file.read() data = json.loads(data) return data - def save_dict(Dict): - with open('mapping_table_for_saving_inference_model', 'w') as file: + def save_dict(Dict, mapping_table_path): + with open(mapping_table_path, 'w') as file: file.write(json.dumps(Dict)) random.seed(0) np.random.seed(0) + tempdir = tempfile.TemporaryDirectory() + mapping_table_path = os.path.join(tempdir.name, 'inference') main = fluid.Program() startup = fluid.Program() @@ -160,7 +163,7 @@ class TestUserDefinedQuantization(unittest.TestCase): executor=exe) test_transform_pass.apply(test_graph) - save_dict(test_graph.out_node_mapping_table) + save_dict(test_graph.out_node_mapping_table, mapping_table_path) add_quant_dequant_pass = AddQuantDequantPass(scope=scope, place=place) add_quant_dequant_pass.apply(main_graph) @@ -202,10 +205,11 @@ class TestUserDefinedQuantization(unittest.TestCase): activation_bits=8, weight_quantize_type=weight_quant_type) - mapping_table = load_dict() + mapping_table = load_dict(mapping_table_path) test_graph.out_node_mapping_table = mapping_table if act_quantize_func == None and weight_quantize_func == None: freeze_pass.apply(test_graph) + tempdir.cleanup() def test_act_preprocess_cuda(self): if fluid.core.is_compiled_with_cuda(): diff --git a/python/paddle/fluid/tests/unittests/test_inference_model_io.py b/python/paddle/fluid/tests/unittests/test_inference_model_io.py index 9abcf2a767662ca11a412b8135fce0f562a9979b..61d086655430e734c59d8fa5d7d0e688cdc3a6a2 100644 --- a/python/paddle/fluid/tests/unittests/test_inference_model_io.py +++ b/python/paddle/fluid/tests/unittests/test_inference_model_io.py @@ -18,6 +18,7 @@ import unittest import os import six +import tempfile import numpy as np import paddle.fluid.core as core import paddle.fluid as fluid @@ -31,6 +32,7 @@ from paddle.fluid.compiler import CompiledProgram from paddle.fluid.framework import Program, program_guard from paddle.fluid.io import save_inference_model, load_inference_model, save_persistables from paddle.fluid.transpiler import memory_optimize + paddle.enable_static() @@ -43,8 +45,9 @@ class InferModel(object): class TestBook(unittest.TestCase): def test_fit_line_inference_model(self): - MODEL_DIR = "./tmp/inference_model" - UNI_MODEL_DIR = "./tmp/inference_model1" + root_path = tempfile.TemporaryDirectory() + MODEL_DIR = os.path.join(root_path.name, "inference_model") + UNI_MODEL_DIR = os.path.join(root_path.name, "inference_model1") init_program = Program() program = Program() @@ -67,8 +70,8 @@ class TestBook(unittest.TestCase): exe.run(init_program, feed={}, fetch_list=[]) for i in six.moves.xrange(100): - tensor_x = np.array( - [[1, 1], [1, 2], [3, 4], [5, 2]]).astype("float32") + tensor_x = np.array([[1, 1], [1, 2], [3, 4], [5, + 2]]).astype("float32") tensor_y = np.array([[-2], [-3], [-7], [-7]]).astype("float32") exe.run(program, @@ -111,13 +114,16 @@ class TestBook(unittest.TestCase): print("fetch %s" % str(model.fetch_vars[0])) self.assertEqual(expected, actual) + root_path.cleanup() + self.assertRaises(ValueError, fluid.io.load_inference_model, None, exe, model_str, None) class TestSaveInferenceModel(unittest.TestCase): def test_save_inference_model(self): - MODEL_DIR = "./tmp/inference_model2" + root_path = tempfile.TemporaryDirectory() + MODEL_DIR = os.path.join(root_path.name, "inference_model2") init_program = Program() program = Program() @@ -136,9 +142,11 @@ class TestSaveInferenceModel(unittest.TestCase): exe.run(init_program, feed={}, fetch_list=[]) save_inference_model(MODEL_DIR, ["x", "y"], [avg_cost], exe, program) + root_path.cleanup() def test_save_inference_model_with_auc(self): - MODEL_DIR = "./tmp/inference_model4" + root_path = tempfile.TemporaryDirectory() + MODEL_DIR = os.path.join(root_path.name, "inference_model4") init_program = Program() program = Program() @@ -160,6 +168,7 @@ class TestSaveInferenceModel(unittest.TestCase): warnings.simplefilter("always") save_inference_model(MODEL_DIR, ["x", "y"], [avg_cost], exe, program) + root_path.cleanup() expected_warn = "please ensure that you have set the auc states to zeros before saving inference model" self.assertTrue(len(w) > 0) self.assertTrue(expected_warn == str(w[0].message)) @@ -167,7 +176,8 @@ class TestSaveInferenceModel(unittest.TestCase): class TestInstance(unittest.TestCase): def test_save_inference_model(self): - MODEL_DIR = "./tmp/inference_model3" + root_path = tempfile.TemporaryDirectory() + MODEL_DIR = os.path.join(root_path.name, "inference_model3") init_program = Program() program = Program() @@ -193,11 +203,13 @@ class TestInstance(unittest.TestCase): save_inference_model(MODEL_DIR, ["x", "y"], [avg_cost], exe, cp_prog) self.assertRaises(TypeError, save_inference_model, [MODEL_DIR, ["x", "y"], [avg_cost], [], cp_prog]) + root_path.cleanup() class TestSaveInferenceModelNew(unittest.TestCase): def test_save_and_load_inference_model(self): - MODEL_DIR = "./tmp/inference_model5" + root_path = tempfile.TemporaryDirectory() + MODEL_DIR = os.path.join(root_path.name, "inference_model5") init_program = fluid.default_startup_program() program = fluid.default_main_program() @@ -292,6 +304,7 @@ class TestSaveInferenceModelNew(unittest.TestCase): model = InferModel( paddle.static.io.load_inference_model(MODEL_DIR, exe)) + root_path.cleanup() outs = exe.run(model.program, feed={ diff --git a/python/paddle/fluid/tests/unittests/test_save_inference_model_conditional_op.py b/python/paddle/fluid/tests/unittests/test_save_inference_model_conditional_op.py index 86431086ac5f963493a9a81522287a53f7cacb42..7c123e3d6e6e1eaee8ca6ae0a4eeab16f73b26cc 100644 --- a/python/paddle/fluid/tests/unittests/test_save_inference_model_conditional_op.py +++ b/python/paddle/fluid/tests/unittests/test_save_inference_model_conditional_op.py @@ -17,6 +17,7 @@ from __future__ import print_function import os import unittest import numpy as np +import tempfile import paddle import paddle.fluid as fluid @@ -31,7 +32,6 @@ def getModelOp(model_path): result = set() for i in range(0, size): - #print(main_block.op(i).type()) result.add(main_block.op(i).type()) return result @@ -90,18 +90,20 @@ class TestConditionalOp(unittest.TestCase): paddle.static.InputSpec( shape=[1, 3, 8, 8], dtype='float32') ]) - paddle.jit.save(net, './while_net') + root_path = tempfile.TemporaryDirectory() + model_file = os.path.join(root_path.name, "while_net") + paddle.jit.save(net, model_file) right_pdmodel = set([ "uniform_random", "shape", "slice", "not_equal", "while", "elementwise_add" ]) paddle.enable_static() - pdmodel = getModelOp("while_net.pdmodel") - #print(len(right_pdmodel.difference(pdmodel))) + pdmodel = getModelOp(model_file + ".pdmodel") self.assertTrue( len(right_pdmodel.difference(pdmodel)) == 0, "The while op is pruned by mistake.") + root_path.cleanup() def test_for_op(self): paddle.disable_static() @@ -110,18 +112,20 @@ class TestConditionalOp(unittest.TestCase): net, input_spec=[paddle.static.InputSpec( shape=[1], dtype='int32')]) - paddle.jit.save(net, './for_net') + root_path = tempfile.TemporaryDirectory() + model_file = os.path.join(root_path.name, "for_net") + paddle.jit.save(net, model_file) right_pdmodel = set([ "randint", "fill_constant", "cast", "less_than", "while", "elementwise_add" ]) paddle.enable_static() - pdmodel = getModelOp("for_net.pdmodel") - #print(len(right_pdmodel.difference(pdmodel))) + pdmodel = getModelOp(model_file + ".pdmodel") self.assertTrue( len(right_pdmodel.difference(pdmodel)) == 0, "The for op is pruned by mistake.") + root_path.cleanup() def test_if_op(self): paddle.disable_static() @@ -130,18 +134,20 @@ class TestConditionalOp(unittest.TestCase): net, input_spec=[paddle.static.InputSpec( shape=[1], dtype='int32')]) - paddle.jit.save(net, './if_net') + root_path = tempfile.TemporaryDirectory() + model_file = os.path.join(root_path.name, "if_net") + paddle.jit.save(net, model_file) right_pdmodel = set([ "assign_value", "greater_than", "cast", "conditional_block", "logical_not", "select_input" ]) paddle.enable_static() - pdmodel = getModelOp("if_net.pdmodel") - #print(len(right_pdmodel.difference(pdmodel))) + pdmodel = getModelOp(model_file + ".pdmodel") self.assertTrue( len(right_pdmodel.difference(pdmodel)) == 0, "The if op is pruned by mistake.") + root_path.cleanup() if __name__ == '__main__':