[Dy2Stat]Modify all jit.save path into tempfile under dygraph_to_static directory (#42842) (#42860)

* [Dy2Stat]Modify all jit.save path into tempfile

* [Dy2Stat]Modify all jit.save path into tempfile

python/paddle/fluid/tests/unittests/dygraph_to_static/test_bmn.py

@@ -12,10 +12,12 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import os
 import math
 import numpy as np
 import unittest
 import paddle
+import tempfile
 from paddle.jit import to_static
 import paddle.fluid as fluid
 from paddle.fluid import ParamAttr
@@ -422,11 +424,6 @@ class Args(object):
     prop_boundary_ratio = 0.5
     num_sample = 2
     num_sample_perbin = 2
-    model_save_dir = "./inference"
-    model_save_prefix = "./inference/bmn"
-    model_filename = "bmn" + INFER_MODEL_SUFFIX
-    params_filename = "bmn" + INFER_PARAMS_SUFFIX
-    dy_param_path = './bmn_dy_param'
 
 
 def optimizer(cfg, parameter_list):
@@ -559,78 +556,6 @@ def fake_data_reader(args, mode='train'):
     return reader
 
 
-def train_bmn(args, place, to_static):
-    program_translator.enable(to_static)
-    loss_data = []
-
-    with fluid.dygraph.guard(place):
-        paddle.seed(SEED)
-        paddle.framework.random._manual_program_seed(SEED)
-        global local_random
-        local_random = np.random.RandomState(SEED)
-
-        bmn = BMN(args)
-        adam = optimizer(args, parameter_list=bmn.parameters())
-
-        train_reader = fake_data_reader(args, 'train')
-
-        for epoch in range(args.epoch):
-            for batch_id, data in enumerate(train_reader()):
-                video_feat = np.array(
-                    [item[0] for item in data]).astype(DATATYPE)
-                gt_iou_map = np.array(
-                    [item[1] for item in data]).astype(DATATYPE)
-                gt_start = np.array([item[2] for item in data]).astype(DATATYPE)
-                gt_end = np.array([item[3] for item in data]).astype(DATATYPE)
-
-                x_data = to_variable(video_feat)
-                gt_iou_map = to_variable(gt_iou_map)
-                gt_start = to_variable(gt_start)
-                gt_end = to_variable(gt_end)
-                gt_iou_map.stop_gradient = True
-                gt_start.stop_gradient = True
-                gt_end.stop_gradient = True
-
-                pred_bm, pred_start, pred_end = bmn(x_data)
-
-                loss, tem_loss, pem_reg_loss, pem_cls_loss = bmn_loss_func(
-                    pred_bm, pred_start, pred_end, gt_iou_map, gt_start, gt_end,
-                    args)
-                avg_loss = fluid.layers.mean(loss)
-
-                avg_loss.backward()
-                adam.minimize(avg_loss)
-                bmn.clear_gradients()
-                # log loss data to verify correctness
-                loss_data += [
-                    avg_loss.numpy()[0], tem_loss.numpy()[0],
-                    pem_reg_loss.numpy()[0], pem_cls_loss.numpy()[0]
-                ]
-
-                if args.log_interval > 0 and (
-                        batch_id % args.log_interval == 0):
-                    print('[TRAIN] Epoch {}, iter {} '.format(epoch, batch_id)
-                                + '\tLoss = {}, \ttem_loss = {}, \tpem_reg_loss = {}, \tpem_cls_loss = {}'.format(
-                        '%f' % avg_loss.numpy()[0], '%f' % tem_loss.numpy()[0], \
-                        '%f' % pem_reg_loss.numpy()[0], '%f' % pem_cls_loss.numpy()[0]))
-
-                # validation
-                if batch_id % args.valid_interval == 0 and batch_id > 0:
-                    bmn.eval()
-                    val_loss_data = val_bmn(bmn, args)
-                    bmn.train()
-                    loss_data += val_loss_data
-
-                if batch_id == args.train_batch_num:
-                    if to_static:
-                        fluid.dygraph.jit.save(bmn, args.model_save_prefix)
-                    else:
-                        fluid.dygraph.save_dygraph(bmn.state_dict(),
-                                                   args.dy_param_path)
-                    break
-        return np.array(loss_data)
-
-
 # Validation
 def val_bmn(model, args):
     val_reader = fake_data_reader(args, 'valid')
@@ -677,10 +602,93 @@ class TestTrain(unittest.TestCase):
         self.place = fluid.CPUPlace() if not fluid.is_compiled_with_cuda() \
             else fluid.CUDAPlace(0)
 
+        self.temp_dir = tempfile.TemporaryDirectory()
+        self.model_save_dir = os.path.join(self.temp_dir.name, 'inference')
+        self.model_save_prefix = os.path.join(self.model_save_dir, 'bmn')
+        self.model_filename = "bmn" + INFER_MODEL_SUFFIX
+        self.params_filename = "bmn" + INFER_PARAMS_SUFFIX
+        self.dy_param_path = os.path.join(self.temp_dir.name, 'bmn_dy_param')
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
+
+    def train_bmn(self, args, place, to_static):
+        program_translator.enable(to_static)
+        loss_data = []
+
+        with fluid.dygraph.guard(place):
+            paddle.seed(SEED)
+            paddle.framework.random._manual_program_seed(SEED)
+            global local_random
+            local_random = np.random.RandomState(SEED)
+
+            bmn = BMN(args)
+            adam = optimizer(args, parameter_list=bmn.parameters())
+
+            train_reader = fake_data_reader(args, 'train')
+
+            for epoch in range(args.epoch):
+                for batch_id, data in enumerate(train_reader()):
+                    video_feat = np.array(
+                        [item[0] for item in data]).astype(DATATYPE)
+                    gt_iou_map = np.array(
+                        [item[1] for item in data]).astype(DATATYPE)
+                    gt_start = np.array(
+                        [item[2] for item in data]).astype(DATATYPE)
+                    gt_end = np.array(
+                        [item[3] for item in data]).astype(DATATYPE)
+
+                    x_data = to_variable(video_feat)
+                    gt_iou_map = to_variable(gt_iou_map)
+                    gt_start = to_variable(gt_start)
+                    gt_end = to_variable(gt_end)
+                    gt_iou_map.stop_gradient = True
+                    gt_start.stop_gradient = True
+                    gt_end.stop_gradient = True
+
+                    pred_bm, pred_start, pred_end = bmn(x_data)
+
+                    loss, tem_loss, pem_reg_loss, pem_cls_loss = bmn_loss_func(
+                        pred_bm, pred_start, pred_end, gt_iou_map, gt_start,
+                        gt_end, args)
+                    avg_loss = fluid.layers.mean(loss)
+
+                    avg_loss.backward()
+                    adam.minimize(avg_loss)
+                    bmn.clear_gradients()
+                    # log loss data to verify correctness
+                    loss_data += [
+                        avg_loss.numpy()[0], tem_loss.numpy()[0],
+                        pem_reg_loss.numpy()[0], pem_cls_loss.numpy()[0]
+                    ]
+
+                    if args.log_interval > 0 and (
+                            batch_id % args.log_interval == 0):
+                        print('[TRAIN] Epoch {}, iter {} '.format(epoch, batch_id)
+                                    + '\tLoss = {}, \ttem_loss = {}, \tpem_reg_loss = {}, \tpem_cls_loss = {}'.format(
+                            '%f' % avg_loss.numpy()[0], '%f' % tem_loss.numpy()[0], \
+                            '%f' % pem_reg_loss.numpy()[0], '%f' % pem_cls_loss.numpy()[0]))
+
+                    # validation
+                    if batch_id % args.valid_interval == 0 and batch_id > 0:
+                        bmn.eval()
+                        val_loss_data = val_bmn(bmn, args)
+                        bmn.train()
+                        loss_data += val_loss_data
+
+                    if batch_id == args.train_batch_num:
+                        if to_static:
+                            fluid.dygraph.jit.save(bmn, self.model_save_prefix)
+                        else:
+                            fluid.dygraph.save_dygraph(bmn.state_dict(),
+                                                       self.dy_param_path)
+                        break
+            return np.array(loss_data)
+
     def test_train(self):
 
-        static_res = train_bmn(self.args, self.place, to_static=True)
-        dygraph_res = train_bmn(self.args, self.place, to_static=False)
+        static_res = self.train_bmn(self.args, self.place, to_static=True)
+        dygraph_res = self.train_bmn(self.args, self.place, to_static=False)
         self.assertTrue(
             np.allclose(dygraph_res, static_res),
             "dygraph_res: {},\n static_res: {}".format(
@@ -726,8 +734,7 @@ class TestTrain(unittest.TestCase):
         with fluid.dygraph.guard(self.place):
             bmn = BMN(self.args)
             # load dygraph trained parameters
-            model_dict, _ = fluid.load_dygraph(self.args.dy_param_path +
-                                               ".pdparams")
+            model_dict, _ = fluid.load_dygraph(self.dy_param_path + ".pdparams")
             bmn.set_dict(model_dict)
             bmn.eval()
 
@@ -743,10 +750,10 @@ class TestTrain(unittest.TestCase):
         # load inference model
         [inference_program, feed_target_names,
          fetch_targets] = fluid.io.load_inference_model(
-             self.args.model_save_dir,
+             self.model_save_dir,
              executor=exe,
-             model_filename=self.args.model_filename,
-             params_filename=self.args.params_filename)
+             model_filename=self.model_filename,
+             params_filename=self.params_filename)
         pred_res = exe.run(inference_program,
                            feed={feed_target_names[0]: data},
                            fetch_list=fetch_targets)
@@ -755,7 +762,7 @@ class TestTrain(unittest.TestCase):
 
     def predict_dygraph_jit(self, data):
         with fluid.dygraph.guard(self.place):
-            bmn = fluid.dygraph.jit.load(self.args.model_save_prefix)
+            bmn = fluid.dygraph.jit.load(self.model_save_prefix)
             bmn.eval()
 
             x = to_variable(data)
@@ -765,9 +772,8 @@ class TestTrain(unittest.TestCase):
             return pred_res
 
     def predict_analysis_inference(self, data):
-        output = PredictorTools(self.args.model_save_dir,
-                                self.args.model_filename,
-                                self.args.params_filename, [data])
+        output = PredictorTools(self.model_save_dir, self.model_filename,
+                                self.params_filename, [data])
         out = output()
         return out
 

python/paddle/fluid/tests/unittests/dygraph_to_static/test_container.py

@@ -12,9 +12,11 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import os
 import paddle
 import unittest
 import numpy as np
+import tempfile
 
 
 class BufferLayers(paddle.nn.Layer):
@@ -66,11 +68,15 @@ class TestSequential(unittest.TestCase):
     def setUp(self):
         paddle.set_device('cpu')
         self.seed = 2021
+        self.temp_dir = tempfile.TemporaryDirectory()
         self._init_config()
 
     def _init_config(self):
         self.net = SequentialNet(BufferLayers, 10, 3)
-        self.model_path = './sequential_net'
+        self.model_path = os.path.join(self.temp_dir.name, 'sequential_net')
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
 
     def _init_seed(self):
         paddle.seed(self.seed)
@@ -108,7 +114,8 @@ class TestSequential(unittest.TestCase):
 class TestNestSequential(TestSequential):
     def _init_config(self):
         self.net = NestSequentialNet()
-        self.model_path = './nested_sequential_net'
+        self.model_path = os.path.join(self.temp_dir.name,
+                                       'nested_sequential_net')
 
 
 if __name__ == '__main__':

python/paddle/fluid/tests/unittests/dygraph_to_static/test_declarative.py

@@ -14,7 +14,8 @@
 
 import numpy as np
 import unittest
-
+import os
+import tempfile
 import paddle
 import paddle.fluid as fluid
 from paddle.static import InputSpec
@@ -100,7 +101,11 @@ class TestStaticFunctionInstance(unittest.TestCase):
 
 class TestInputSpec(unittest.TestCase):
     def setUp(self):
-        pass
+        self.temp_dir = tempfile.TemporaryDirectory()
+        self.model_path = os.path.join(self.temp_dir.name, 'simple_net')
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
 
     def test_with_input_spec(self):
         with fluid.dygraph.guard(fluid.CPUPlace()):
@@ -116,8 +121,8 @@ class TestInputSpec(unittest.TestCase):
 
             # 2. test save load
             net.inner_function(x)
-            jit.save(net, './simple_net')
-            infer_net = fluid.dygraph.jit.load('./simple_net')
+            jit.save(net, self.model_path)
+            infer_net = fluid.dygraph.jit.load(self.model_path)
             pred = infer_net(x)
             self.assertTrue(np.allclose(out.numpy(), pred.numpy()))
 
@@ -438,12 +443,19 @@ class SetBuffersNet2(paddle.nn.Layer):
 
 
 class TestSetBuffers(unittest.TestCase):
+    def setUp(self):
+        self.temp_dir = tempfile.TemporaryDirectory()
+        self.model_path = os.path.join(self.temp_dir.name, 'SetBuffersNet1')
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
+
     def test_set_buffers1(self):
         paddle.disable_static()
         net = SetBuffersNet1()
         out = net()
         self.assertEqual(out.numpy().tolist(), [2])
-        paddle.jit.save(net, './SetBuffersNet1')
+        paddle.jit.save(net, self.model_path)
         paddle.enable_static()
 
     def test_set_buffers2(self):

python/paddle/fluid/tests/unittests/dygraph_to_static/test_for_enumerate.py

@@ -16,6 +16,8 @@ from __future__ import print_function
 
 import numpy as np
 import unittest
+import os
+import tempfile
 
 import paddle
 import paddle.fluid as fluid
@@ -532,12 +534,20 @@ class TestForOriginalTuple(TestTransformForOriginalList):
 
 
 class TestForZip(unittest.TestCase):
+    def setUp(self):
+        self.temp_dir = tempfile.TemporaryDirectory()
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
+
     def test_for_zip_error(self):
         with self.assertRaises(RuntimeError):
-            paddle.jit.save(for_zip_error, './for_zip_error')
+            model_path = os.path.join(self.temp_dir.name, 'for_zip_error')
+            paddle.jit.save(for_zip_error, model_path)
 
     def test_for_zip(self):
-        paddle.jit.save(for_zip, './for_zip')
+        model_path = os.path.join(self.temp_dir.name, 'for_zip')
+        paddle.jit.save(for_zip, model_path)
 
 
 if __name__ == '__main__':

python/paddle/fluid/tests/unittests/dygraph_to_static/test_grad.py

@@ -17,6 +17,8 @@ from __future__ import print_function
 import numpy as np
 import paddle
 import unittest
+import os
+import tempfile
 
 
 class GradLayer(paddle.nn.Layer):
@@ -88,8 +90,15 @@ class TestGradLinear(TestGrad):
         self.func = GradLinearLayer()
         self.x = paddle.ones(shape=[10, 2, 5], dtype='float32')
         self.x.stop_gradient = False
-        self.infer_model_path = "double_grad_infer_model"
-        self.train_model_path = "double_grad_train_model"
+
+        self.temp_dir = tempfile.TemporaryDirectory()
+        self.infer_model_path = os.path.join(self.temp_dir.name,
+                                             'double_grad_infer_model')
+        self.train_model_path = os.path.join(self.temp_dir.name,
+                                             'double_grad_train_model')
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
 
     def test_save_infer_program(self):
         input_spec = [
@@ -113,7 +122,7 @@ class TestGradLinear(TestGrad):
             avg_loss = paddle.mean(paddle.abs(out - 1))
             avg_loss.backward()
             optimizer.minimize(avg_loss)
-            print(self.x.grad.mean())
+
             self.func.clear_gradients()
 
         paddle.jit.save(self.func, self.train_model_path)
@@ -129,8 +138,15 @@ class TestNoGradLinear(TestGradLinear):
         self.func = NoGradLinearLayer()
         self.x = paddle.ones(shape=[10, 2, 5], dtype='float32')
         self.x.stop_gradient = False
-        self.infer_model_path = "no_grad_infer_model"
-        self.train_model_path = "no_grad_train_model"
+
+        self.temp_dir = tempfile.TemporaryDirectory()
+        self.infer_model_path = os.path.join(self.temp_dir.name,
+                                             'no_grad_infer_model')
+        self.train_model_path = os.path.join(self.temp_dir.name,
+                                             'no_grad_train_model')
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
 
 
 if __name__ == '__main__':

python/paddle/fluid/tests/unittests/dygraph_to_static/test_lac.py

@@ -19,6 +19,7 @@ import numpy as np
 import unittest
 
 import os
+import tempfile
 os.environ["CUDA_VISIBLE_DEVICES"] = "2"
 
 import paddle
@@ -406,11 +407,6 @@ class Args(object):
     base_learning_rate = 0.01
     bigru_num = 2
     print_steps = 1
-    model_save_dir = "./inference"
-    model_save_prefix = "./inference/lac"
-    model_filename = "lac" + INFER_MODEL_SUFFIX
-    params_filename = "lac" + INFER_PARAMS_SUFFIX
-    dy_param_path = "./lac_dy_param"
 
 
 def get_random_input_data(batch_size, vocab_size, num_labels, max_seq_len=64):
@@ -458,84 +454,86 @@ def create_dataloader(reader, place):
     return data_loader
 
 
-def do_train(args, to_static):
-    program_translator.enable(to_static)
-    place = fluid.CUDAPlace(0) if fluid.is_compiled_with_cuda(
-    ) else fluid.CPUPlace()
-    with fluid.dygraph.guard(place):
-        paddle.seed(SEED)
-        paddle.framework.random._manual_program_seed(SEED)
-
-        reader = get_random_input_data(args.batch_size, args.vocab_size,
-                                       args.num_labels)
-        train_loader = create_dataloader(reader, place)
-
-        model = LexNet(args)
-        optimizer = fluid.optimizer.AdamOptimizer(
-            learning_rate=args.base_learning_rate,
-            parameter_list=model.parameters())
-        chunk_eval = ChunkEval(
-            int(math.ceil((args.num_labels - 1) / 2.0)), "IOB")
-
-        step = 0
-        chunk_evaluator = fluid.metrics.ChunkEvaluator()
-        chunk_evaluator.reset()
-
-        loss_data = []
-        for epoch_id in range(args.epoch):
-            for batch in train_loader():
-                words, targets, length = batch
-                start_time = time.time()
-                avg_cost, crf_decode = model(words, targets, length)
-                loss_data.append(avg_cost.numpy()[0])
-
-                # backward and optimization
-                avg_cost.backward()
-                optimizer.minimize(avg_cost)
-                model.clear_gradients()
-                end_time = time.time()
-
-                if step % args.print_steps == 0:
-                    (precision, recall, f1_score, num_infer_chunks,
-                     num_label_chunks, num_correct_chunks) = chunk_eval(
-                         input=crf_decode, label=targets, seq_length=length)
-                    outputs = [avg_cost, precision, recall, f1_score]
-                    avg_cost, precision, recall, f1_score = [
-                        np.mean(x.numpy()) for x in outputs
-                    ]
-
-                    print(
-                        "[train] step = %d, loss = %f, P: %f, R: %f, F1: %f, elapsed time %f"
-                        % (step, avg_cost, precision, recall, f1_score,
-                           end_time - start_time))
-
-                step += 1
-        # save inference model
-        if to_static:
-            fluid.dygraph.jit.save(
-                layer=model,
-                path=args.model_save_prefix,
-                input_spec=[input_specs[0], input_specs[-1]],
-                output_spec=[crf_decode])
-        else:
-            fluid.dygraph.save_dygraph(model.state_dict(), args.dy_param_path)
-
-        return np.array(loss_data)
-
-
 class TestLACModel(unittest.TestCase):
     def setUp(self):
         self.args = Args()
         self.place = fluid.CUDAPlace(0) if fluid.is_compiled_with_cuda(
         ) else fluid.CPUPlace()
+        self.temp_dir = tempfile.TemporaryDirectory()
+        self.model_save_dir = os.path.join(self.temp_dir.name, 'inference')
+        self.model_save_prefix = os.path.join(self.model_save_dir, 'lac')
+        self.model_filename = "lac" + INFER_MODEL_SUFFIX
+        self.params_filename = "lac" + INFER_PARAMS_SUFFIX
+        self.dy_param_path = os.path.join(self.temp_dir.name, 'lac_dy_param')
+
+    def train(self, args, to_static):
+        program_translator.enable(to_static)
+        place = fluid.CUDAPlace(0) if fluid.is_compiled_with_cuda(
+        ) else fluid.CPUPlace()
+        with fluid.dygraph.guard(place):
+            paddle.seed(SEED)
+            paddle.framework.random._manual_program_seed(SEED)
+
+            reader = get_random_input_data(args.batch_size, args.vocab_size,
+                                           args.num_labels)
+            train_loader = create_dataloader(reader, place)
+
+            model = LexNet(args)
+            optimizer = fluid.optimizer.AdamOptimizer(
+                learning_rate=args.base_learning_rate,
+                parameter_list=model.parameters())
+            chunk_eval = ChunkEval(
+                int(math.ceil((args.num_labels - 1) / 2.0)), "IOB")
+
+            step = 0
+            chunk_evaluator = fluid.metrics.ChunkEvaluator()
+            chunk_evaluator.reset()
+
+            loss_data = []
+            for epoch_id in range(args.epoch):
+                for batch in train_loader():
+                    words, targets, length = batch
+                    start_time = time.time()
+                    avg_cost, crf_decode = model(words, targets, length)
+                    loss_data.append(avg_cost.numpy()[0])
+
+                    # backward and optimization
+                    avg_cost.backward()
+                    optimizer.minimize(avg_cost)
+                    model.clear_gradients()
+                    end_time = time.time()
+
+                    if step % args.print_steps == 0:
+                        (precision, recall, f1_score, num_infer_chunks,
+                         num_label_chunks, num_correct_chunks) = chunk_eval(
+                             input=crf_decode, label=targets, seq_length=length)
+                        outputs = [avg_cost, precision, recall, f1_score]
+                        avg_cost, precision, recall, f1_score = [
+                            np.mean(x.numpy()) for x in outputs
+                        ]
+
+                        print(
+                            "[train] step = %d, loss = %f, P: %f, R: %f, F1: %f, elapsed time %f"
+                            % (step, avg_cost, precision, recall, f1_score,
+                               end_time - start_time))
+
+                    step += 1
+            # save inference model
+            if to_static:
+                fluid.dygraph.jit.save(
+                    layer=model,
+                    path=self.model_save_prefix,
+                    input_spec=[input_specs[0], input_specs[-1]],
+                    output_spec=[crf_decode])
+            else:
+                fluid.dygraph.save_dygraph(model.state_dict(),
+                                           self.dy_param_path)
 
-    def train(self, to_static):
-        out = do_train(self.args, to_static)
-        return out
+            return np.array(loss_data)
 
     def test_train(self):
-        st_out = self.train(to_static=True)
-        dy_out = self.train(to_static=False)
+        st_out = self.train(self.args, to_static=True)
+        dy_out = self.train(self.args, to_static=False)
         self.assertTrue(
             np.allclose(dy_out, st_out),
             msg="dygraph output:\n{},\nstatic output:\n {}.".format(dy_out,
@@ -565,8 +563,7 @@ class TestLACModel(unittest.TestCase):
         with fluid.dygraph.guard(self.place):
             model = LexNet(self.args)
             # load dygraph trained parameters
-            model_dict, _ = fluid.load_dygraph(self.args.dy_param_path +
-                                               ".pdparams")
+            model_dict, _ = fluid.load_dygraph(self.dy_param_path + ".pdparams")
             model.set_dict(model_dict)
             model.eval()
 
@@ -585,10 +582,10 @@ class TestLACModel(unittest.TestCase):
         # load inference model
         [inference_program, feed_target_names,
          fetch_targets] = fluid.io.load_inference_model(
-             self.args.model_save_dir,
+             self.model_save_dir,
              executor=exe,
-             model_filename=self.args.model_filename,
-             params_filename=self.args.params_filename)
+             model_filename=self.model_filename,
+             params_filename=self.params_filename)
 
         words, targets, length = batch
         pred_res = exe.run(
@@ -601,7 +598,7 @@ class TestLACModel(unittest.TestCase):
     def predict_dygraph_jit(self, batch):
         words, targets, length = batch
         with fluid.dygraph.guard(self.place):
-            model = fluid.dygraph.jit.load(self.args.model_save_prefix)
+            model = fluid.dygraph.jit.load(self.model_save_prefix)
             model.eval()
 
             pred_res = model(to_variable(words), to_variable(length))

python/paddle/fluid/tests/unittests/dygraph_to_static/test_layer_hook.py

@@ -14,8 +14,9 @@
 
 import unittest
 import paddle
-
+import os
 import numpy as np
+import tempfile
 
 
 def forward_post_hook1(layer, input, output):
@@ -54,7 +55,11 @@ class TestNestLayerHook(unittest.TestCase):
     def setUp(self):
         paddle.seed(2022)
         self.x = paddle.randn([4, 10])
-        self.path = "./net_hook"
+        self.temp_dir = tempfile.TemporaryDirectory()
+        self.path = os.path.join(self.temp_dir.name, 'net_hook')
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
 
     def train_net(self, to_static=False):
         paddle.seed(2022)

python/paddle/fluid/tests/unittests/dygraph_to_static/test_lstm.py

@@ -16,6 +16,8 @@ import numpy as np
 import paddle
 import unittest
 from paddle import nn
+import os
+import tempfile
 
 
 class LSTMLayer(nn.Layer):
@@ -40,6 +42,12 @@ class Net(nn.Layer):
 
 
 class TestLstm(unittest.TestCase):
+    def setUp(self):
+        self.temp_dir = tempfile.TemporaryDirectory()
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
+
     def run_lstm(self, to_static):
         paddle.jit.ProgramTranslator().enable(to_static)
 
@@ -78,11 +86,12 @@ class TestLstm(unittest.TestCase):
         x = paddle.randn((2, 10, 12))
         net = paddle.jit.to_static(
             net, input_spec=[paddle.static.InputSpec(shape=[-1, 10, 12])])
-        paddle.jit.save(net, 'simple_lstm')
+        model_path = os.path.join(self.temp_dir.name, 'simple_lstm')
+        paddle.jit.save(net, model_path)
 
         dygraph_out = net(x)
         # load saved model
-        load_net = paddle.jit.load('simple_lstm')
+        load_net = paddle.jit.load(model_path)
 
         static_out = load_net(x)
         self.assertTrue(
@@ -115,6 +124,12 @@ class LinearNet(nn.Layer):
 
 
 class TestSaveInEvalMode(unittest.TestCase):
+    def setUp(self):
+        self.temp_dir = tempfile.TemporaryDirectory()
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
+
     def test_save_in_eval(self):
         paddle.jit.ProgramTranslator().enable(True)
         net = LinearNet()
@@ -131,9 +146,11 @@ class TestSaveInEvalMode(unittest.TestCase):
         # save directly
         net = paddle.jit.to_static(
             net, input_spec=[paddle.static.InputSpec(shape=[-1, 10])])
-        paddle.jit.save(net, 'linear_net')
+
+        model_path = os.path.join(self.temp_dir.name, 'linear_net')
+        paddle.jit.save(net, model_path)
         # load saved model
-        load_net = paddle.jit.load('linear_net')
+        load_net = paddle.jit.load(model_path)
 
         x = paddle.randn((2, 10))
         eval_out = net(x)
@@ -146,6 +163,12 @@ class TestSaveInEvalMode(unittest.TestCase):
 
 
 class TestEvalAfterSave(unittest.TestCase):
+    def setUp(self):
+        self.temp_dir = tempfile.TemporaryDirectory()
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
+
     def test_eval_after_save(self):
         x = paddle.randn((2, 10, 12)).astype('float32')
         net = Net(12, 2)
@@ -159,8 +182,9 @@ class TestEvalAfterSave(unittest.TestCase):
         x = paddle.randn((2, 10, 12)).astype('float32')
         dy_out = net(x)
         # save model
-        paddle.jit.save(net, 'jit.save/lstm', input_spec=[x])
-        load_net = paddle.jit.load('jit.save/lstm')
+        model_path = os.path.join(self.temp_dir.name, 'jit.save/lstm')
+        paddle.jit.save(net, model_path, input_spec=[x])
+        load_net = paddle.jit.load(model_path)
         load_out = load_net(x)
         self.assertTrue(np.allclose(dy_out.numpy(), load_out.numpy()))
         # eval

python/paddle/fluid/tests/unittests/dygraph_to_static/test_mnist.py

@@ -15,6 +15,8 @@
 from __future__ import print_function
 
 import unittest
+import os
+import tempfile
 from time import time
 
 import numpy as np
@@ -134,6 +136,10 @@ class TestMNIST(unittest.TestCase):
             paddle.dataset.mnist.train(),
             batch_size=self.batch_size,
             drop_last=True)
+        self.temp_dir = tempfile.TemporaryDirectory()
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
 
 
 class TestMNISTWithToStatic(TestMNIST):
@@ -227,9 +233,10 @@ class TestMNISTWithToStatic(TestMNIST):
 
     def check_jit_save_load(self, model, inputs, input_spec, to_static, gt_out):
         if to_static:
-            infer_model_path = "./test_mnist_inference_model_by_jit_save"
-            model_save_dir = "./inference"
-            model_save_prefix = "./inference/mnist"
+            infer_model_path = os.path.join(
+                self.temp_dir.name, 'test_mnist_inference_model_by_jit_save')
+            model_save_dir = os.path.join(self.temp_dir.name, 'inference')
+            model_save_prefix = os.path.join(model_save_dir, 'mnist')
             model_filename = "mnist" + INFER_MODEL_SUFFIX
             params_filename = "mnist" + INFER_PARAMS_SUFFIX
             fluid.dygraph.jit.save(

python/paddle/fluid/tests/unittests/dygraph_to_static/test_mobile_net.py

@@ -12,6 +12,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import os
+import tempfile
 import time
 import numpy as np
 import paddle
@@ -439,11 +441,11 @@ class Args(object):
     train_step = 10
     place = fluid.CUDAPlace(0) if fluid.is_compiled_with_cuda(
     ) else fluid.CPUPlace()
-    model_save_dir = "./inference"
-    model_save_prefix = "./inference/" + model
-    model_filename = model + INFER_MODEL_SUFFIX
-    params_filename = model + INFER_PARAMS_SUFFIX
-    dy_state_dict_save_path = model + ".dygraph"
+    model_save_dir = None
+    model_save_prefix = None
+    model_filename = None
+    params_filename = None
+    dy_state_dict_save_path = None
 
 
 def train_mobilenet(args, to_static):
@@ -571,13 +573,21 @@ def predict_analysis_inference(args, data):
 class TestMobileNet(unittest.TestCase):
     def setUp(self):
         self.args = Args()
+        self.temp_dir = tempfile.TemporaryDirectory()
+        self.args.model_save_dir = os.path.join(self.temp_dir.name,
+                                                "./inference")
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
 
     def train(self, model_name, to_static):
         self.args.model = model_name
-        self.args.model_save_prefix = "./inference/" + model_name
+        self.args.model_save_prefix = os.path.join(self.temp_dir.name,
+                                                   "./inference/" + model_name)
         self.args.model_filename = model_name + INFER_MODEL_SUFFIX
         self.args.params_filename = model_name + INFER_PARAMS_SUFFIX
-        self.args.dy_state_dict_save_path = model_name + ".dygraph"
+        self.args.dy_state_dict_save_path = os.path.join(
+            self.temp_dir.name, model_name + ".dygraph")
         out = train_mobilenet(self.args, to_static)
         return out
 
@@ -590,10 +600,12 @@ class TestMobileNet(unittest.TestCase):
 
     def assert_same_predict(self, model_name):
         self.args.model = model_name
-        self.args.model_save_prefix = "./inference/" + model_name
+        self.args.model_save_prefix = os.path.join(self.temp_dir.name,
+                                                   "./inference/" + model_name)
         self.args.model_filename = model_name + INFER_MODEL_SUFFIX
         self.args.params_filename = model_name + INFER_PARAMS_SUFFIX
-        self.args.dy_state_dict_save_path = model_name + ".dygraph"
+        self.args.dy_state_dict_save_path = os.path.join(
+            self.temp_dir.name, model_name + ".dygraph")
         local_random = np.random.RandomState(SEED)
         image = local_random.random_sample([1, 3, 224, 224]).astype('float32')
         dy_pre = predict_dygraph(self.args, image)

python/paddle/fluid/tests/unittests/dygraph_to_static/test_resnet_v2.py

@@ -19,6 +19,7 @@ os.environ["FLAGS_enable_eager_mode"] = "0"
 import math
 import time
 import unittest
+import tempfile
 
 import numpy as np
 
@@ -37,11 +38,6 @@ epoch_num = 1
 place = paddle.CUDAPlace(0) if paddle.is_compiled_with_cuda() \
     else paddle.CPUPlace()
 
-MODEL_SAVE_DIR = "./inference"
-MODEL_SAVE_PREFIX = "./inference/resnet_v2"
-MODEL_FILENAME = "resnet_v2" + paddle.fluid.dygraph.io.INFER_MODEL_SUFFIX
-PARAMS_FILENAME = "resnet_v2" + paddle.fluid.dygraph.io.INFER_PARAMS_SUFFIX
-DY_STATE_DICT_SAVE_PATH = "./resnet_v2.dygraph"
 program_translator = paddle.jit.ProgramTranslator()
 
 if paddle.is_compiled_with_cuda():
@@ -210,133 +206,145 @@ def reader_decorator(reader):
     return __reader__
 
 
-def train(to_static):
-    """
-    Tests model decorated by `dygraph_to_static_output` in static mode. For users, the model is defined in dygraph mode and trained in static mode.
-    """
-    paddle.disable_static(place)
-    np.random.seed(SEED)
-    paddle.seed(SEED)
-    paddle.framework.random._manual_program_seed(SEED)
-
-    train_reader = paddle.batch(
-        reader_decorator(paddle.dataset.flowers.train(use_xmap=False)),
-        batch_size=batch_size,
-        drop_last=True)
-    data_loader = paddle.io.DataLoader.from_generator(capacity=5, iterable=True)
-    data_loader.set_sample_list_generator(train_reader)
-
-    resnet = ResNet()
-    optimizer = optimizer_setting(parameter_list=resnet.parameters())
-
-    for epoch in range(epoch_num):
-        total_loss = 0.0
-        total_acc1 = 0.0
-        total_acc5 = 0.0
-        total_sample = 0
-
-        for batch_id, data in enumerate(data_loader()):
-            start_time = time.time()
-            img, label = data
-
-            pred = resnet(img)
-            loss = paddle.nn.functional.cross_entropy(input=pred, label=label)
-            avg_loss = paddle.mean(x=loss)
-            acc_top1 = paddle.metric.accuracy(input=pred, label=label, k=1)
-            acc_top5 = paddle.metric.accuracy(input=pred, label=label, k=5)
-
-            avg_loss.backward()
-            optimizer.minimize(avg_loss)
-            resnet.clear_gradients()
-
-            total_loss += avg_loss
-            total_acc1 += acc_top1
-            total_acc5 += acc_top5
-            total_sample += 1
-
-            end_time = time.time()
-            if batch_id % 2 == 0:
-                print( "epoch %d | batch step %d, loss %0.3f, acc1 %0.3f, acc5 %0.3f, time %f" % \
-                    ( epoch, batch_id, total_loss.numpy() / total_sample, \
-                        total_acc1.numpy() / total_sample, total_acc5.numpy() / total_sample, end_time-start_time))
-            if batch_id == 10:
-                if to_static:
-                    paddle.jit.save(resnet, MODEL_SAVE_PREFIX)
-                else:
-                    paddle.fluid.dygraph.save_dygraph(resnet.state_dict(),
-                                                      DY_STATE_DICT_SAVE_PATH)
-                    # avoid dataloader throw abort signaal
-                data_loader._reset()
-                break
-    paddle.enable_static()
-
-    return total_loss.numpy()
-
-
-def predict_dygraph(data):
-    program_translator.enable(False)
-    paddle.disable_static(place)
-    resnet = ResNet()
-
-    model_dict, _ = paddle.fluid.dygraph.load_dygraph(DY_STATE_DICT_SAVE_PATH)
-    resnet.set_dict(model_dict)
-    resnet.eval()
-
-    pred_res = resnet(
-        paddle.to_tensor(
-            data=data, dtype=None, place=None, stop_gradient=True))
-
-    ret = pred_res.numpy()
-    paddle.enable_static()
-    return ret
-
-
-def predict_static(data):
-    exe = paddle.static.Executor(place)
-    [inference_program, feed_target_names,
-     fetch_targets] = paddle.static.load_inference_model(
-         MODEL_SAVE_DIR,
-         executor=exe,
-         model_filename=MODEL_FILENAME,
-         params_filename=PARAMS_FILENAME)
-
-    pred_res = exe.run(inference_program,
-                       feed={feed_target_names[0]: data},
-                       fetch_list=fetch_targets)
-
-    return pred_res[0]
-
-
-def predict_dygraph_jit(data):
-    paddle.disable_static(place)
-    resnet = paddle.jit.load(MODEL_SAVE_PREFIX)
-    resnet.eval()
-
-    pred_res = resnet(data)
-
-    ret = pred_res.numpy()
-    paddle.enable_static()
-    return ret
-
-
-def predict_analysis_inference(data):
-    output = PredictorTools(MODEL_SAVE_DIR, MODEL_FILENAME, PARAMS_FILENAME,
-                            [data])
-    out = output()
-    return out
-
-
 class TestResnet(unittest.TestCase):
+    def setUp(self):
+        self.temp_dir = tempfile.TemporaryDirectory()
+
+        self.model_save_dir = os.path.join(self.temp_dir.name, "./inference")
+        self.model_save_prefix = os.path.join(self.temp_dir.name,
+                                              "./inference/resnet_v2")
+        self.model_filename = "resnet_v2" + paddle.fluid.dygraph.io.INFER_MODEL_SUFFIX
+        self.params_filename = "resnet_v2" + paddle.fluid.dygraph.io.INFER_PARAMS_SUFFIX
+        self.dy_state_dict_save_path = os.path.join(self.temp_dir.name,
+                                                    "./resnet_v2.dygraph")
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
+
+    def do_train(self, to_static):
+        """
+        Tests model decorated by `dygraph_to_static_output` in static mode. For users, the model is defined in dygraph mode and trained in static mode.
+        """
+        paddle.disable_static(place)
+        np.random.seed(SEED)
+        paddle.seed(SEED)
+        paddle.framework.random._manual_program_seed(SEED)
+
+        train_reader = paddle.batch(
+            reader_decorator(paddle.dataset.flowers.train(use_xmap=False)),
+            batch_size=batch_size,
+            drop_last=True)
+        data_loader = paddle.io.DataLoader.from_generator(
+            capacity=5, iterable=True)
+        data_loader.set_sample_list_generator(train_reader)
+
+        resnet = ResNet()
+        optimizer = optimizer_setting(parameter_list=resnet.parameters())
+
+        for epoch in range(epoch_num):
+            total_loss = 0.0
+            total_acc1 = 0.0
+            total_acc5 = 0.0
+            total_sample = 0
+
+            for batch_id, data in enumerate(data_loader()):
+                start_time = time.time()
+                img, label = data
+
+                pred = resnet(img)
+                loss = paddle.nn.functional.cross_entropy(
+                    input=pred, label=label)
+                avg_loss = paddle.mean(x=loss)
+                acc_top1 = paddle.metric.accuracy(input=pred, label=label, k=1)
+                acc_top5 = paddle.metric.accuracy(input=pred, label=label, k=5)
+
+                avg_loss.backward()
+                optimizer.minimize(avg_loss)
+                resnet.clear_gradients()
+
+                total_loss += avg_loss
+                total_acc1 += acc_top1
+                total_acc5 += acc_top5
+                total_sample += 1
+
+                end_time = time.time()
+                if batch_id % 2 == 0:
+                    print( "epoch %d | batch step %d, loss %0.3f, acc1 %0.3f, acc5 %0.3f, time %f" % \
+                        ( epoch, batch_id, total_loss.numpy() / total_sample, \
+                            total_acc1.numpy() / total_sample, total_acc5.numpy() / total_sample, end_time-start_time))
+                if batch_id == 10:
+                    if to_static:
+                        paddle.jit.save(resnet, self.model_save_prefix)
+                    else:
+                        paddle.fluid.dygraph.save_dygraph(
+                            resnet.state_dict(), self.dy_state_dict_save_path)
+                        # avoid dataloader throw abort signaal
+                    data_loader._reset()
+                    break
+        paddle.enable_static()
+
+        return total_loss.numpy()
+
+    def predict_dygraph(self, data):
+        program_translator.enable(False)
+        paddle.disable_static(place)
+        resnet = ResNet()
+
+        model_dict, _ = paddle.fluid.dygraph.load_dygraph(
+            self.dy_state_dict_save_path)
+        resnet.set_dict(model_dict)
+        resnet.eval()
+
+        pred_res = resnet(
+            paddle.to_tensor(
+                data=data, dtype=None, place=None, stop_gradient=True))
+
+        ret = pred_res.numpy()
+        paddle.enable_static()
+        return ret
+
+    def predict_static(self, data):
+        exe = paddle.static.Executor(place)
+        [inference_program, feed_target_names,
+         fetch_targets] = paddle.static.load_inference_model(
+             self.model_save_dir,
+             executor=exe,
+             model_filename=self.model_filename,
+             params_filename=self.params_filename)
+
+        pred_res = exe.run(inference_program,
+                           feed={feed_target_names[0]: data},
+                           fetch_list=fetch_targets)
+
+        return pred_res[0]
+
+    def predict_dygraph_jit(self, data):
+        paddle.disable_static(place)
+        resnet = paddle.jit.load(self.model_save_prefix)
+        resnet.eval()
+
+        pred_res = resnet(data)
+
+        ret = pred_res.numpy()
+        paddle.enable_static()
+        return ret
+
+    def predict_analysis_inference(self, data):
+        output = PredictorTools(self.model_save_dir, self.model_filename,
+                                self.params_filename, [data])
+        out = output()
+        return out
+
     def train(self, to_static):
         program_translator.enable(to_static)
-        return train(to_static)
+        return self.do_train(to_static)
 
     def verify_predict(self):
         image = np.random.random([1, 3, 224, 224]).astype('float32')
-        dy_pre = predict_dygraph(image)
-        st_pre = predict_static(image)
-        dy_jit_pre = predict_dygraph_jit(image)
-        predictor_pre = predict_analysis_inference(image)
+        dy_pre = self.predict_dygraph(image)
+        st_pre = self.predict_static(image)
+        dy_jit_pre = self.predict_dygraph_jit(image)
+        predictor_pre = self.predict_analysis_inference(image)
         self.assertTrue(
             np.allclose(dy_pre, st_pre),
             msg="dy_pre:\n {}\n, st_pre: \n{}.".format(dy_pre, st_pre))
@@ -361,7 +369,7 @@ class TestResnet(unittest.TestCase):
         paddle.fluid.set_flags({'FLAGS_use_mkldnn': True})
         try:
             if paddle.fluid.core.is_compiled_with_mkldnn():
-                train(to_static=True)
+                self.train(to_static=True)
         finally:
             paddle.fluid.set_flags({'FLAGS_use_mkldnn': False})
 

python/paddle/fluid/tests/unittests/dygraph_to_static/test_save_inference_model.py

@@ -15,6 +15,7 @@
 from __future__ import print_function
 
 import os
+import tempfile
 import unittest
 import numpy as np
 
@@ -48,6 +49,12 @@ class SimpleFcLayer(fluid.dygraph.Layer):
 
 
 class TestDyToStaticSaveInferenceModel(unittest.TestCase):
+    def setUp(self):
+        self.temp_dir = tempfile.TemporaryDirectory()
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
+
     def test_save_inference_model(self):
         fc_size = 20
         x_data = np.random.random((fc_size, fc_size)).astype('float32')
@@ -66,8 +73,10 @@ class TestDyToStaticSaveInferenceModel(unittest.TestCase):
                 adam.minimize(loss)
                 layer.clear_gradients()
             # test for saving model in dygraph.guard
-            infer_model_prefix = "./test_dy2stat_inference_in_guard/model"
-            infer_model_dir = "./test_dy2stat_inference_in_guard"
+            infer_model_prefix = os.path.join(
+                self.temp_dir.name, "test_dy2stat_inference_in_guard/model")
+            infer_model_dir = os.path.join(self.temp_dir.name,
+                                           "test_dy2stat_inference_in_guard")
             fluid.dygraph.jit.save(
                 layer=layer,
                 path=infer_model_prefix,
@@ -90,8 +99,10 @@ class TestDyToStaticSaveInferenceModel(unittest.TestCase):
 
         expected_persistable_vars = set([p.name for p in model.parameters()])
 
-        infer_model_prefix = "./test_dy2stat_inference/model"
-        infer_model_dir = "./test_dy2stat_inference"
+        infer_model_prefix = os.path.join(self.temp_dir.name,
+                                          "test_dy2stat_inference/model")
+        infer_model_dir = os.path.join(self.temp_dir.name,
+                                       "test_dy2stat_inference")
         model_filename = "model" + INFER_MODEL_SUFFIX
         params_filename = "model" + INFER_PARAMS_SUFFIX
         fluid.dygraph.jit.save(

python/paddle/fluid/tests/unittests/dygraph_to_static/test_save_load.py

@@ -15,6 +15,8 @@
 from __future__ import print_function
 
 import unittest
+import os
+import tempfile
 
 import numpy as np
 import paddle.fluid as fluid
@@ -30,6 +32,14 @@ place = fluid.CUDAPlace(0) if fluid.is_compiled_with_cuda() else fluid.CPUPlace(
 
 
 class TestDyToStaticSaveLoad(unittest.TestCase):
+    def setUp(self):
+        self.temp_dir = tempfile.TemporaryDirectory()
+        self.model_path = os.path.join(self.temp_dir.name,
+                                       "test_dy2stat_save_load")
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
+
     def test_save_load_same_result(self):
         program_translator = ProgramTranslator()
         x_data = np.random.randn(30, 10, 32).astype('float32')
@@ -50,7 +60,8 @@ class TestDyToStaticSaveLoad(unittest.TestCase):
                 adam.minimize(static_loss)
                 net.clear_gradients()
             # Save parameters
-            fluid.save_dygraph(net.state_dict(), "./test_dy2stat_save_load")
+
+            fluid.save_dygraph(net.state_dict(), self.model_path)
             # minimize() will update parameter, call net() to get output and avg_loss.
             # Switch into eval mode.
             net.eval()
@@ -61,7 +72,7 @@ class TestDyToStaticSaveLoad(unittest.TestCase):
             dygraph_net = Linear(32, 64)
 
             # Load parameters
-            model_dict, _ = fluid.load_dygraph("./test_dy2stat_save_load")
+            model_dict, _ = fluid.load_dygraph(self.model_path)
             dygraph_net.set_dict(model_dict)
             # Switch into eval mode.
             dygraph_net.eval()

python/paddle/fluid/tests/unittests/dygraph_to_static/test_se_resnet.py

@@ -16,6 +16,8 @@ import logging
 import math
 import time
 import unittest
+import os
+import tempfile
 import numpy as np
 
 import paddle
@@ -35,11 +37,6 @@ BATCH_SIZE = 8
 EPOCH_NUM = 1
 PRINT_STEP = 2
 STEP_NUM = 10
-MODEL_SAVE_DIR = "./inference"
-MODEL_SAVE_PREFIX = "./inference/se_resnet"
-MODEL_FILENAME = "se_resnet" + INFER_MODEL_SUFFIX
-PARAMS_FILENAME = "se_resnet" + INFER_PARAMS_SUFFIX
-DY_STATE_DICT_SAVE_PATH = "./se_resnet.dygraph"
 
 place = fluid.CUDAPlace(0) if fluid.is_compiled_with_cuda() \
     else fluid.CPUPlace()
@@ -327,129 +324,6 @@ class SeResNeXt(fluid.dygraph.Layer):
         return out, avg_loss, acc_top1, acc_top5
 
 
-def train(train_reader, to_static):
-    program_translator = ProgramTranslator()
-    program_translator.enable(to_static)
-
-    np.random.seed(SEED)
-
-    with fluid.dygraph.guard(place):
-        paddle.seed(SEED)
-        paddle.framework.random._manual_program_seed(SEED)
-        se_resnext = SeResNeXt()
-        optimizer = optimizer_setting(train_parameters, se_resnext.parameters())
-
-        for epoch_id in range(EPOCH_NUM):
-            total_loss = 0.0
-            total_acc1 = 0.0
-            total_acc5 = 0.0
-            total_sample = 0
-            step_idx = 0
-            speed_list = []
-            for step_id, data in enumerate(train_reader()):
-                dy_x_data = np.array([x[0].reshape(3, 224, 224)
-                                      for x in data]).astype('float32')
-                y_data = np.array([x[1] for x in data]).astype('int64').reshape(
-                    BATCH_SIZE, 1)
-
-                img = to_variable(dy_x_data)
-                label = to_variable(y_data)
-                label.stop_gradient = True
-
-                pred, avg_loss, acc_top1, acc_top5 = se_resnext(img, label)
-
-                dy_out = avg_loss.numpy()
-                avg_loss.backward()
-
-                optimizer.minimize(avg_loss)
-                se_resnext.clear_gradients()
-
-                lr = optimizer._global_learning_rate().numpy()
-                total_loss += dy_out
-                total_acc1 += acc_top1.numpy()
-                total_acc5 += acc_top5.numpy()
-                total_sample += 1
-                if step_id % PRINT_STEP == 0:
-                    if step_id == 0:
-                        logging.info( "epoch %d | step %d, loss %0.3f, acc1 %0.3f, acc5 %0.3f" % \
-                                      ( epoch_id, step_id, total_loss / total_sample, \
-                                        total_acc1 / total_sample, total_acc5 / total_sample))
-                        avg_batch_time = time.time()
-                    else:
-                        speed = PRINT_STEP / (time.time() - avg_batch_time)
-                        speed_list.append(speed)
-                        logging.info( "epoch %d | step %d, loss %0.3f, acc1 %0.3f, acc5 %0.3f, speed %.3f steps/s" % \
-                                      ( epoch_id, step_id, total_loss / total_sample, \
-                                        total_acc1 / total_sample, total_acc5 / total_sample, speed))
-                        avg_batch_time = time.time()
-
-                step_idx += 1
-                if step_idx == STEP_NUM:
-                    if to_static:
-                        fluid.dygraph.jit.save(
-                            se_resnext,
-                            MODEL_SAVE_PREFIX, [img],
-                            output_spec=[pred])
-                    else:
-                        fluid.dygraph.save_dygraph(se_resnext.state_dict(),
-                                                   DY_STATE_DICT_SAVE_PATH)
-                    break
-        return pred.numpy(), avg_loss.numpy(), acc_top1.numpy(), acc_top5.numpy(
-        )
-
-
-def predict_dygraph(data):
-    program_translator = ProgramTranslator()
-    program_translator.enable(False)
-    with fluid.dygraph.guard(place):
-        se_resnext = SeResNeXt()
-
-        model_dict, _ = fluid.dygraph.load_dygraph(DY_STATE_DICT_SAVE_PATH)
-        se_resnext.set_dict(model_dict)
-        se_resnext.eval()
-
-        label = np.random.random([1, 1]).astype("int64")
-        img = fluid.dygraph.to_variable(data)
-        label = fluid.dygraph.to_variable(label)
-        pred_res, _, _, _ = se_resnext(img, label)
-
-        return pred_res.numpy()
-
-
-def predict_static(data):
-    paddle.enable_static()
-    exe = fluid.Executor(place)
-    [inference_program, feed_target_names,
-     fetch_targets] = fluid.io.load_inference_model(
-         MODEL_SAVE_DIR,
-         executor=exe,
-         model_filename=MODEL_FILENAME,
-         params_filename=PARAMS_FILENAME)
-
-    pred_res = exe.run(inference_program,
-                       feed={feed_target_names[0]: data},
-                       fetch_list=fetch_targets)
-
-    return pred_res[0]
-
-
-def predict_dygraph_jit(data):
-    with fluid.dygraph.guard(place):
-        se_resnext = fluid.dygraph.jit.load(MODEL_SAVE_PREFIX)
-        se_resnext.eval()
-
-        pred_res = se_resnext(data)
-
-        return pred_res.numpy()
-
-
-def predict_analysis_inference(data):
-    output = PredictorTools(MODEL_SAVE_DIR, MODEL_FILENAME, PARAMS_FILENAME,
-                            [data])
-    out = output()
-    return out
-
-
 class TestSeResnet(unittest.TestCase):
     def setUp(self):
         self.train_reader = paddle.batch(
@@ -457,13 +331,148 @@ class TestSeResnet(unittest.TestCase):
                 use_xmap=False, cycle=True),
             batch_size=BATCH_SIZE,
             drop_last=True)
+        self.temp_dir = tempfile.TemporaryDirectory()
+
+        self.model_save_dir = os.path.join(self.temp_dir.name, "inference")
+        self.model_save_prefix = os.path.join(self.temp_dir.name,
+                                              "inference/se_resnet")
+        self.model_filename = "se_resnet" + INFER_MODEL_SUFFIX
+        self.params_filename = "se_resnet" + INFER_PARAMS_SUFFIX
+        self.dy_state_dict_save_path = os.path.join(self.temp_dir.name,
+                                                    "se_resnet.dygraph")
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
+
+    def train(self, train_reader, to_static):
+        program_translator = ProgramTranslator()
+        program_translator.enable(to_static)
+
+        np.random.seed(SEED)
+
+        with fluid.dygraph.guard(place):
+            paddle.seed(SEED)
+            paddle.framework.random._manual_program_seed(SEED)
+            se_resnext = SeResNeXt()
+            optimizer = optimizer_setting(train_parameters,
+                                          se_resnext.parameters())
+
+            for epoch_id in range(EPOCH_NUM):
+                total_loss = 0.0
+                total_acc1 = 0.0
+                total_acc5 = 0.0
+                total_sample = 0
+                step_idx = 0
+                speed_list = []
+                for step_id, data in enumerate(train_reader()):
+                    dy_x_data = np.array(
+                        [x[0].reshape(3, 224, 224)
+                         for x in data]).astype('float32')
+                    y_data = np.array(
+                        [x[1] for x in data]).astype('int64').reshape(
+                            BATCH_SIZE, 1)
+
+                    img = to_variable(dy_x_data)
+                    label = to_variable(y_data)
+                    label.stop_gradient = True
+
+                    pred, avg_loss, acc_top1, acc_top5 = se_resnext(img, label)
+
+                    dy_out = avg_loss.numpy()
+                    avg_loss.backward()
+
+                    optimizer.minimize(avg_loss)
+                    se_resnext.clear_gradients()
+
+                    lr = optimizer._global_learning_rate().numpy()
+                    total_loss += dy_out
+                    total_acc1 += acc_top1.numpy()
+                    total_acc5 += acc_top5.numpy()
+                    total_sample += 1
+                    if step_id % PRINT_STEP == 0:
+                        if step_id == 0:
+                            logging.info( "epoch %d | step %d, loss %0.3f, acc1 %0.3f, acc5 %0.3f" % \
+                                        ( epoch_id, step_id, total_loss / total_sample, \
+                                            total_acc1 / total_sample, total_acc5 / total_sample))
+                            avg_batch_time = time.time()
+                        else:
+                            speed = PRINT_STEP / (time.time() - avg_batch_time)
+                            speed_list.append(speed)
+                            logging.info( "epoch %d | step %d, loss %0.3f, acc1 %0.3f, acc5 %0.3f, speed %.3f steps/s" % \
+                                        ( epoch_id, step_id, total_loss / total_sample, \
+                                            total_acc1 / total_sample, total_acc5 / total_sample, speed))
+                            avg_batch_time = time.time()
+
+                    step_idx += 1
+                    if step_idx == STEP_NUM:
+                        if to_static:
+                            fluid.dygraph.jit.save(
+                                se_resnext,
+                                self.model_save_prefix, [img],
+                                output_spec=[pred])
+                        else:
+                            fluid.dygraph.save_dygraph(
+                                se_resnext.state_dict(),
+                                self.dy_state_dict_save_path)
+                        break
+            return pred.numpy(), avg_loss.numpy(), acc_top1.numpy(
+            ), acc_top5.numpy()
+
+    def predict_dygraph(self, data):
+        program_translator = ProgramTranslator()
+        program_translator.enable(False)
+        with fluid.dygraph.guard(place):
+            se_resnext = SeResNeXt()
+
+            model_dict, _ = fluid.dygraph.load_dygraph(
+                self.dy_state_dict_save_path)
+            se_resnext.set_dict(model_dict)
+            se_resnext.eval()
+
+            label = np.random.random([1, 1]).astype("int64")
+            img = fluid.dygraph.to_variable(data)
+            label = fluid.dygraph.to_variable(label)
+            pred_res, _, _, _ = se_resnext(img, label)
+
+            return pred_res.numpy()
+
+    def predict_static(self, data):
+        paddle.enable_static()
+        exe = fluid.Executor(place)
+        [inference_program, feed_target_names,
+         fetch_targets] = fluid.io.load_inference_model(
+             self.model_save_dir,
+             executor=exe,
+             model_filename=self.model_filename,
+             params_filename=self.params_filename)
+
+        pred_res = exe.run(inference_program,
+                           feed={feed_target_names[0]: data},
+                           fetch_list=fetch_targets)
+
+        return pred_res[0]
+
+    def predict_dygraph_jit(self, data):
+        with fluid.dygraph.guard(place):
+            se_resnext = fluid.dygraph.jit.load(self.model_save_prefix)
+            se_resnext.eval()
+
+            pred_res = se_resnext(data)
+
+            return pred_res.numpy()
+
+    def predict_analysis_inference(self, data):
+        output = PredictorTools(self.model_save_dir, self.model_filename,
+                                self.params_filename, [data])
+        out = output()
+        return out
 
     def verify_predict(self):
         image = np.random.random([1, 3, 224, 224]).astype('float32')
-        dy_pre = predict_dygraph(image)
-        st_pre = predict_static(image)
-        dy_jit_pre = predict_dygraph_jit(image)
-        predictor_pre = predict_analysis_inference(image)
+        dy_pre = self.predict_dygraph(image)
+        st_pre = self.predict_static(image)
+        dy_jit_pre = self.predict_dygraph_jit(image)
+        predictor_pre = self.predict_analysis_inference(image)
         self.assertTrue(
             np.allclose(dy_pre, st_pre),
             msg="dy_pre:\n {}\n, st_pre: \n{}.".format(dy_pre, st_pre))
@@ -483,9 +492,9 @@ class TestSeResnet(unittest.TestCase):
                     flat_predictor_pre[i], flat_st_pre[i]))
 
     def test_check_result(self):
-        pred_1, loss_1, acc1_1, acc5_1 = train(
+        pred_1, loss_1, acc1_1, acc5_1 = self.train(
             self.train_reader, to_static=False)
-        pred_2, loss_2, acc1_2, acc5_2 = train(
+        pred_2, loss_2, acc1_2, acc5_2 = self.train(
             self.train_reader, to_static=True)
 
         self.assertTrue(

python/paddle/fluid/tests/unittests/dygraph_to_static/test_seq2seq.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 import os
+import tempfile
 import time
 import unittest
 
@@ -22,7 +23,7 @@ from paddle.fluid.clip import GradientClipByGlobalNorm
 from paddle.fluid.dygraph.dygraph_to_static import ProgramTranslator
 
 from seq2seq_dygraph_model import BaseModel, AttentionModel
-from seq2seq_utils import Seq2SeqModelHyperParams as args
+from seq2seq_utils import Seq2SeqModelHyperParams
 from seq2seq_utils import get_data_iter
 place = fluid.CUDAPlace(0) if fluid.is_compiled_with_cuda() else fluid.CPUPlace(
 )
@@ -43,7 +44,7 @@ def prepare_input(batch):
     return inputs, np.sum(tar_mask)
 
 
-def train(attn_model=False):
+def train(args, attn_model=False):
     with fluid.dygraph.guard(place):
         fluid.default_startup_program().random_seed = 2020
         fluid.default_main_program().random_seed = 2020
@@ -117,7 +118,7 @@ def train(attn_model=False):
         return loss.numpy()
 
 
-def infer(attn_model=False):
+def infer(args, attn_model=False):
     with fluid.dygraph.guard(place):
 
         if attn_model:
@@ -160,19 +161,32 @@ def infer(attn_model=False):
 
 
 class TestSeq2seq(unittest.TestCase):
+    def setUp(self):
+        self.args = Seq2SeqModelHyperParams
+        self.temp_dir = tempfile.TemporaryDirectory()
+        self.args.base_model_path = os.path.join(self.temp_dir.name,
+                                                 self.args.base_model_path)
+        self.args.attn_model_path = os.path.join(self.temp_dir.name,
+                                                 self.args.attn_model_path)
+        self.args.reload_model = os.path.join(self.temp_dir.name,
+                                              self.args.reload_model)
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
+
     def run_dygraph(self, mode="train", attn_model=False):
         program_translator.enable(False)
         if mode == "train":
-            return train(attn_model)
+            return train(self.args, attn_model)
         else:
-            return infer(attn_model)
+            return infer(self.args, attn_model)
 
     def run_static(self, mode="train", attn_model=False):
         program_translator.enable(True)
         if mode == "train":
-            return train(attn_model)
+            return train(self.args, attn_model)
         else:
-            return infer(attn_model)
+            return infer(self.args, attn_model)
 
     def _test_train(self, attn_model=False):
         dygraph_loss = self.run_dygraph(mode="train", attn_model=attn_model)

python/paddle/fluid/tests/unittests/dygraph_to_static/test_slice.py

@@ -13,7 +13,8 @@
 # limitations under the License.
 
 from __future__ import print_function
-
+import os
+import tempfile
 import unittest
 import numpy as np
 
@@ -166,15 +167,20 @@ class TestSetValue(TestSliceWithoutControlFlow):
 
 
 class TestSetValueWithLayerAndSave(unittest.TestCase):
+    def setUp(self):
+        self.temp_dir = tempfile.TemporaryDirectory()
+        self.model_path = os.path.join(self.temp_dir.name,
+                                       "layer_use_set_value")
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
+
     def test_set_value_with_save(self):
         prog_trans.enable(True)
         model = LayerWithSetValue(input_dim=10, hidden=1)
         x = paddle.full(shape=[5, 10], fill_value=5.0, dtype="float32")
         paddle.jit.save(
-            layer=model,
-            path="./layer_use_set_value",
-            input_spec=[x],
-            output_spec=None)
+            layer=model, path=self.model_path, input_spec=[x], output_spec=None)
 
 
 class TestSliceSupplementSpecialCase(unittest.TestCase):

python/paddle/fluid/tests/unittests/dygraph_to_static/test_typing.py

@@ -11,7 +11,8 @@
 # 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 os
+import tempfile
 import paddle
 import unittest
 import numpy as np
@@ -72,11 +73,16 @@ class TestTyping(unittest.TestCase):
         self.x = paddle.randn([4, 16])
         self.spec = [paddle.static.InputSpec(shape=[None, 16], dtype='float32')]
 
+        self.temp_dir = tempfile.TemporaryDirectory()
+
+    def tearDown(self):
+        self.temp_dir.cleanup()
+
     def build_net(self):
         return LinearNetWithTuple(self.in_num, self.out_num)
 
     def save_and_load(self, suffix=''):
-        path = './layer_typing_' + suffix
+        path = os.path.join(self.temp_dir.name, 'layer_typing_' + suffix)
         paddle.jit.save(self.net, path, input_spec=self.spec)
         return paddle.jit.load(path)