diff --git a/python/paddle/fluid/tests/unittests/rnn/rnn_numpy.py b/python/paddle/fluid/tests/unittests/rnn/rnn_numpy.py index bfaf6430f27221473bcc914695f07305de37461b..dd1e18b89d29d0bb9446a0f8439c72d64e289266 100644 --- a/python/paddle/fluid/tests/unittests/rnn/rnn_numpy.py +++ b/python/paddle/fluid/tests/unittests/rnn/rnn_numpy.py @@ -414,9 +414,9 @@ class SimpleRNN(RNNMixin): time_major=False, dtype="float64"): super(SimpleRNN, self).__init__() - - if direction in ["forward", "backward"]: - is_reverse = direction == "backward" + bidirectional_list = ["bidirectional", "bidirect"] + if direction in ["forward"]: + is_reverse = False cell = SimpleRNNCell( input_size, hidden_size, nonlinearity=nonlinearity, dtype=dtype) self.append(RNN(cell, is_reverse, time_major)) @@ -427,7 +427,7 @@ class SimpleRNN(RNNMixin): nonlinearity=nonlinearity, dtype=dtype) self.append(RNN(cell, is_reverse, time_major)) - elif direction == "bidirectional": + elif direction in bidirectional_list: cell_fw = SimpleRNNCell( input_size, hidden_size, nonlinearity=nonlinearity, dtype=dtype) cell_bw = SimpleRNNCell( @@ -447,7 +447,7 @@ class SimpleRNN(RNNMixin): self.input_size = input_size self.hidden_size = hidden_size self.dropout = dropout - self.num_directions = 2 if direction == "bidirectional" else 1 + self.num_directions = 2 if direction in bidirectional_list else 1 self.time_major = time_major self.num_layers = num_layers self.state_components = 1 @@ -464,14 +464,15 @@ class LSTM(RNNMixin): dtype="float64"): super(LSTM, self).__init__() - if direction in ["forward", "backward"]: - is_reverse = direction == "backward" + bidirectional_list = ["bidirectional", "bidirect"] + if direction in ["forward"]: + is_reverse = False cell = LSTMCell(input_size, hidden_size, dtype=dtype) self.append(RNN(cell, is_reverse, time_major)) for i in range(1, num_layers): cell = LSTMCell(hidden_size, hidden_size, dtype=dtype) self.append(RNN(cell, is_reverse, time_major)) - elif direction == "bidirectional": + elif direction in bidirectional_list: cell_fw = LSTMCell(input_size, hidden_size, dtype=dtype) cell_bw = LSTMCell(input_size, hidden_size, dtype=dtype) self.append(BiRNN(cell_fw, cell_bw, time_major)) @@ -487,7 +488,7 @@ class LSTM(RNNMixin): self.input_size = input_size self.hidden_size = hidden_size self.dropout = dropout - self.num_directions = 2 if direction == "bidirectional" else 1 + self.num_directions = 2 if direction in bidirectional_list else 1 self.time_major = time_major self.num_layers = num_layers self.state_components = 2 @@ -504,14 +505,15 @@ class GRU(RNNMixin): dtype="float64"): super(GRU, self).__init__() - if direction in ["forward", "backward"]: - is_reverse = direction == "backward" + bidirectional_list = ["bidirectional", "bidirect"] + if direction in ["forward"]: + is_reverse = False cell = GRUCell(input_size, hidden_size, dtype=dtype) self.append(RNN(cell, is_reverse, time_major)) for i in range(1, num_layers): cell = GRUCell(hidden_size, hidden_size, dtype=dtype) self.append(RNN(cell, is_reverse, time_major)) - elif direction == "bidirectional": + elif direction in bidirectional_list: cell_fw = GRUCell(input_size, hidden_size, dtype=dtype) cell_bw = GRUCell(input_size, hidden_size, dtype=dtype) self.append(BiRNN(cell_fw, cell_bw, time_major)) @@ -527,7 +529,7 @@ class GRU(RNNMixin): self.input_size = input_size self.hidden_size = hidden_size self.dropout = dropout - self.num_directions = 2 if direction == "bidirectional" else 1 + self.num_directions = 2 if direction in bidirectional_list else 1 self.time_major = time_major self.num_layers = num_layers self.state_components = 1 diff --git a/python/paddle/fluid/tests/unittests/rnn/test_rnn_nets.py b/python/paddle/fluid/tests/unittests/rnn/test_rnn_nets.py index f0aa42495161e02ce2b03ec2d182820dd1b81a82..badabbd8ceabd889ec26e52d8e461cad7a3b90af 100755 --- a/python/paddle/fluid/tests/unittests/rnn/test_rnn_nets.py +++ b/python/paddle/fluid/tests/unittests/rnn/test_rnn_nets.py @@ -22,13 +22,15 @@ import unittest from convert import convert_params_for_net from rnn_numpy import SimpleRNN, LSTM, GRU +bidirectional_list = ["bidirectional", "bidirect"] + class TestSimpleRNN(unittest.TestCase): def __init__(self, time_major=True, direction="forward", place="cpu"): super(TestSimpleRNN, self).__init__("runTest") self.time_major = time_major self.direction = direction - self.num_directions = 2 if direction == "bidirectional" else 1 + self.num_directions = 2 if direction in bidirectional_list else 1 self.place = place def setUp(self): @@ -109,7 +111,7 @@ class TestGRU(unittest.TestCase): super(TestGRU, self).__init__("runTest") self.time_major = time_major self.direction = direction - self.num_directions = 2 if direction == "bidirectional" else 1 + self.num_directions = 2 if direction in bidirectional_list else 1 self.place = place def setUp(self): @@ -196,7 +198,7 @@ class TestLSTM(unittest.TestCase): super(TestLSTM, self).__init__("runTest") self.time_major = time_major self.direction = direction - self.num_directions = 2 if direction == "bidirectional" else 1 + self.num_directions = 2 if direction in bidirectional_list else 1 self.place = place def setUp(self): @@ -339,7 +341,7 @@ def load_tests(loader, tests, pattern): suite = unittest.TestSuite() devices = ["cpu", "gpu"] if paddle.fluid.is_compiled_with_cuda() \ else ["cpu"] - for direction in ["forward", "backward", "bidirectional"]: + for direction in ["forward", "bidirectional", "bidirect"]: for time_major in [True, False]: for device in devices: for test_class in [TestSimpleRNN, TestLSTM, TestGRU]: diff --git a/python/paddle/fluid/tests/unittests/rnn/test_rnn_nets_static.py b/python/paddle/fluid/tests/unittests/rnn/test_rnn_nets_static.py index 950d942b7917e17a106ddcb5bb071d193d965e63..5de539ebf39393ee4bf3baad8cd824066cd67aef 100755 --- a/python/paddle/fluid/tests/unittests/rnn/test_rnn_nets_static.py +++ b/python/paddle/fluid/tests/unittests/rnn/test_rnn_nets_static.py @@ -23,13 +23,15 @@ import unittest from convert import convert_params_for_net_static from rnn_numpy import SimpleRNN, LSTM, GRU +bidirectional_list = ["bidirectional", "bidirect"] + class TestSimpleRNN(unittest.TestCase): def __init__(self, time_major=True, direction="forward", place="cpu"): super(TestSimpleRNN, self).__init__("runTest") self.time_major = time_major self.direction = direction - self.num_directions = 2 if direction == "bidirectional" else 1 + self.num_directions = 2 if direction in bidirectional_list else 1 self.place = place def setUp(self): @@ -173,7 +175,7 @@ class TestGRU(unittest.TestCase): super(TestGRU, self).__init__("runTest") self.time_major = time_major self.direction = direction - self.num_directions = 2 if direction == "bidirectional" else 1 + self.num_directions = 2 if direction in bidirectional_list else 1 self.place = place def setUp(self): @@ -319,7 +321,7 @@ class TestLSTM(unittest.TestCase): super(TestLSTM, self).__init__("runTest") self.time_major = time_major self.direction = direction - self.num_directions = 2 if direction == "bidirectional" else 1 + self.num_directions = 2 if direction in bidirectional_list else 1 self.place = place def setUp(self): @@ -469,9 +471,13 @@ def load_tests(loader, tests, pattern): suite = unittest.TestSuite() devices = ["cpu", "gpu"] if paddle.fluid.is_compiled_with_cuda() \ else ["cpu"] - for direction in ["forward", "backward", "bidirectional"]: + for direction in ["forward", "bidirectional", "bidirect"]: for time_major in [True, False]: for device in devices: for test_class in [TestSimpleRNN, TestLSTM, TestGRU]: suite.addTest(test_class(time_major, direction, device)) return suite + + +if __name__ == "__main__": + unittest.main() diff --git a/python/paddle/nn/layer/rnn.py b/python/paddle/nn/layer/rnn.py index d06623a2b910f2cbc17cc13bd1108c53da5ecc3a..fefef52ba6b1988f5942c4258ff85d35421eee3a 100644 --- a/python/paddle/nn/layer/rnn.py +++ b/python/paddle/nn/layer/rnn.py @@ -858,11 +858,12 @@ class RNNBase(LayerList): bias_ih_attr=None, bias_hh_attr=None): super(RNNBase, self).__init__() + bidirectional_list = ["bidirectional", "bidirect"] self.mode = mode self.input_size = input_size self.hidden_size = hidden_size self.dropout = dropout - self.num_directions = 2 if direction == "bidirectional" else 1 + self.num_directions = 2 if direction in bidirectional_list else 1 self.time_major = time_major self.num_layers = num_layers self.state_components = 2 if mode == "LSTM" else 1 @@ -882,14 +883,14 @@ class RNNBase(LayerList): rnn_cls = SimpleRNNCell kwargs["activation"] = self.activation - if direction in ["forward", "backward"]: - is_reverse = direction == "backward" + if direction in ["forward"]: + is_reverse = False cell = rnn_cls(input_size, hidden_size, **kwargs) self.append(RNN(cell, is_reverse, time_major)) for i in range(1, num_layers): cell = rnn_cls(hidden_size, hidden_size, **kwargs) self.append(RNN(cell, is_reverse, time_major)) - elif direction == "bidirectional": + elif direction in bidirectional_list: cell_fw = rnn_cls(input_size, hidden_size, **kwargs) cell_bw = rnn_cls(input_size, hidden_size, **kwargs) self.append(BiRNN(cell_fw, cell_bw, time_major)) @@ -899,13 +900,12 @@ class RNNBase(LayerList): self.append(BiRNN(cell_fw, cell_bw, time_major)) else: raise ValueError( - "direction should be forward, backward or bidirectional, " + "direction should be forward or bidirect (or bidirectional), " "received direction = {}".format(direction)) self.could_use_cudnn = True - self.could_use_cudnn &= direction != "backward" self.could_use_cudnn &= len(self.parameters()) == num_layers * 4 * ( - 2 if direction == "bidirectional" else 1) + 2 if direction in bidirectional_list else 1) # Expose params as RNN's attribute, which can make it compatible when # replacing small ops composed rnn with cpp rnn kernel. @@ -1079,8 +1079,8 @@ class SimpleRNN(RNNBase): input_size (int): The input size for the first layer's cell. hidden_size (int): The hidden size for each layer's cell. num_layers (int, optional): Number of layers. Defaults to 1. - direction (str, optional): The direction of the network. It can be "forward", - "backward" and "bidirectional". When "bidirectional", the way to merge + direction (str, optional): The direction of the network. It can be "forward" + or "bidirect"(or "bidirectional"). When "bidirect", the way to merge outputs of forward and backward is concatenating. Defaults to "forward". time_major (bool, optional): Whether the first dimension of the input means the time steps. Defaults to False. @@ -1195,8 +1195,8 @@ class LSTM(RNNBase): input_size (int): The input size for the first layer's cell. hidden_size (int): The hidden size for each layer's cell. num_layers (int, optional): Number of layers. Defaults to 1. - direction (str, optional): The direction of the network. It can be "forward", - "backward" and "bidirectional". When "bidirectional", the way to merge + direction (str, optional): The direction of the network. It can be "forward" + or "bidirect"(or "bidirectional"). When "bidirect", the way to merge outputs of forward and backward is concatenating. Defaults to "forward". time_major (bool, optional): Whether the first dimension of the input means the time steps. Defaults to False. @@ -1300,8 +1300,8 @@ class GRU(RNNBase): input_size (int): The input size for the first layer's cell. hidden_size (int): The hidden size for each layer's cell. num_layers (int, optional): Number of layers. Defaults to 1. - direction (str, optional): The direction of the network. It can be "forward", - "backward" and "bidirectional". When "bidirectional", the way to merge + direction (str, optional): The direction of the network. It can be "forward" + or "bidirect"(or "bidirectional"). When "bidirect", the way to merge outputs of forward and backward is concatenating. Defaults to "forward". time_major (bool, optional): Whether the first dimension of the input means the time steps. Defaults to False. diff --git a/python/paddle/text/datasets/wmt14.py b/python/paddle/text/datasets/wmt14.py index 36cb6dfd3e5b7652da3e4e9233dd5b16076a53b6..b080824d72410e3f6a66f35a9368d2e0960d95b5 100644 --- a/python/paddle/text/datasets/wmt14.py +++ b/python/paddle/text/datasets/wmt14.py @@ -43,7 +43,7 @@ class WMT14(Dataset): Implementation of `WMT14 `_ test dataset. The original WMT14 dataset is too large and a small set of data for set is provided. This module will download dataset from - http://paddlepaddle.bj.bcebos.com/demo/wmt_shrinked_data/wmt14.tgz + http://paddlemodels.bj.bcebos.com/wmt/wmt14.tgz . Args: data_file(str): path to data tar file, can be set None if @@ -70,8 +70,6 @@ class WMT14(Dataset): def forward(self, src_ids, trg_ids, trg_ids_next): return paddle.sum(src_ids), paddle.sum(trg_ids), paddle.sum(trg_ids_next) - paddle.disable_static() - wmt14 = WMT14(mode='train', dict_size=50) for i in range(10):