Merge branch 'develop' of https://github.com/PaddlePaddle/models into my-dev

fluid/DeepASR/data_utils/async_data_reader.py

@@ -30,11 +30,12 @@ class SampleInfo(object):
         label_bin_path (str): File containing the label data.
         label_size (int): Byte count of the sample's label data.
         label_frame_num (int): Label number of the sample.
+        sample_name (str): Key of the sample
     """
 
     def __init__(self, feature_bin_path, feature_start, feature_size,
                  feature_frame_num, feature_dim, label_bin_path, label_start,
-                 label_size, label_frame_num):
+                 label_size, label_frame_num, sample_name):
         self.feature_bin_path = feature_bin_path
         self.feature_start = feature_start
         self.feature_size = feature_size
@@ -45,6 +46,7 @@ class SampleInfo(object):
         self.label_start = label_start
         self.label_size = label_size
         self.label_frame_num = label_frame_num
+        self.sample_name = sample_name
 
 
 class SampleInfoBucket(object):
@@ -102,19 +104,28 @@ class SampleInfoBucket(object):
             feature_bin_path = self._feature_bin_paths[block_idx]
             feature_desc_path = self._feature_desc_paths[block_idx]
 
-            label_desc_lines = open(label_desc_path).readlines()
             feature_desc_lines = open(feature_desc_path).readlines()
 
-            sample_num = int(label_desc_lines[0].split()[1])
-            assert sample_num == int(feature_desc_lines[0].split()[1])
+            label_desc_lines = []
+            if label_desc_path != "":
+                label_desc_lines = open(label_desc_path).readlines()
+            sample_num = int(feature_desc_lines[0].split()[1])
+
+            if label_desc_path != "":
+                assert sample_num == int(label_desc_lines[0].split()[1])
 
             for i in xrange(sample_num):
                 feature_desc_split = feature_desc_lines[i + 1].split()
+                sample_name = feature_desc_split[0]
                 feature_start = int(feature_desc_split[2])
                 feature_size = int(feature_desc_split[3])
                 feature_frame_num = int(feature_desc_split[4])
                 feature_dim = int(feature_desc_split[5])
 
+                label_start = -1
+                label_size = -1
+                label_frame_num = feature_frame_num
+                if label_desc_path != "":
                     label_desc_split = label_desc_lines[i + 1].split()
                     label_start = int(label_desc_split[2])
                     label_size = int(label_desc_split[3])
@@ -129,7 +140,7 @@ class SampleInfoBucket(object):
                         SampleInfo(feature_bin_path, feature_start,
                                    feature_size, feature_frame_num, feature_dim,
                                    label_bin_path, label_start, label_size,
-                                   label_frame_num))
+                                   label_frame_num, sample_name))
                 #split sentence
                 else:
                     cur_frame_pos = 0
@@ -150,13 +161,12 @@ class SampleInfoBucket(object):
                                 * feature_dim * 4, cur_frame_len * feature_dim *
                                 4, cur_frame_len, feature_dim, label_bin_path,
                                 label_start + cur_frame_pos * 4, cur_frame_len *
-                                4, cur_frame_len))
+                                4, cur_frame_len, sample_name))
 
                         remain_frame_num -= cur_frame_len
                         cur_frame_pos += cur_frame_len
                         if remain_frame_num <= 0:
                             break
-
         return sample_info_list
 
 
@@ -192,7 +202,7 @@ class AsyncDataReader(object):
 
     def __init__(self,
                  feature_file_list,
-                 label_file_list,
+                 label_file_list="",
                  drop_frame_len=512,
                  proc_num=10,
                  sample_buffer_size=1024,
@@ -221,9 +231,11 @@ class AsyncDataReader(object):
     def generate_bucket_list(self, is_shuffle):
         if self._block_info_list is None:
             block_feature_info_lines = open(self._feature_file_list).readlines()
-            block_label_info_lines = open(self._label_file_list).readlines()
-            assert len(block_feature_info_lines) == len(block_label_info_lines)
             self._block_info_list = []
+            if self._label_file_list != "":
+                block_label_info_lines = open(self._label_file_list).readlines()
+                assert len(block_feature_info_lines) == len(
+                    block_label_info_lines)
                 for i in xrange(0, len(block_feature_info_lines), 2):
                     block_info = (block_feature_info_lines[i],
                                   block_feature_info_lines[i + 1],
@@ -231,6 +243,12 @@ class AsyncDataReader(object):
                                   block_label_info_lines[i + 1])
                     self._block_info_list.append(
                         map(lambda line: line.strip(), block_info))
+            else:
+                for i in xrange(0, len(block_feature_info_lines), 2):
+                    block_info = (block_feature_info_lines[i],
+                                  block_feature_info_lines[i + 1], "", "")
+                    self._block_info_list.append(
+                        map(lambda line: line.strip(), block_info))
 
         if is_shuffle:
             self._rng.shuffle(self._block_info_list)
@@ -310,19 +328,25 @@ class AsyncDataReader(object):
                          sample_info.feature_dim,
                          len(feature_bytes))
 
+                label_data = None
+                if sample_info.label_bin_path != "":
                     label_bytes = read_bytes(sample_info.label_bin_path,
                                              sample_info.label_start,
                                              sample_info.label_size)
 
-                assert sample_info.label_frame_num * 4 == len(label_bytes), (
-                    sample_info.label_bin_path, sample_info.label_array,
+                    assert sample_info.label_frame_num * 4 == len(
+                        label_bytes), (sample_info.label_bin_path,
+                                       sample_info.label_array,
                                        len(label_bytes))
 
-                label_array = struct.unpack('I' * sample_info.label_frame_num,
-                                            label_bytes)
+                    label_array = struct.unpack(
+                        'I' * sample_info.label_frame_num, label_bytes)
                     label_data = np.array(
                         label_array, dtype='int64').reshape(
                             (sample_info.label_frame_num, 1))
+                else:
+                    label_data = np.zeros(
+                        (sample_info.label_frame_num, 1), dtype='int64')
 
                 feature_frame_num = sample_info.feature_frame_num
                 feature_dim = sample_info.feature_dim
@@ -332,12 +356,11 @@ class AsyncDataReader(object):
                 feature_data = np.array(
                     feature_array, dtype='float32').reshape((
                         sample_info.feature_frame_num, sample_info.feature_dim))
-
-                sample_data = (feature_data, label_data)
+                sample_data = (feature_data, label_data,
+                               sample_info.sample_name)
                 for transformer in self._transformers:
                     # @TODO(pkuyym) to make transfomer only accept feature_data
                     sample_data = transformer.perform_trans(sample_data)
-
                 while order_id != out_order[0]:
                     time.sleep(0.001)
 
@@ -387,12 +410,14 @@ class AsyncDataReader(object):
             batch_feature = np.zeros((lod[-1], frame_dim), dtype="float32")
             batch_label = np.zeros((lod[-1], 1), dtype="int64")
             start = 0
+            name_lst = []
             for sample in batch_samples:
                 frame_num = sample[0].shape[0]
                 batch_feature[start:start + frame_num, :] = sample[0]
                 batch_label[start:start + frame_num, :] = sample[1]
                 start += frame_num
-            return (batch_feature, batch_label)
+                name_lst.append(sample[2])
+            return (batch_feature, batch_label, name_lst)
 
         @suppress_complaints(verbose=self._verbose, notify=self._force_exit)
         def batch_assembling_task(sample_generator, batch_queue):
@@ -402,16 +427,16 @@ class AsyncDataReader(object):
                 batch_samples.append(sample)
                 lod.append(lod[-1] + sample[0].shape[0])
                 if len(batch_samples) == batch_size:
-                    (batch_feature, batch_label) = batch_to_ndarray(
+                    (batch_feature, batch_label, name_lst) = batch_to_ndarray(
                         batch_samples, lod)
-                    batch_queue.put((batch_feature, batch_label, lod))
+                    batch_queue.put((batch_feature, batch_label, lod, name_lst))
                     batch_samples = []
                     lod = [0]
 
             if len(batch_samples) >= minimum_batch_size:
-                (batch_feature, batch_label) = batch_to_ndarray(batch_samples,
-                                                                lod)
-                batch_queue.put((batch_feature, batch_label, lod))
+                (batch_feature, batch_label, name_lst) = batch_to_ndarray(
+                    batch_samples, lod)
+                batch_queue.put((batch_feature, batch_label, lod, name_lst))
 
             batch_queue.put(EpochEndSignal())
 

fluid/DeepASR/data_utils/augmentor/tests/test_data_trans.py

@@ -22,7 +22,7 @@ class TestTransMeanVarianceNorm(unittest.TestCase):
         feature = np.zeros((2, 120), dtype="float32")
         feature.fill(1)
         trans = trans_mean_variance_norm.TransMeanVarianceNorm(self._file_path)
-        (feature1, label1) = trans.perform_trans((feature, None))
+        (feature1, label1, name) = trans.perform_trans((feature, None, None))
         (mean, var) = trans.get_mean_var()
         feature_flat1 = feature1.flatten()
         feature_flat = feature.flatten()
@@ -70,7 +70,7 @@ class TestTransAddDelta(unittest.TestCase):
         feature[2, 0:40].fill(3)
         feature[3, 0:40].fill(4)
         trans = trans_add_delta.TransAddDelta()
-        (feature, label) = trans.perform_trans((feature, None))
+        (feature, label, name) = trans.perform_trans((feature, None, None))
         self.assertAlmostEqual(feature.shape[0], 4)
         self.assertAlmostEqual(feature.shape[1], 120)
         self.assertAlmostEqual(1.0, feature[0][0])
@@ -93,7 +93,7 @@ class TestTransSplict(unittest.TestCase):
             feature[i, :].fill(i)
 
         trans = trans_splice.TransSplice()
-        (feature, label) = trans.perform_trans((feature, None))
+        (feature, label, name) = trans.perform_trans((feature, None, None))
         self.assertEqual(feature.shape[1], 110)
 
         for i in xrange(8):

fluid/DeepASR/data_utils/augmentor/trans_add_delta.py

@@ -32,9 +32,9 @@ class TransAddDelta(object):
             Args: 
                 sample(object,tuple): contain feature numpy and label numpy
             Returns:
-                (feature, label)
+                (feature, label, name)
         """
-        (feature, label) = sample
+        (feature, label, name) = sample
         frame_dim = feature.shape[1]
         d_frame_dim = frame_dim * 3
         head_filled = 5
@@ -64,7 +64,7 @@ class TransAddDelta(object):
                       start * d_frame_dim + 2 * frame_dim, frame_dim, nframe,
                       d_frame_dim)
         mat.shape = tmp_shape
-        return (mat[head_filled:mat.shape[0] - tail_filled, :], label)
+        return (mat[head_filled:mat.shape[0] - tail_filled, :], label, name)
 
     def _regress(self, data_in, start_in, data_out, start_out, size, n, step):
         """ regress

fluid/DeepASR/data_utils/augmentor/trans_mean_variance_norm.py

@@ -53,9 +53,9 @@ class TransMeanVarianceNorm(object):
             Args:
                 sample(object):input sample, contain feature numpy and label numpy
             Returns:
-                (feature, label)
+                (feature, label, name)
         """
-        (feature, label) = sample
+        (feature, label, name) = sample
         shape = feature.shape
         assert len(shape) == 2
         nfeature_len = shape[0] * shape[1]
@@ -68,4 +68,4 @@ class TransMeanVarianceNorm(object):
             feature[ncur_idx:ncur_idx + self._nLen] = block
             ncur_idx += self._nLen
         feature = feature.reshape(shape)
-        return (feature, label)
+        return (feature, label, name)

fluid/DeepASR/data_utils/augmentor/trans_splice.py

@@ -30,9 +30,9 @@ class TransSplice(object):
         Args:
             sample(object): input sample(feature, label)
         Return:
-            (feature, label)
+            (feature, label, name)
         """
-        (feature, label) = sample
+        (feature, label, name) = sample
         nframe_num = feature.shape[0]
         nframe_dim = feature.shape[1]
         nnew_frame_dim = nframe_dim * (
@@ -61,4 +61,4 @@ class TransSplice(object):
             np.copyto(ret[i * nnew_frame_dim:(i + 1) * nnew_frame_dim],
                       mat[i * nframe_dim:i * nframe_dim + nnew_frame_dim])
         ret = ret.reshape((nframe_num, nnew_frame_dim))
-        return (ret, label)
+        return (ret, label, name)

fluid/DeepASR/train.py

@@ -210,6 +210,7 @@ def train(args):
     # train data reader
     train_data_reader = reader.AsyncDataReader(args.train_feature_lst,
                                                args.train_label_lst, -1)
+
     train_data_reader.set_transformers(ltrans)
     # train
     for pass_id in xrange(args.pass_num):
@@ -218,7 +219,7 @@ def train(args):
                 train_data_reader.batch_iterator(args.batch_size,
                                                  args.minimum_batch_size)):
             # load_data
-            (features, labels, lod) = batch_data
+            (features, labels, lod, name_lst) = batch_data
             feature_t.set(features, place)
             feature_t.set_lod([lod])
             label_t.set(labels, place)

fluid/neural_machine_translation/transformer/config.py

@@ -92,7 +92,9 @@ pos_enc_param_names = (
 encoder_input_data_names = (
     "src_word",
     "src_pos",
-    "src_slf_attn_bias", )
+    "src_slf_attn_bias",
+    "src_slf_attn_pre_softmax_shape",
+    "src_slf_attn_post_softmax_shape", )
 
 # Names of all data layers in decoder listed in order.
 decoder_input_data_names = (
@@ -100,6 +102,10 @@ decoder_input_data_names = (
     "trg_pos",
     "trg_slf_attn_bias",
     "trg_src_attn_bias",
+    "trg_slf_attn_pre_softmax_shape",
+    "trg_slf_attn_post_softmax_shape",
+    "trg_src_attn_pre_softmax_shape",
+    "trg_src_attn_post_softmax_shape",
     "enc_output", )
 
 # Names of label related data layers listed in order.

fluid/neural_machine_translation/transformer/infer.py

@@ -27,7 +27,14 @@ def translate_batch(exe, src_words, encoder, enc_in_names, enc_out_names,
         is_target=False,
         return_pos=True,
         return_attn_bias=True,
-        return_max_len=True)
+        return_max_len=False)
+    # Append the shape inputs to reshape before and after softmax in encoder
+    # self attention.
+    enc_in_data = enc_in_data + [
+        np.array(
+            [-1, enc_in_data[2].shape[-1]], dtype="int32"), np.array(
+                enc_in_data[2].shape, dtype="int32")
+    ]
     enc_output = exe.run(encoder,
                          feed=dict(zip(enc_in_names, enc_in_data)),
                          fetch_list=enc_out_names)[0]
@@ -35,8 +42,8 @@ def translate_batch(exe, src_words, encoder, enc_in_names, enc_out_names,
     # Beam Search.
     # To store the beam info.
     scores = np.zeros((batch_size, beam_size), dtype="float32")
-    prev_branchs = [[]] * batch_size
-    next_ids = [[]] * batch_size
+    prev_branchs = [[] for i in range(batch_size)]
+    next_ids = [[] for i in range(batch_size)]
     # Use beam_map to map the instance idx in batch to beam idx, since the
     # size of feeded batch is changing.
     beam_map = range(batch_size)
@@ -64,8 +71,8 @@ def translate_batch(exe, src_words, encoder, enc_in_names, enc_out_names,
         trg_words = np.array(
             [[bos_idx]] * batch_size * beam_size, dtype="int64")
         trg_pos = np.array([[1]] * batch_size * beam_size, dtype="int64")
-        src_max_length, src_slf_attn_bias, trg_max_len = enc_in_data[
-            -1], enc_in_data[-2], 1
+        src_max_length, src_slf_attn_bias, trg_max_len = enc_in_data[2].shape[
+            -1], enc_in_data[2], 1
         # This is used to remove attention on subsequent words.
         trg_slf_attn_bias = np.ones((batch_size * beam_size, trg_max_len,
                                      trg_max_len))
@@ -77,15 +84,33 @@ def translate_batch(exe, src_words, encoder, enc_in_names, enc_out_names,
         trg_src_attn_bias = np.tile(
             src_slf_attn_bias[:, :, ::src_max_length, :],
             [beam_size, 1, trg_max_len, 1])
+        # Append the shape inputs to reshape before and after softmax in
+        # decoder self attention.
+        trg_slf_attn_pre_softmax_shape = np.array(
+            [-1, trg_slf_attn_bias.shape[-1]], dtype="int32")
+        trg_slf_attn_post_softmax_shape = np.array(
+            trg_slf_attn_bias.shape, dtype="int32")
+        # Append the shape inputs to reshape before and after softmax in
+        # encoder-decoder attention.
+        trg_src_attn_pre_softmax_shape = np.array(
+            [-1, trg_src_attn_bias.shape[-1]], dtype="int32")
+        trg_src_attn_post_softmax_shape = np.array(
+            trg_src_attn_bias.shape, dtype="int32")
         enc_output = np.tile(enc_output, [beam_size, 1, 1])
-        return trg_words, trg_pos, trg_slf_attn_bias, trg_src_attn_bias, enc_output
+        return trg_words, trg_pos, trg_slf_attn_bias, trg_src_attn_bias, \
+            trg_slf_attn_pre_softmax_shape, trg_slf_attn_post_softmax_shape, \
+            trg_src_attn_pre_softmax_shape, trg_src_attn_post_softmax_shape, \
+            enc_output
 
     def update_dec_in_data(dec_in_data, next_ids, active_beams):
         """
         Update the input data of decoder mainly by slicing from the previous
         input data and dropping the finished instance beams.
         """
-        trg_words, trg_pos, trg_slf_attn_bias, trg_src_attn_bias, enc_output = dec_in_data
+        trg_words, trg_pos, trg_slf_attn_bias, trg_src_attn_bias, \
+            trg_slf_attn_pre_softmax_shape, trg_slf_attn_post_softmax_shape, \
+            trg_src_attn_pre_softmax_shape, trg_src_attn_post_softmax_shape, \
+            enc_output = dec_in_data
         trg_cur_len = len(next_ids[0]) + 1  # include the <bos>
         trg_words = np.array(
             [
@@ -112,8 +137,23 @@ def translate_batch(exe, src_words, encoder, enc_in_names, enc_out_names,
         trg_src_attn_bias = np.tile(trg_src_attn_bias[
             active_beams_indice, :, ::trg_src_attn_bias.shape[2], :],
                                     [1, 1, trg_cur_len, 1])
+        # Append the shape inputs to reshape before and after softmax in
+        # decoder self attention.
+        trg_slf_attn_pre_softmax_shape = np.array(
+            [-1, trg_slf_attn_bias.shape[-1]], dtype="int32")
+        trg_slf_attn_post_softmax_shape = np.array(
+            trg_slf_attn_bias.shape, dtype="int32")
+        # Append the shape inputs to reshape before and after softmax in
+        # encoder-decoder attention.
+        trg_src_attn_pre_softmax_shape = np.array(
+            [-1, trg_src_attn_bias.shape[-1]], dtype="int32")
+        trg_src_attn_post_softmax_shape = np.array(
+            trg_src_attn_bias.shape, dtype="int32")
         enc_output = enc_output[active_beams_indice, :, :]
-        return trg_words, trg_pos, trg_slf_attn_bias, trg_src_attn_bias, enc_output
+        return trg_words, trg_pos, trg_slf_attn_bias, trg_src_attn_bias, \
+            trg_slf_attn_pre_softmax_shape, trg_slf_attn_post_softmax_shape, \
+            trg_src_attn_pre_softmax_shape, trg_src_attn_post_softmax_shape, \
+            enc_output
 
     dec_in_data = init_dec_in_data(batch_size, beam_size, enc_in_data,
                                    enc_output)

fluid/neural_machine_translation/transformer/model.py

@@ -32,7 +32,9 @@ def multi_head_attention(queries,
                          d_value,
                          d_model,
                          n_head=1,
-                         dropout_rate=0.):
+                         dropout_rate=0.,
+                         pre_softmax_shape=None,
+                         post_softmax_shape=None):
     """
     Multi-Head Attention. Note that attn_bias is added to the logit before
     computing softmax activiation to mask certain selected positions so that
@@ -111,26 +113,16 @@ def multi_head_attention(queries,
         """
         Scaled Dot-Product Attention
         """
-
-        # FIXME(guosheng): Optimize the shape in reshape_op or softmax_op.
-
-        # The current implementation of softmax_op only supports 2D tensor,
-        # consequently it cannot be directly used here.
-        # If to use the reshape_op, Besides, the shape of product inferred in
-        # compile-time is not the actual shape in run-time. It cann't be used
-        # to set the attribute of reshape_op.
-        # So, here define the softmax for temporary solution.
-
-        def __softmax(x, eps=1e-9):
-            exp_out = layers.exp(x=x)
-            sum_out = layers.reduce_sum(exp_out, dim=-1, keep_dim=False)
-            return layers.elementwise_div(x=exp_out, y=sum_out, axis=0)
-
         scaled_q = layers.scale(x=q, scale=d_model**-0.5)
         product = layers.matmul(x=scaled_q, y=k, transpose_y=True)
-        weights = __softmax(
-            layers.elementwise_add(
-                x=product, y=attn_bias) if attn_bias else product)
+        weights = layers.reshape(
+            x=layers.elementwise_add(
+                x=product, y=attn_bias) if attn_bias else product,
+            shape=[-1, product.shape[-1]],
+            actual_shape=pre_softmax_shape,
+            act="softmax")
+        weights = layers.reshape(
+            x=weights, shape=product.shape, actual_shape=post_softmax_shape)
         if dropout_rate:
             weights = layers.dropout(
                 weights, dropout_prob=dropout_rate, is_test=False)
@@ -177,7 +169,7 @@ def positionwise_feed_forward(x, d_inner_hid, d_hid):
     return out
 
 
-def pre_post_process_layer(prev_out, out, process_cmd, dropout=0.):
+def pre_post_process_layer(prev_out, out, process_cmd, dropout_rate=0.):
     """
     Add residual connection, layer normalization and droput to the out tensor
     optionally according to the value of process_cmd.
@@ -195,8 +187,9 @@ def pre_post_process_layer(prev_out, out, process_cmd, dropout=0.):
                 param_attr=fluid.initializer.Constant(1.),
                 bias_attr=fluid.initializer.Constant(0.))
         elif cmd == "d":  # add dropout
-            if dropout:
-                out = layers.dropout(out, dropout_prob=dropout, is_test=False)
+            if dropout_rate:
+                out = layers.dropout(
+                    out, dropout_prob=dropout_rate, is_test=False)
     return out
 
 
@@ -210,7 +203,7 @@ def prepare_encoder(src_word,
                     src_emb_dim,
                     src_pad_idx,
                     src_max_len,
-                    dropout=0.,
+                    dropout_rate=0.,
                     pos_pad_idx=0,
                     pos_enc_param_name=None):
     """Add word embeddings and position encodings.
@@ -235,8 +228,8 @@ def prepare_encoder(src_word,
     # FIXME(guosheng): Decouple the program desc with batch_size.
     enc_input = layers.reshape(x=enc_input, shape=[batch_size, -1, src_emb_dim])
     return layers.dropout(
-        enc_input, dropout_prob=dropout,
-        is_test=False) if dropout else enc_input
+        enc_input, dropout_prob=dropout_rate,
+        is_test=False) if dropout_rate else enc_input
 
 
 prepare_encoder = partial(
@@ -252,7 +245,9 @@ def encoder_layer(enc_input,
                   d_value,
                   d_model,
                   d_inner_hid,
-                  dropout_rate=0.):
+                  dropout_rate=0.,
+                  pre_softmax_shape=None,
+                  post_softmax_shape=None):
     """The encoder layers that can be stacked to form a deep encoder.
 
     This module consits of a multi-head (self) attention followed by
@@ -260,9 +255,9 @@ def encoder_layer(enc_input,
     with the post_process_layer to add residual connection, layer normalization
     and droput.
     """
-    attn_output = multi_head_attention(enc_input, enc_input, enc_input,
-                                       attn_bias, d_key, d_value, d_model,
-                                       n_head, dropout_rate)
+    attn_output = multi_head_attention(
+        enc_input, enc_input, enc_input, attn_bias, d_key, d_value, d_model,
+        n_head, dropout_rate, pre_softmax_shape, post_softmax_shape)
     attn_output = post_process_layer(enc_input, attn_output, "dan",
                                      dropout_rate)
     ffd_output = positionwise_feed_forward(attn_output, d_inner_hid, d_model)
@@ -277,7 +272,9 @@ def encoder(enc_input,
             d_value,
             d_model,
             d_inner_hid,
-            dropout_rate=0.):
+            dropout_rate=0.,
+            pre_softmax_shape=None,
+            post_softmax_shape=None):
     """
     The encoder is composed of a stack of identical layers returned by calling
     encoder_layer.
@@ -291,7 +288,9 @@ def encoder(enc_input,
             d_value,
             d_model,
             d_inner_hid,
-            dropout_rate, )
+            dropout_rate,
+            pre_softmax_shape,
+            post_softmax_shape, )
         enc_input = enc_output
     return enc_output
 
@@ -305,7 +304,11 @@ def decoder_layer(dec_input,
                   d_value,
                   d_model,
                   d_inner_hid,
-                  dropout_rate=0.):
+                  dropout_rate=0.,
+                  slf_attn_pre_softmax_shape=None,
+                  slf_attn_post_softmax_shape=None,
+                  src_attn_pre_softmax_shape=None,
+                  src_attn_post_softmax_shape=None):
     """ The layer to be stacked in decoder part.
 
     The structure of this module is similar to that in the encoder part except
@@ -320,7 +323,9 @@ def decoder_layer(dec_input,
         d_value,
         d_model,
         n_head,
-        dropout_rate, )
+        dropout_rate,
+        slf_attn_pre_softmax_shape,
+        slf_attn_post_softmax_shape, )
     slf_attn_output = post_process_layer(
         dec_input,
         slf_attn_output,
@@ -335,7 +340,9 @@ def decoder_layer(dec_input,
         d_value,
         d_model,
         n_head,
-        dropout_rate, )
+        dropout_rate,
+        src_attn_pre_softmax_shape,
+        src_attn_post_softmax_shape, )
     enc_attn_output = post_process_layer(
         slf_attn_output,
         enc_attn_output,
@@ -363,7 +370,11 @@ def decoder(dec_input,
             d_value,
             d_model,
             d_inner_hid,
-            dropout_rate=0.):
+            dropout_rate=0.,
+            slf_attn_pre_softmax_shape=None,
+            slf_attn_post_softmax_shape=None,
+            src_attn_pre_softmax_shape=None,
+            src_attn_post_softmax_shape=None):
     """
     The decoder is composed of a stack of identical decoder_layer layers.
     """
@@ -378,7 +389,11 @@ def decoder(dec_input,
             d_value,
             d_model,
             d_inner_hid,
-            dropout_rate, )
+            dropout_rate,
+            slf_attn_pre_softmax_shape,
+            slf_attn_post_softmax_shape,
+            src_attn_pre_softmax_shape,
+            src_attn_post_softmax_shape, )
         dec_input = dec_output
     return dec_output
 
@@ -391,7 +406,9 @@ def make_inputs(input_data_names,
                 is_pos,
                 slf_attn_bias_flag,
                 src_attn_bias_flag,
-                enc_output_flag=False):
+                enc_output_flag=False,
+                slf_attn_shape_flag=True,
+                src_attn_shape_flag=True):
     """
     Define the input data layers for the transformer model.
     """
@@ -429,6 +446,32 @@ def make_inputs(input_data_names,
             dtype="float32",
             append_batch_size=False)
         input_layers += [src_attn_bias]
+    if slf_attn_shape_flag:
+        slf_attn_pre_softmax_shape = layers.data(
+            name=input_data_names[len(input_layers)],
+            shape=[3],
+            dtype="int32",
+            append_batch_size=False)
+        input_layers += [slf_attn_pre_softmax_shape]
+        slf_attn_post_softmax_shape = layers.data(
+            name=input_data_names[len(input_layers)],
+            shape=[3],
+            dtype="int32",
+            append_batch_size=False)
+        input_layers += [slf_attn_post_softmax_shape]
+    if src_attn_shape_flag:
+        src_attn_pre_softmax_shape = layers.data(
+            name=input_data_names[len(input_layers)],
+            shape=[3],
+            dtype="int32",
+            append_batch_size=False)
+        input_layers += [src_attn_pre_softmax_shape]
+        src_attn_post_softmax_shape = layers.data(
+            name=input_data_names[len(input_layers)],
+            shape=[3],
+            dtype="int32",
+            append_batch_size=False)
+        input_layers += [src_attn_post_softmax_shape]
     if enc_output_flag:
         enc_output = layers.data(
             name=input_data_names[len(input_layers)],
@@ -436,6 +479,7 @@ def make_inputs(input_data_names,
             dtype="float32",
             append_batch_size=False)
         input_layers += [enc_output]
+
     return input_layers
 
 
@@ -453,8 +497,18 @@ def transformer(
         src_pad_idx,
         trg_pad_idx,
         pos_pad_idx, ):
-    enc_input_layers = make_inputs(encoder_input_data_names, n_head, d_model,
-                                   batch_size, max_length, True, True, False)
+    enc_input_layers = make_inputs(
+        encoder_input_data_names,
+        n_head,
+        d_model,
+        batch_size,
+        max_length,
+        is_pos=True,
+        slf_attn_bias_flag=True,
+        src_attn_bias_flag=False,
+        enc_output_flag=False,
+        slf_attn_shape_flag=True,
+        src_attn_shape_flag=False)
 
     enc_output = wrap_encoder(
         src_vocab_size,
@@ -470,8 +524,18 @@ def transformer(
         pos_pad_idx,
         enc_input_layers, )
 
-    dec_input_layers = make_inputs(decoder_input_data_names, n_head, d_model,
-                                   batch_size, max_length, True, True, True)
+    dec_input_layers = make_inputs(
+        decoder_input_data_names,
+        n_head,
+        d_model,
+        batch_size,
+        max_length,
+        is_pos=True,
+        slf_attn_bias_flag=True,
+        src_attn_bias_flag=True,
+        enc_output_flag=False,
+        slf_attn_shape_flag=True,
+        src_attn_shape_flag=True)
 
     predict = wrap_decoder(
         trg_vocab_size,
@@ -490,9 +554,19 @@ def transformer(
 
     # Padding index do not contribute to the total loss. The weights is used to
     # cancel padding index in calculating the loss.
-    gold, weights = make_inputs(label_data_names, n_head, d_model, batch_size,
-                                max_length, False, False, False)
-    cost = layers.cross_entropy(input=predict, label=gold)
+    gold, weights = make_inputs(
+        label_data_names,
+        n_head,
+        d_model,
+        batch_size,
+        max_length,
+        is_pos=False,
+        slf_attn_bias_flag=False,
+        src_attn_bias_flag=False,
+        enc_output_flag=False,
+        slf_attn_shape_flag=False,
+        src_attn_shape_flag=False)
+    cost = layers.softmax_with_cross_entropy(logits=predict, label=gold)
     weighted_cost = cost * weights
     return layers.reduce_sum(weighted_cost), predict
 
@@ -514,11 +588,22 @@ def wrap_encoder(src_vocab_size,
     """
     if enc_input_layers is None:
         # This is used to implement independent encoder program in inference.
-        src_word, src_pos, src_slf_attn_bias = make_inputs(
-            encoder_input_data_names, n_head, d_model, batch_size, max_length,
-            True, True, False)
+        src_word, src_pos, src_slf_attn_bias, slf_attn_pre_softmax_shape, \
+            slf_attn_post_softmax_shape = make_inputs(
+                encoder_input_data_names,
+                n_head,
+                d_model,
+                batch_size,
+                max_length,
+                is_pos=True,
+                slf_attn_bias_flag=True,
+                src_attn_bias_flag=False,
+                enc_output_flag=False,
+                slf_attn_shape_flag=True,
+                src_attn_shape_flag=False)
     else:
-        src_word, src_pos, src_slf_attn_bias = enc_input_layers
+        src_word, src_pos, src_slf_attn_bias, slf_attn_pre_softmax_shape, \
+            slf_attn_post_softmax_shape = enc_input_layers
     enc_input = prepare_encoder(
         src_word,
         src_pos,
@@ -536,7 +621,9 @@ def wrap_encoder(src_vocab_size,
         d_value,
         d_model,
         d_inner_hid,
-        dropout_rate, )
+        dropout_rate,
+        slf_attn_pre_softmax_shape,
+        slf_attn_post_softmax_shape, )
     return enc_output
 
 
@@ -558,11 +645,26 @@ def wrap_decoder(trg_vocab_size,
     """
     if dec_input_layers is None:
         # This is used to implement independent decoder program in inference.
-        trg_word, trg_pos, trg_slf_attn_bias, trg_src_attn_bias, enc_output = make_inputs(
-            decoder_input_data_names, n_head, d_model, batch_size, max_length,
-            True, True, True, True)
+        trg_word, trg_pos, trg_slf_attn_bias, trg_src_attn_bias, \
+            slf_attn_pre_softmax_shape, slf_attn_post_softmax_shape, \
+            src_attn_pre_softmax_shape, src_attn_post_softmax_shape, \
+            enc_output = make_inputs(
+                decoder_input_data_names,
+                n_head,
+                d_model,
+                batch_size,
+                max_length,
+                is_pos=True,
+                slf_attn_bias_flag=True,
+                src_attn_bias_flag=True,
+                enc_output_flag=True,
+                slf_attn_shape_flag=True,
+                src_attn_shape_flag=True)
     else:
-        trg_word, trg_pos, trg_slf_attn_bias, trg_src_attn_bias = dec_input_layers
+        trg_word, trg_pos, trg_slf_attn_bias, trg_src_attn_bias, \
+            slf_attn_pre_softmax_shape, slf_attn_post_softmax_shape, \
+            src_attn_pre_softmax_shape, src_attn_post_softmax_shape = \
+                dec_input_layers
 
     dec_input = prepare_decoder(
         trg_word,
@@ -583,13 +685,17 @@ def wrap_decoder(trg_vocab_size,
         d_value,
         d_model,
         d_inner_hid,
-        dropout_rate, )
-
+        dropout_rate,
+        slf_attn_pre_softmax_shape,
+        slf_attn_post_softmax_shape,
+        src_attn_pre_softmax_shape,
+        src_attn_post_softmax_shape, )
+    # Return logits for training and probs for inference.
     predict = layers.reshape(
         x=layers.fc(input=dec_output,
                     size=trg_vocab_size,
                     bias_attr=False,
                     num_flatten_dims=2),
         shape=[-1, trg_vocab_size],
-        act="softmax")
+        act="softmax" if dec_input_layers is None else None)
     return predict

fluid/neural_machine_translation/transformer/train.py

@@ -66,13 +66,29 @@ def prepare_batch_input(insts, input_data_names, src_pad_idx, trg_pad_idx,
         [inst[1] for inst in insts], trg_pad_idx, n_head, is_target=True)
     trg_src_attn_bias = np.tile(src_slf_attn_bias[:, :, ::src_max_len, :],
                                 [1, 1, trg_max_len, 1]).astype("float32")
+    src_slf_attn_pre_softmax_shape = np.array(
+        [-1, src_slf_attn_bias.shape[-1]], dtype="int32")
+    src_slf_attn_post_softmax_shape = np.array(
+        src_slf_attn_bias.shape, dtype="int32")
+    trg_slf_attn_pre_softmax_shape = np.array(
+        [-1, trg_slf_attn_bias.shape[-1]], dtype="int32")
+    trg_slf_attn_post_softmax_shape = np.array(
+        trg_slf_attn_bias.shape, dtype="int32")
+    trg_src_attn_pre_softmax_shape = np.array(
+        [-1, trg_src_attn_bias.shape[-1]], dtype="int32")
+    trg_src_attn_post_softmax_shape = np.array(
+        trg_src_attn_bias.shape, dtype="int32")
     lbl_word = pad_batch_data([inst[2] for inst in insts], trg_pad_idx, n_head,
                               False, False, False, False)
     lbl_weight = (lbl_word != trg_pad_idx).astype("float32").reshape([-1, 1])
     input_dict = dict(
         zip(input_data_names, [
-            src_word, src_pos, src_slf_attn_bias, trg_word, trg_pos,
-            trg_slf_attn_bias, trg_src_attn_bias, lbl_word, lbl_weight
+            src_word, src_pos, src_slf_attn_bias,
+            src_slf_attn_pre_softmax_shape, src_slf_attn_post_softmax_shape,
+            trg_word, trg_pos, trg_slf_attn_bias, trg_src_attn_bias,
+            trg_slf_attn_pre_softmax_shape, trg_slf_attn_post_softmax_shape,
+            trg_src_attn_pre_softmax_shape, trg_src_attn_post_softmax_shape,
+            lbl_word, lbl_weight
         ]))
     return input_dict