Update some en api docs (#20496)

* Fix api docs. test=document-fix

* Fix en docs. test=develop

* Fix the doc of dynamic_gru. test=document_fix

* Update API.spec. test=document_fix

* Fix codestyle test=develop, test=document_fix

paddle/fluid/API.spec

@@ -127,15 +127,15 @@ paddle.fluid.layers.center_loss (ArgSpec(args=['input', 'label', 'num_classes',
 paddle.fluid.layers.embedding (ArgSpec(args=['input', 'size', 'is_sparse', 'is_distributed', 'padding_idx', 'param_attr', 'dtype'], varargs=None, keywords=None, defaults=(False, False, None, None, 'float32')), ('document', 'c51fcac7a4f5786ca41f27fa60bd22c5'))
 paddle.fluid.layers.dynamic_lstm (ArgSpec(args=['input', 'size', 'h_0', 'c_0', 'param_attr', 'bias_attr', 'use_peepholes', 'is_reverse', 'gate_activation', 'cell_activation', 'candidate_activation', 'dtype', 'name'], varargs=None, keywords=None, defaults=(None, None, None, None, True, False, 'sigmoid', 'tanh', 'tanh', 'float32', None)), ('document', 'd4a82e2f5feb20c4a23ced8054e047ed'))
 paddle.fluid.layers.dynamic_lstmp (ArgSpec(args=['input', 'size', 'proj_size', 'param_attr', 'bias_attr', 'use_peepholes', 'is_reverse', 'gate_activation', 'cell_activation', 'candidate_activation', 'proj_activation', 'dtype', 'name', 'h_0', 'c_0', 'cell_clip', 'proj_clip'], varargs=None, keywords=None, defaults=(None, None, True, False, 'sigmoid', 'tanh', 'tanh', 'tanh', 'float32', None, None, None, None, None)), ('document', 'b35fe3e0c2ecca15a8be658277e064ec'))
-paddle.fluid.layers.dynamic_gru (ArgSpec(args=['input', 'size', 'param_attr', 'bias_attr', 'is_reverse', 'gate_activation', 'candidate_activation', 'h_0', 'origin_mode'], varargs=None, keywords=None, defaults=(None, None, False, 'sigmoid', 'tanh', None, False)), ('document', '83617c165827e030636c80486d5de6f3'))
-paddle.fluid.layers.gru_unit (ArgSpec(args=['input', 'hidden', 'size', 'param_attr', 'bias_attr', 'activation', 'gate_activation', 'origin_mode'], varargs=None, keywords=None, defaults=(None, None, 'tanh', 'sigmoid', False)), ('document', '33974b9bfa69f2f1eb85e6f956dff04e'))
+paddle.fluid.layers.dynamic_gru (ArgSpec(args=['input', 'size', 'param_attr', 'bias_attr', 'is_reverse', 'gate_activation', 'candidate_activation', 'h_0', 'origin_mode'], varargs=None, keywords=None, defaults=(None, None, False, 'sigmoid', 'tanh', None, False)), ('document', 'a3364b36fb3190b9bd75e419aa75573b'))
+paddle.fluid.layers.gru_unit (ArgSpec(args=['input', 'hidden', 'size', 'param_attr', 'bias_attr', 'activation', 'gate_activation', 'origin_mode'], varargs=None, keywords=None, defaults=(None, None, 'tanh', 'sigmoid', False)), ('document', '0b10a755b469d0b85b3a5cac38b4cf01'))
 paddle.fluid.layers.linear_chain_crf (ArgSpec(args=['input', 'label', 'param_attr', 'length'], varargs=None, keywords=None, defaults=(None, None)), ('document', 'b28bdb43160e9667be2a3457d19d9f5b'))
 paddle.fluid.layers.crf_decoding (ArgSpec(args=['input', 'param_attr', 'label', 'length'], varargs=None, keywords=None, defaults=(None, None)), ('document', '708ce0348b74d3e0c7885c2c524b7fa7'))
 paddle.fluid.layers.cos_sim (ArgSpec(args=['X', 'Y'], varargs=None, keywords=None, defaults=None), ('document', '48ec1ba2d75c4e2faf8d9a47350462ae'))
 paddle.fluid.layers.cross_entropy (ArgSpec(args=['input', 'label', 'soft_label', 'ignore_index'], varargs=None, keywords=None, defaults=(False, -100)), ('document', 'd1985a930a59c3bd41a7c1d72594f5b9'))
 paddle.fluid.layers.bpr_loss (ArgSpec(args=['input', 'label', 'name'], varargs=None, keywords=None, defaults=(None,)), ('document', 'ae57e6e5136dade436f0df1f11770afa'))
 paddle.fluid.layers.square_error_cost (ArgSpec(args=['input', 'label'], varargs=None, keywords=None, defaults=None), ('document', '4ed09e115b50ec7393674c4c09d223a2'))
-paddle.fluid.layers.chunk_eval (ArgSpec(args=['input', 'label', 'chunk_scheme', 'num_chunk_types', 'excluded_chunk_types', 'seq_length'], varargs=None, keywords=None, defaults=(None, None)), ('document', 'b02844e0ad4bd713c5fe6802aa13219c'))
+paddle.fluid.layers.chunk_eval (ArgSpec(args=['input', 'label', 'chunk_scheme', 'num_chunk_types', 'excluded_chunk_types', 'seq_length'], varargs=None, keywords=None, defaults=(None, None)), ('document', 'a8aa2071cae18df1e8dde9183d64bfb1'))
 paddle.fluid.layers.sequence_conv (ArgSpec(args=['input', 'num_filters', 'filter_size', 'filter_stride', 'padding', 'padding_start', 'bias_attr', 'param_attr', 'act', 'name'], varargs=None, keywords=None, defaults=(3, 1, True, None, None, None, None, None)), ('document', 'ebddcc5a1073ef065d22b4673e36b1d2'))
 paddle.fluid.layers.conv2d (ArgSpec(args=['input', 'num_filters', 'filter_size', 'stride', 'padding', 'dilation', 'groups', 'param_attr', 'bias_attr', 'use_cudnn', 'act', 'name', 'data_format'], varargs=None, keywords=None, defaults=(1, 0, 1, None, None, None, True, None, None, 'NCHW')), ('document', 'e91c63b8ac8c35982c0ac518537e44bf'))
 paddle.fluid.layers.conv3d (ArgSpec(args=['input', 'num_filters', 'filter_size', 'stride', 'padding', 'dilation', 'groups', 'param_attr', 'bias_attr', 'use_cudnn', 'act', 'name', 'data_format'], varargs=None, keywords=None, defaults=(1, 0, 1, None, None, None, True, None, None, 'NCDHW')), ('document', 'feff9c8ebb4d4d0be5345f9042f57c8e'))
@@ -149,14 +149,14 @@ paddle.fluid.layers.adaptive_pool3d (ArgSpec(args=['input', 'pool_size', 'pool_t
 paddle.fluid.layers.batch_norm (ArgSpec(args=['input', 'act', 'is_test', 'momentum', 'epsilon', 'param_attr', 'bias_attr', 'data_layout', 'in_place', 'name', 'moving_mean_name', 'moving_variance_name', 'do_model_average_for_mean_and_var', 'fuse_with_relu', 'use_global_stats'], varargs=None, keywords=None, defaults=(None, False, 0.9, 1e-05, None, None, 'NCHW', False, None, None, None, False, False, False)), ('document', '1400433bae7876d0407ae205be39b7a1'))
 paddle.fluid.layers.instance_norm (ArgSpec(args=['input', 'epsilon', 'param_attr', 'bias_attr', 'name'], varargs=None, keywords=None, defaults=(1e-05, None, None, None)), ('document', '23d6fba8ad8495f67a66d8878be5b0be'))
 paddle.fluid.layers.data_norm (ArgSpec(args=['input', 'act', 'epsilon', 'param_attr', 'data_layout', 'in_place', 'name', 'moving_mean_name', 'moving_variance_name', 'do_model_average_for_mean_and_var'], varargs=None, keywords=None, defaults=(None, 1e-05, None, 'NCHW', False, None, None, None, False)), ('document', '5ba4cdb4ea5c03382da545335ffc05b7'))
-paddle.fluid.layers.beam_search_decode (ArgSpec(args=['ids', 'scores', 'beam_size', 'end_id', 'name'], varargs=None, keywords=None, defaults=(None,)), ('document', '83e08f21af41ac8bac37aeab1f86fdd0'))
+paddle.fluid.layers.beam_search_decode (ArgSpec(args=['ids', 'scores', 'beam_size', 'end_id', 'name'], varargs=None, keywords=None, defaults=(None,)), ('document', 'eafa177a7fed6178a51c1affa7f46a40'))
 paddle.fluid.layers.conv2d_transpose (ArgSpec(args=['input', 'num_filters', 'output_size', 'filter_size', 'padding', 'stride', 'dilation', 'groups', 'param_attr', 'bias_attr', 'use_cudnn', 'act', 'name', 'data_format'], varargs=None, keywords=None, defaults=(None, None, 0, 1, 1, None, None, None, True, None, None, 'NCHW')), ('document', 'ed24c2d0f82cd9a3b40488157285a584'))
 paddle.fluid.layers.conv3d_transpose (ArgSpec(args=['input', 'num_filters', 'output_size', 'filter_size', 'padding', 'stride', 'dilation', 'groups', 'param_attr', 'bias_attr', 'use_cudnn', 'act', 'name', 'data_format'], varargs=None, keywords=None, defaults=(None, None, 0, 1, 1, None, None, None, True, None, None, 'NCDHW')), ('document', 'efb1e3bc87339cb26faa2edae210e8b0'))
 paddle.fluid.layers.sequence_expand (ArgSpec(args=['x', 'y', 'ref_level', 'name'], varargs=None, keywords=None, defaults=(-1, None)), ('document', '10e122eb755c2bd1f78ef2332b28f1a0'))
 paddle.fluid.layers.sequence_expand_as (ArgSpec(args=['x', 'y', 'name'], varargs=None, keywords=None, defaults=(None,)), ('document', '858c432e7cbd8bb952cc2eb555457d50'))
 paddle.fluid.layers.sequence_pad (ArgSpec(args=['x', 'pad_value', 'maxlen', 'name'], varargs=None, keywords=None, defaults=(None, None)), ('document', 'df08b9c499ab3a90f95d08ab5b6c6c62'))
 paddle.fluid.layers.sequence_unpad (ArgSpec(args=['x', 'length', 'name'], varargs=None, keywords=None, defaults=(None,)), ('document', 'e478180d5bc010a84f35af958cafa62c'))
-paddle.fluid.layers.lstm_unit (ArgSpec(args=['x_t', 'hidden_t_prev', 'cell_t_prev', 'forget_bias', 'param_attr', 'bias_attr', 'name'], varargs=None, keywords=None, defaults=(0.0, None, None, None)), ('document', 'fe126c58e4339410e875ab1eba246d21'))
+paddle.fluid.layers.lstm_unit (ArgSpec(args=['x_t', 'hidden_t_prev', 'cell_t_prev', 'forget_bias', 'param_attr', 'bias_attr', 'name'], varargs=None, keywords=None, defaults=(0.0, None, None, None)), ('document', 'f5a878b6166f34878376a58d7e6fa95c'))
 paddle.fluid.layers.reduce_sum (ArgSpec(args=['input', 'dim', 'keep_dim', 'name'], varargs=None, keywords=None, defaults=(None, False, None)), ('document', 'ecb55075fdf89a866bcede85e60aebad'))
 paddle.fluid.layers.reduce_mean (ArgSpec(args=['input', 'dim', 'keep_dim', 'name'], varargs=None, keywords=None, defaults=(None, False, None)), ('document', '968c9b17affaf714e5021c3dc8d68c73'))
 paddle.fluid.layers.reduce_max (ArgSpec(args=['input', 'dim', 'keep_dim', 'name'], varargs=None, keywords=None, defaults=(None, False, None)), ('document', 'd37e3a9a05c00e032d4b7876c4f6b414'))
@@ -181,7 +181,7 @@ paddle.fluid.layers.im2sequence (ArgSpec(args=['input', 'filter_size', 'stride',
 paddle.fluid.layers.nce (ArgSpec(args=['input', 'label', 'num_total_classes', 'sample_weight', 'param_attr', 'bias_attr', 'num_neg_samples', 'name', 'sampler', 'custom_dist', 'seed', 'is_sparse'], varargs=None, keywords=None, defaults=(None, None, None, None, None, 'uniform', None, 0, False)), ('document', '38297567127888e01542857839058d52'))
 paddle.fluid.layers.sampled_softmax_with_cross_entropy (ArgSpec(args=['logits', 'label', 'num_samples', 'num_true', 'remove_accidental_hits', 'use_customized_samples', 'customized_samples', 'customized_probabilities', 'seed'], varargs=None, keywords=None, defaults=(1, True, False, None, None, 0)), ('document', 'd4435a63d34203339831ee6a86ef9242'))
 paddle.fluid.layers.hsigmoid (ArgSpec(args=['input', 'label', 'num_classes', 'param_attr', 'bias_attr', 'name', 'path_table', 'path_code', 'is_custom', 'is_sparse'], varargs=None, keywords=None, defaults=(None, None, None, None, None, False, False)), ('document', 'b83e7dfa81059b39bb137922dc914f50'))
-paddle.fluid.layers.beam_search (ArgSpec(args=['pre_ids', 'pre_scores', 'ids', 'scores', 'beam_size', 'end_id', 'level', 'is_accumulated', 'name', 'return_parent_idx'], varargs=None, keywords=None, defaults=(0, True, None, False)), ('document', '1270395ce97a4e1b556104abbb14f096'))
+paddle.fluid.layers.beam_search (ArgSpec(args=['pre_ids', 'pre_scores', 'ids', 'scores', 'beam_size', 'end_id', 'level', 'is_accumulated', 'name', 'return_parent_idx'], varargs=None, keywords=None, defaults=(0, True, None, False)), ('document', '2b505ddaa309fd7b9be5445e41ca76d5'))
 paddle.fluid.layers.row_conv (ArgSpec(args=['input', 'future_context_size', 'param_attr', 'act'], varargs=None, keywords=None, defaults=(None, None)), ('document', 'a6477957b44907787b3c74157400b80c'))
 paddle.fluid.layers.multiplex (ArgSpec(args=['inputs', 'index'], varargs=None, keywords=None, defaults=None), ('document', '2c4d1ae83da6ed35e3b36ba1b3b51d23'))
 paddle.fluid.layers.layer_norm (ArgSpec(args=['input', 'scale', 'shift', 'begin_norm_axis', 'epsilon', 'param_attr', 'bias_attr', 'act', 'name'], varargs=None, keywords=None, defaults=(True, True, 1, 1e-05, None, None, None, None)), ('document', '678de6d6d0c93da74189990b039daae8'))
@@ -281,7 +281,7 @@ paddle.fluid.layers.similarity_focus (ArgSpec(args=['input', 'axis', 'indexes',
 paddle.fluid.layers.hash (ArgSpec(args=['input', 'hash_size', 'num_hash', 'name'], varargs=None, keywords=None, defaults=(1, None)), ('document', 'a0b73c21be618cec0281e7903039e5e3'))
 paddle.fluid.layers.grid_sampler (ArgSpec(args=['x', 'grid', 'name'], varargs=None, keywords=None, defaults=(None,)), ('document', '90c74742f48c70b103f1fbb9eb129066'))
 paddle.fluid.layers.log_loss (ArgSpec(args=['input', 'label', 'epsilon', 'name'], varargs=None, keywords=None, defaults=(0.0001, None)), ('document', 'ef1701e11d60508fe8f02dd2a8c60bdf'))
-paddle.fluid.layers.add_position_encoding (ArgSpec(args=['input', 'alpha', 'beta', 'name'], varargs=None, keywords=None, defaults=(None,)), ('document', 'e399f9436fed5f7ff480d8532e42c937'))
+paddle.fluid.layers.add_position_encoding (ArgSpec(args=['input', 'alpha', 'beta', 'name'], varargs=None, keywords=None, defaults=(None,)), ('document', 'bd8b28e6c1640b13a42b0524f86f7800'))
 paddle.fluid.layers.bilinear_tensor_product (ArgSpec(args=['x', 'y', 'size', 'act', 'name', 'param_attr', 'bias_attr'], varargs=None, keywords=None, defaults=(None, None, None, None)), ('document', '6755168c4b2308e1e4f54cb56fa7dcb2'))
 paddle.fluid.layers.merge_selected_rows (ArgSpec(args=['x', 'name'], varargs=None, keywords=None, defaults=(None,)), ('document', 'b2b0e5d5c155ce24bafc38b78cd0b164'))
 paddle.fluid.layers.get_tensor_from_selected_rows (ArgSpec(args=['x', 'name'], varargs=None, keywords=None, defaults=(None,)), ('document', '2c568321feb4d16c41a83df43f95089d'))
@@ -922,7 +922,7 @@ paddle.fluid.transpiler.DistributeTranspilerConfig.__init__ (ArgSpec(args=['self
 paddle.fluid.nets.simple_img_conv_pool (ArgSpec(args=['input', 'num_filters', 'filter_size', 'pool_size', 'pool_stride', 'pool_padding', 'pool_type', 'global_pooling', 'conv_stride', 'conv_padding', 'conv_dilation', 'conv_groups', 'param_attr', 'bias_attr', 'act', 'use_cudnn'], varargs=None, keywords=None, defaults=(0, 'max', False, 1, 0, 1, 1, None, None, None, True)), ('document', '5e89c978199c4ecce2b26d5fed1ec52b'))
 paddle.fluid.nets.sequence_conv_pool (ArgSpec(args=['input', 'num_filters', 'filter_size', 'param_attr', 'act', 'pool_type', 'bias_attr'], varargs=None, keywords=None, defaults=(None, 'sigmoid', 'max', None)), ('document', 'b2d435f782ac8ea3ca480b8d24e7f5b4'))
 paddle.fluid.nets.glu (ArgSpec(args=['input', 'dim'], varargs=None, keywords=None, defaults=(-1,)), ('document', '3efe197c8e3e75f84a4c464d8b74e943'))
-paddle.fluid.nets.scaled_dot_product_attention (ArgSpec(args=['queries', 'keys', 'values', 'num_heads', 'dropout_rate'], varargs=None, keywords=None, defaults=(1, 0.0)), ('document', 'b1a07a0000eb9103e3a143ca8c13de5b'))
+paddle.fluid.nets.scaled_dot_product_attention (ArgSpec(args=['queries', 'keys', 'values', 'num_heads', 'dropout_rate'], varargs=None, keywords=None, defaults=(1, 0.0)), ('document', '375898e47266633635f4c2096e1ac296'))
 paddle.fluid.nets.img_conv_group (ArgSpec(args=['input', 'conv_num_filter', 'pool_size', 'conv_padding', 'conv_filter_size', 'conv_act', 'param_attr', 'conv_with_batchnorm', 'conv_batchnorm_drop_rate', 'pool_stride', 'pool_type', 'use_cudnn'], varargs=None, keywords=None, defaults=(1, 3, None, None, False, 0.0, 1, 'max', True)), ('document', 'a59c581d5969266427e841abe69f694a'))
 paddle.fluid.optimizer.SGDOptimizer ('paddle.fluid.optimizer.SGDOptimizer', ('document', 'fc09d6e6c1083cec2dce51f6f9f4ecaf'))
 paddle.fluid.optimizer.SGDOptimizer.__init__ (ArgSpec(args=['self', 'learning_rate', 'regularization', 'name'], varargs=None, keywords=None, defaults=(None, None)), ('document', '6adf97f83acf6453d4a6a4b1070f3754'))

python/paddle/fluid/layers/nn.py

@@ -452,6 +452,7 @@ def center_loss(input,
     centers_param = helper.create_parameter(
         attr=param_attr, shape=centers_shape, dtype=dtype)
     centers_param.stop_gradient = True
+
     if isinstance(alpha, Variable):
         alpha_param = alpha
     else:
@@ -1215,13 +1216,16 @@ def dynamic_gru(input,
                 h_0=None,
                 origin_mode=False):
     """
-    **Gated Recurrent Unit (GRU) Layer**
+    **Note: The input type of this must be LoDTensor. If the input type to be
+    processed is Tensor, use** :ref:`api_fluid_layers_StaticRNN` .
 
-    if origin_mode is False, then the equation of a gru step is from paper
-    `Empirical Evaluation of Gated Recurrent Neural Networks on Sequence
-    Modeling <https://arxiv.org/pdf/1412.3555.pdf>`_ .
+    This operator is used to perform the calculations for a single layer of
+    Gated Recurrent Unit (GRU) on full sequences step by step. The calculations
+    in one time step support these two modes:
 
-    The formula is as follows:
+    If ``origin_mode`` is True, then the formula used is from paper
+    `Learning Phrase Representations using RNN Encoder Decoder for Statistical
+    Machine Translation <https://arxiv.org/pdf/1406.1078.pdf>`_ .
 
     .. math::
 
@@ -1231,12 +1235,12 @@ def dynamic_gru(input,
 
         \\tilde{h_t} & = act_c(W_{cx}x_{t} + W_{ch}(r_t \odot h_{t-1}) + b_c)
 
-        h_t & = (1-u_t) \odot h_{t-1} + u_t \odot \\tilde{h_t}
+        h_t & = u_t \odot h_{t-1} + (1-u_t) \odot \\tilde{h_t}
 
 
-    if origin_mode is True then the equation is from paper
-    Learning Phrase Representations using RNN Encoder-Decoder for Statistical
-    Machine Translation <https://arxiv.org/pdf/1406.1078.pdf>`_
+    if ``origin_mode`` is False, then the formula used is from paper
+    `Empirical Evaluation of Gated Recurrent Neural Networks on Sequence
+    Modeling  <https://arxiv.org/pdf/1412.3555.pdf>`_
 
     .. math::
 
@@ -1246,59 +1250,56 @@ def dynamic_gru(input,
 
         \\tilde{h_t} & = act_c(W_{cx}x_{t} + W_{ch}(r_t \odot h_{t-1}) + b_c)
 
-        h_t & = u_t \odot h_{t-1} + (1-u_t) \odot \\tilde{h_t}
+        h_t & = (1-u_t) \odot h_{t-1} + u_t \odot \\tilde{h_t}
 
-    The :math:`\odot` is the element-wise product of the vectors. :math:`act_g`
-    is the update gate and reset gate activation function and :math:`sigmoid`
-    is usually used for it. :math:`act_c` is the activation function for
-    candidate hidden state and :math:`tanh` is usually used for it.
+    :math:`x_t` is the input of current time step, but it is not from ``input`` .
+    This operator does not include the calculations :math:`W_{ux}x_{t}, W_{rx}x_{t}, W_{cx}x_{t}` ,
+    **Note** thus a fully-connect layer whose size is 3 times of ``size`` should
+    be used before this operator, and the output should be used as ``input`` here.
+    :math:`h_{t-1}` is the hidden state from previous time step. 
+    :math:`u_t` , :math:`r_t` , :math:`\\tilde{h_t}` and :math:`h_t` stand for
+    update gate, reset gate, candidate hidden and hidden output separately.
+    :math:`W_{uh}, b_u` , :math:`W_{rh}, b_r` and :math:`W_{ch}, b_c` stand for
+    the weight matrix and bias used in update gate, reset gate, candidate hidden
+    calculations. For implementation, the three weight matrix are merged into a
+    tensor shaped :math:`[D, D \\times 3]` , the three bias are concatenated as
+    a tensor shaped :math:`[1, D \\times 3]` , where :math:`D` stands for the
+    hidden size; The data layout of weight tensor is: :math:`W_{uh}` and :math:`W_{rh}`
+    are concatenated with shape :math:`[D, D  \\times 2]` lying on the first part,
+    and :math:`W_{ch}` lying on the latter part with shape :math:`[D, D]` .
 
-    Note that these :math:`W_{ux}x_{t}, W_{rx}x_{t}, W_{cx}x_{t}` operations on
-    the input :math:`x_{t}` are NOT included in this operator. Users can choose
-    to use fully-connect layer before GRU layer.
 
     Args:
-        input(Variable): The input of dynamic_gru layer, which supports
-            variable-time length input sequence. The underlying tensor in this
-            Variable is a matrix with shape :math:`(T \\times 3D)`, where
-            :math:`T` is the total time steps in this mini-batch, :math:`D`
-            is the hidden size.
-        size(int): The dimension of the gru cell.
-        param_attr(ParamAttr|None): The parameter attribute for the learnable
-            hidden-hidden weight matrix. Note:
-
-            - The shape of the weight matrix is :math:`(T \\times 3D)`, where
-              :math:`D` is the hidden size.
-            - All elements in the weight matrix can be divided into two parts.
-              The first part are weights of the update gate and reset gate with
-              shape :math:`(D \\times 2D)`, and the second part are weights for
-              candidate hidden state with shape :math:`(D \\times D)`.
-
-            If it is set to None or one attribute of ParamAttr, dynamic_gru will
-            create ParamAttr as param_attr. If the Initializer of the param_attr
-            is not set, the parameter is initialized with Xavier. Default: None.
-        bias_attr (ParamAttr|bool|None): The parameter attribute for the bias
-            of GRU.Note that the bias with :math:`(1 \\times 3D)` concatenates
-            the bias in the update gate, reset gate and candidate calculations.
-            If it is set to False, no bias will be applied to the update gate,
-            reset gate and candidate calculations. If it is set to None or one
-            attribute of ParamAttr, dynamic_gru will create ParamAttr as
-            bias_attr. If the Initializer of the bias_attr is not set, the bias
-            is initialized zero. Default: None.
-        is_reverse(bool): Whether to compute reversed GRU, default
-            :attr:`False`.
-        gate_activation(str): The activation for update gate and reset gate.
-            Choices = ["sigmoid", "tanh", "relu", "identity"], default "sigmoid".
-        candidate_activation(str): The activation for candidate hidden state.
-            Choices = ["sigmoid", "tanh", "relu", "identity"], default "tanh".
-        h_0 (Variable): This is initial hidden state. If not set, default is
-            zero. This is a tensor with shape (N x D), where N is the number of
-            total time steps of input mini-batch feature and D is the hidden
-            size.
+        input(Variable): A LoDTensor whose lod level is 1, representing the input
+            after linear projection. Its shape should be :math:`[T, D \\times 3]` ,
+            where :math:`T` stands for the total sequence lengths in this mini-batch,
+            :math:`D` for the hidden size. The data type should be float32 or float64.
+        size(int): Indicate the hidden size.
+        param_attr(ParamAttr, optional):  To specify the weight parameter property.
+            Default: None, which means the default weight parameter property is used.
+            See usage for details in :ref:`api_fluid_ParamAttr` .
+        bias_attr (ParamAttr, optional): To specify the bias parameter property.
+            Default: None, which means the default bias parameter property is used.
+            See usage for details in :ref:`api_fluid_ParamAttr` .
+        is_reverse(bool, optional): Whether to compute in the reversed order of
+            input sequences. Default False.
+        gate_activation(str, optional): The activation fuction corresponding to
+            :math:`act_g` in the formula. "sigmoid", "tanh", "relu" and "identity"
+            are supported. Default "sigmoid".
+        candidate_activation(str, optional): The activation fuction corresponding to
+            :math:`act_c` in the formula. "sigmoid", "tanh", "relu" and "identity"
+            are supported. Default "tanh".
+        h_0 (Variable, optional): A Tensor representing the initial hidden state.
+            It not provided, the default initial hidden state is 0. The shape is
+            :math:`[N, D]` , where :math:`N` is the number of sequences in the
+            mini-batch, :math:`D` for the hidden size. The data type should be
+            same as ``input`` . Default None.
 
     Returns:
-        Variable: The hidden state of GRU. The shape is :math:`(T \\times D)`, \
-            and sequence length is the same with the input.
+        Variable: A LoDTensor whose lod level is 1 and shape is :math:`[T, D]` , \
+            where :math:`T` stands for the total sequence lengths in this mini-batch \
+            :math:`D` for the hidden size. It represents GRU transformed sequence output, \
+            and has the same lod and data type with ``input`` .
 
     Examples:
 
@@ -1307,9 +1308,11 @@ def dynamic_gru(input,
             import paddle.fluid as fluid
 
             dict_dim, emb_dim = 128, 64
-            data = fluid.layers.data(name='sequence', shape=[1],
-                                     dtype='int32', lod_level=1)
-            emb = fluid.layers.embedding(input=data, size=[dict_dim, emb_dim])
+            data = fluid.data(name='sequence',
+                      shape=[None],
+                      dtype='int64',
+                      lod_level=1)
+            emb = fluid.embedding(input=data, size=[dict_dim, emb_dim])
             hidden_dim = 512
             x = fluid.layers.fc(input=emb, size=hidden_dim * 3)
             hidden = fluid.layers.dynamic_gru(input=x, size=hidden_dim)
@@ -1365,79 +1368,85 @@ def gru_unit(input,
              gate_activation='sigmoid',
              origin_mode=False):
     """
-    **GRU unit layer**
+    Gated Recurrent Unit (GRU) RNN cell. This operator performs GRU calculations for
+    one time step and it supports these two modes:
 
-    if origin_mode is True, then the equation of a gru step is from paper
-    `Learning Phrase Representations using RNN Encoder-Decoder for Statistical
-    Machine Translation <https://arxiv.org/pdf/1406.1078.pdf>`_
+    If ``origin_mode`` is True, then the formula used is from paper
+    `Learning Phrase Representations using RNN Encoder Decoder for Statistical
+    Machine Translation <https://arxiv.org/pdf/1406.1078.pdf>`_ .
 
     .. math::
-            u_t & = actGate(xu_{t} + W_u h_{t-1} + b_u)
 
-            r_t & = actGate(xr_{t} + W_r h_{t-1} + b_r)
+        u_t & = act_g(W_{ux}x_{t} + W_{uh}h_{t-1} + b_u)
+
+        r_t & = act_g(W_{rx}x_{t} + W_{rh}h_{t-1} + b_r)
 
-            m_t & = actNode(xm_t + W_c dot(r_t, h_{t-1}) + b_m)
+        \\tilde{h_t} & = act_c(W_{cx}x_{t} + W_{ch}(r_t \odot h_{t-1}) + b_c)
+
+        h_t & = u_t \odot h_{t-1} + (1-u_t) \odot \\tilde{h_t}
 
-            h_t & = dot(u_t, h_{t-1}) + dot((1-u_t), m_t)
 
-    if origin_mode is False, then the equation of a gru step is from paper
+    if ``origin_mode`` is False, then the formula used is from paper
     `Empirical Evaluation of Gated Recurrent Neural Networks on Sequence
     Modeling  <https://arxiv.org/pdf/1412.3555.pdf>`_
 
     .. math::
-            u_t & = actGate(xu_{t} + W_u h_{t-1} + b_u)
 
-            r_t & = actGate(xr_{t} + W_r h_{t-1} + b_r)
+        u_t & = act_g(W_{ux}x_{t} + W_{uh}h_{t-1} + b_u)
 
-            m_t & = actNode(xm_t + W_c dot(r_t, h_{t-1}) + b_m)
+        r_t & = act_g(W_{rx}x_{t} + W_{rh}h_{t-1} + b_r)
 
-            h_t & = dot((1-u_t), h_{t-1}) + dot(u_t, m_t)
+        \\tilde{h_t} & = act_c(W_{cx}x_{t} + W_{ch}(r_t \odot h_{t-1}) + b_c)
 
+        h_t & = (1-u_t) \odot h_{t-1} + u_t \odot \\tilde{h_t}
 
-    The inputs of gru unit includes :math:`z_t`, :math:`h_{t-1}`. In terms
-    of the equation above, the :math:`z_t` is split into 3 parts -
-    :math:`xu_t`, :math:`xr_t` and :math:`xm_t`. This means that in order to
-    implement a full GRU unit operator for an input, a fully
-    connected layer has to be applied, such that :math:`z_t = W_{fc}x_t`.
+    :math:`x_t` is the input of current time step, but it is not ``input`` .
+    This operator does not include the calculations :math:`W_{ux}x_{t}, W_{rx}x_{t}, W_{cx}x_{t}` ,
+    **Note** thus a fully-connect layer whose size is 3 times of GRU hidden size should
+    be used before this operator, and the output should be used as ``input`` here.
+    :math:`h_{t-1}` is the hidden state from previous time step. 
+    :math:`u_t` , :math:`r_t` , :math:`\\tilde{h_t}` and :math:`h_t` stand for
+    update gate, reset gate, candidate hidden and hidden output separately.
+    :math:`W_{uh}, b_u` , :math:`W_{rh}, b_r` and :math:`W_{ch}, b_c` stand for
+    the weight matrix and bias used in update gate, reset gate, candidate hidden
+    calculations. For implementation, the three weight matrix are merged into a
+    tensor shaped :math:`[D, D \\times 3]` , the three bias are concatenated as
+    a tensor shaped :math:`[1, D \\times 3]` , where :math:`D` stands for the
+    hidden size; The data layout of weight tensor is: :math:`W_{uh}` and :math:`W_{rh}`
+    are concatenated with shape :math:`[D, D  \\times 2]` lying on the first part,
+    and :math:`W_{ch}` lying on the latter part with shape :math:`[D, D]` .
 
-    The terms :math:`u_t` and :math:`r_t` represent the update and reset gates
-    of the GRU cell. Unlike LSTM, GRU has one lesser gate. However, there is
-    an intermediate candidate hidden output, which is denoted by :math:`m_t`.
-    This layer has three outputs :math:`h_t`, :math:`dot(r_t, h_{t-1})`
-    and concatenation of :math:`u_t`, :math:`r_t` and :math:`m_t`.
 
     Args:
-        input (Variable): The fc transformed input value of current step.
-        hidden (Variable): The hidden value of gru unit from previous step.
-        size (integer): The input dimension value.
-        param_attr(ParamAttr|None): The parameter attribute for the learnable
-            hidden-hidden weight matrix. Note:
-
-            - The shape of the weight matrix is :math:`(T \\times 3D)`, where
-              :math:`D` is the hidden size.
-            - All elements in the weight matrix can be divided into two parts.
-              The first part are weights of the update gate and reset gate with
-              shape :math:`(D \\times 2D)`, and the second part are weights for
-              candidate hidden state with shape :math:`(D \\times D)`.
-
-            If it is set to None or one attribute of ParamAttr, gru_unit will
-            create ParamAttr as param_attr. If the Initializer of the param_attr
-            is not set, the parameter is initialized with Xavier. Default: None.
-        bias_attr (ParamAttr|bool|None): The parameter attribute for the bias
-            of GRU.Note that the bias with :math:`(1 \\times 3D)` concatenates
-            the bias in the update gate, reset gate and candidate calculations.
-            If it is set to False, no bias will be applied to the update gate,
-            reset gate and candidate calculations. If it is set to None or one
-            attribute of ParamAttr, gru_unit will create ParamAttr as
-            bias_attr. If the Initializer of the bias_attr is not set, the bias
-            is initialized zero. Default: None.
-        activation (string): The activation type for cell (actNode).
-                             Default: 'tanh'
-        gate_activation (string): The activation type for gates (actGate).
-                                  Default: 'sigmoid'
+        input(Variable): A 2D Tensor representing the input after linear projection
+            after linear projection. Its shape should be :math:`[N, D \\times 3]` ,
+            where :math:`N` stands for batch size, :math:`D` for the hidden size.
+            The data type should be float32 or float64.
+        hidden(Variable): A 2D Tensor representing the hidden state from previous step.
+            Its shape should be :math:`[N, D]` , where :math:`N` stands for batch size,
+            :math:`D` for the hidden size. The data type should be same as ``input`` .
+        size(int): Indicate the hidden size.
+        param_attr(ParamAttr, optional):  To specify the weight parameter property.
+            Default: None, which means the default weight parameter property is used.
+            See usage for details in :ref:`api_fluid_ParamAttr` .
+        bias_attr (ParamAttr, optional): To specify the bias parameter property.
+            Default: None, which means the default bias parameter property is used.
+            See usage for details in :ref:`api_fluid_ParamAttr` .
+        activation(str, optional): The activation fuction corresponding to
+            :math:`act_c` in the formula. "sigmoid", "tanh", "relu" and "identity"
+            are supported. Default "tanh".
+        gate_activation(str, optional): The activation fuction corresponding to
+            :math:`act_g` in the formula. "sigmoid", "tanh", "relu" and "identity"
+            are supported. Default "sigmoid".
 
     Returns:
-        tuple: The hidden value, reset-hidden value and gate values.
+        tuple: The tuple contains three Tensor variables with the same data type \
+            as ``input`` . They represent the hidden state for next time step ( :math:`h_t` ), \
+            reseted previous hidden state ( :math:`r_t \odot h_{t-1}` ), and the \
+            concatenation of :math:`h_t, r_t, \\tilde{h_t}` . And they have shape \
+            :math:`[N, D]` , :math:`[N, D]` , :math:`[N, D \times 3]` separately. \
+            Usually only the hidden state for next time step ( :math:`h_t` ) is used \
+            as output and state, the other two are intermediate results of calculations.
 
     Examples:
 
@@ -1446,12 +1455,12 @@ def gru_unit(input,
             import paddle.fluid as fluid
 
             dict_dim, emb_dim = 128, 64
-            data = fluid.layers.data(name='step_data', shape=[1], dtype='int32')
-            emb = fluid.layers.embedding(input=data, size=[dict_dim, emb_dim])
+            data = fluid.data(name='step_data', shape=[None], dtype='int64')
+            emb = fluid.embedding(input=data, size=[dict_dim, emb_dim])
             hidden_dim = 512
             x = fluid.layers.fc(input=emb, size=hidden_dim * 3)
-            pre_hidden = fluid.layers.data(
-                name='pre_hidden', shape=[hidden_dim], dtype='float32')
+            pre_hidden = fluid.data(
+                name='pre_hidden', shape=[None, hidden_dim], dtype='float32')
             hidden = fluid.layers.gru_unit(
                 input=x, hidden=pre_hidden, size=hidden_dim * 3)
 
@@ -2028,17 +2037,14 @@ def chunk_eval(input,
                excluded_chunk_types=None,
                seq_length=None):
     """
-    **Chunk Evaluator**
-
-    This function computes and outputs the precision, recall and
-    F1-score of chunk detection.
+    This operator computes the precision, recall and F1-score for chunk detection.
+    It is often used in sequence tagging tasks, such as Named Entity Recognition(NER).
 
     For some basics of chunking, please refer to
     `Chunking with Support Vector Machines <https://aclanthology.info/pdf/N/N01/N01-1025.pdf>`_ .
 
-    ChunkEvalOp computes the precision, recall, and F1-score of chunk detection,
-    and supports IOB, IOE, IOBES and IO (also known as plain) tagging schemes.
-    Here is a NER example of labeling for these tagging schemes:
+    This operator supports IOB, IOE, IOBES and IO (also known as plain) tagging schemes.
+    Here is a NER example for the usage of these tagging schemes:
 
     .. code-block:: python
 
@@ -2052,11 +2058,11 @@ def chunk_eval(input,
        ====== ====== ======  =====  ==  ============   =====  ===== =====  ==  =========
 
     There are three chunk types(named entity types) including PER(person), ORG(organization)
-    and LOC(LOCATION), and we can see that the labels have the form <tag type>-<chunk type>.
+    and LOC(location), and we can see that the labels have the form `<tag type>-<chunk type>` .
 
-    Since the calculations actually use label ids rather than labels, extra attention
-    should be paid when mapping labels to ids to make CheckEvalOp work. The key point
-    is that the listed equations are satisfied by ids.
+    Since the implementation of this operator actually uses label ids rather than
+    label strings, to make it work, there should be a way to map label ids to
+    tag types and chunk types. This operator uses the following way to do mapping:
 
     .. code-block:: python
 
@@ -2074,8 +2080,8 @@ def chunk_eval(input,
         IOE     -     0      1     -
         IOBES   0     1      2     3
 
-    Still use NER as example, assuming the tagging scheme is IOB while chunk types are ORG,
-    PER and LOC. To satisfy the above equations, the label map can be like this:
+    Accordingly, in the above NER example, if the tagging scheme is IOB and chunk
+    types are ORG, PER and LOC, then the label ids would be as follows:
 
     .. code-block:: python
 
@@ -2087,23 +2093,32 @@ def chunk_eval(input,
        I-LOC  5
        O      6
 
-    It's not hard to verify the equations noting that the num of chunk types
-    is 3 and the num of tag types in IOB scheme is 2. For example, the label
-    id of I-LOC is 5, the tag type id of I-LOC is 1, and the chunk type id of
-    I-LOC is 2, which consistent with the results from the equations.
+    With which we can map each label id to the corresponding tag type and chunk
+    type correctly.
 
     Args:
-        input (Variable): prediction output of the network.
-        label (Variable): label of the test data set.
-        chunk_scheme (str): ${chunk_scheme_comment}
-        num_chunk_types (int): ${num_chunk_types_comment}
-        excluded_chunk_types (list): ${excluded_chunk_types_comment}
-        seq_length(Variable): 1-D Tensor specifying sequence length when input and label are Tensor type.
+        input (Variable): A Tensor or LoDTensor, representing the predicted labels
+            from the network. When it is a Tensor, its shape would be `[N, M, 1]`,
+            where `N` stands for batch size, `M` for sequence length; When it is
+            a LoDTensor, its shape would be `[N, 1]` where `N` stands for the total
+            sequence lengths in this mini-batch. The data type should be int64.
+        label (Variable): A Tensor or LoDTensor representing the ground-truth labels.
+            It shoud have the same shape, lod and data type as ``input`` .
+        chunk_scheme (str): Indicate the tagging schemes used here. The value must
+            be IOB, IOE, IOBES or plain.
+        num_chunk_types (int): The number of chunk types.
+        excluded_chunk_types (list, optional): Indicate the chunk types shouldn't
+            be taken into account. It should be a list of chunk type ids(integer).
+            Default None.
+        seq_length(Variable, optional): A 1D Tensor containing the length of each
+            sequence when ``input`` and ``label`` are Tensor. It needn't be
+            provided if ``input`` and ``label`` are LoDTensor. Default None.
 
     Returns:
-        tuple: tuple containing: precision, recall, f1_score,
-        num_infer_chunks, num_label_chunks,
-        num_correct_chunks
+        tuple: A tuple including precision, recall, F1-score, chunk number detected, \
+            chunk number in ground-truth, chunk number correctly detected. Each \
+            is a Tensor with shape `[1]`. The data type of precision, recall and \
+            F1-score all is float32, and the others' data type all is int64.
 
     Examples:
         .. code-block:: python
@@ -2112,9 +2127,9 @@ def chunk_eval(input,
 
             dict_size = 10000
             label_dict_len = 7
-            sequence = fluid.layers.data(
-                name='id', shape=[1], lod_level=1, dtype='int64')
-            embedding = fluid.layers.embedding(
+            sequence = fluid.data(
+                name='id', shape=[-1, 1], lod_level=1, dtype='int64')
+            embedding = fluid.embedding(
                 input=sequence, size=[dict_size, 512])
             hidden = fluid.layers.fc(input=embedding, size=512)
             label = fluid.layers.data(
@@ -5644,64 +5659,71 @@ def beam_search(pre_ids,
     Refer to `Beam search <https://en.wikipedia.org/wiki/Beam_search>`_
     for more details.
 
-    This layer does the search in beams for one time step. Specifically, it
-    selects the top-K candidate word ids of current step from :attr:`ids`
-    according to their :attr:`scores` for all source sentences, where K is
-    :attr:`beam_size` and :attr:`ids, scores` are predicted results from the
-    computation cell. If :attr:`ids` is not set, it will be calculated out
-    according to :attr:`scores`. Additionally, :attr:`pre_ids` and
-    :attr:`pre_scores` are the output of beam_search at previous step, they
+    **This operator only supports LoDTensor.** It is used after finishing
+    scores calculation to perform beam search for one time step. Specifically,
+    after ``ids`` and ``scores`` have been produced, it selects the top-K
+    ( `k` is ``beam_size`` ) candidate word ids of current step from ``ids``
+    according to the correspongding ``scores``. Additionally, ``pre_id`` and
+    ``pre_scores`` are the output of `beam_search` at previous step, they
     are needed for special use to handle ended candidate translations.
 
-    Note that if :attr:`is_accumulated` is :attr:`True`, the :attr:`scores`
-    passed in should be accumulated scores. Else, the :attr:`scores` are
-    considered as the straightforward scores and will be transformed to the
-    log field and accumulated the :attr:`pre_scores` in this operator.
-    Length penalty should be done with extra operators before calculating the
-    accumulated scores if needed.
+    Note that if ``is_accumulated`` is True, the ``scores`` passed in should
+    be accumulated scores. Otherwise, the ``scores`` are
+    considered as the probabilities of single step and would be transformed to
+    the log field and added up with ``pre_scores`` for final scores in this
+    operator. Length penalty should be done with extra operators before calculating
+    the accumulated scores if needed.
 
     Please see the following demo for a fully beam search usage example:
 
         fluid/tests/book/test_machine_translation.py
 
     Args:
-        pre_ids(Variable): The LodTensor variable which is the output of
-            beam_search at previous step. It should be a LodTensor with shape
-            :math:`(batch_size, 1)` and lod
-            :math:`[[0, 1, ... , batch_size], [0, 1, ..., batch_size]]` at the
-            first step.
-        pre_scores(Variable): The LodTensor variable which is the output of
-            beam_search at previous step.
-        ids(Variable): The LodTensor variable containing the candidates ids.
-            Its shape should be :math:`(batch_size \\times beam_size, K)`,
-            where :math:`K` supposed to be :attr:`beam_size`.
-        scores(Variable): The LodTensor variable containing the accumulated
-            scores corresponding to :attr:`ids` and its shape is the same as
-            the shape of :attr:`ids`.
+        pre_ids(Variable): A LodTensor variable (lod level is 2), representing
+            the selected ids of previous step. It is the output of beam_search
+            at previous step. Its shape is `[batch_size, 1]` and its lod is
+            `[[0, 1, ... , batch_size], [0, 1, ..., batch_size]]` at the
+            first step. The data type should be int64.
+        pre_scores(Variable): A LodTensor variable has the same shape and lod
+            with ``pre_ids`` , representing the accumulated scores corresponding
+            to the selected ids of previous step. It is the output of
+            beam_search at previous step. The data type should be float32.
+        ids(Variable|None): A LodTensor variable containing the candidates ids.
+            It has the same lod with ``pre_ids`` and its shape should be
+            `[batch_size * beam_size, K]`, where `K` supposed to be greater than
+            ``beam_size`` and the first dimension size (decrease as samples reach
+            to the end) should be same as that of ``pre_ids`` . The data type
+            should be int64. It can be None, which use indice in ``scores`` as
+            ids.
+        scores(Variable): A LodTensor variable containing the accumulated
+            scores corresponding to ``ids`` . Both its shape and lod are same as
+            thoes of ``ids`` . The data type should be float32.
         beam_size(int): The beam width used in beam search.
         end_id(int): The id of end token.
-        level(int, default 0): It can be ignored and mustn't change currently.
-            It means the source level of lod, which is explained as following.
-            The lod level of :attr:`ids` should be 2. The first level is source
-            level which describes how many prefixes (branchs) for each source
-            sentece (beam), and the second level is sentence level which
-            describes how these candidates belong to the prefix. The paths
-            linking prefixes and selected candidates are organized and reserved
-            in lod.
-        is_accumulated(bool, default True): Whether the input :attr:`score` is
-             accumulated scores.
-        name(str|None): A name for this layer(optional). If set None, the layer
-                        will be named automatically.
-        return_parent_idx(bool): Whether to return an extra Tensor variable 
-                        preserving the selected_ids' parent indice in pre_ids
-                        in output, which can be used to gather cell states at
-                        the next time step.
+        level(int): **It can be ignored and mustn't change currently.**
+            The 2 level lod used in this operator has the following
+            meaning: The first level describes how many beams each sample has,
+            which would change to 0 when beams of the sample all end (batch reduce);
+            The second level describes how many times each beam is selected.
+            Default 0, which shouldn't be changed currently.
+        is_accumulated(bool): Whether the input ``score`` is accumulated scores.
+            Default True.
+        name(str, optional): For detailed information, please refer 
+            to :ref:`api_guide_Name`. Usually name is no need to set and 
+            None by default.
+        return_parent_idx(bool, optional): Whether to return an extra Tensor variable
+            in output, which stores the selected ids' parent indice in
+            ``pre_ids`` and can be used to update RNN's states by gather operator.
+            Default False.
 
     Returns:
-        Variable: The LodTensor tuple containing the selected ids and the \
-            corresponding scores. If :attr:`return_parent_idx` is :attr:`True`, \
-            an extra Tensor variable preserving the selected_ids' parent indice \
-            is included.
+        tuple: The tuple contains two or three LodTensor variables. The two LodTensor, \
+            representing the selected ids and the corresponding accumulated scores of \
+            current step, have the same shape `[batch_size, beam_size]` and lod with 2 levels, \
+            and have data types int64 and float32. If ``return_parent_idx`` is True, \
+            an extra Tensor variable preserving the selected ids' parent indice \
+            is included, whose shape is `[batch_size * beam_size]` and data type \
+            is int64.
 
     Examples:
         .. code-block:: python
@@ -5713,12 +5735,12 @@ def beam_search(pre_ids,
             # at previous step.
             beam_size = 4
             end_id = 1
-            pre_ids = fluid.layers.data(
-                name='pre_id', shape=[1], lod_level=2, dtype='int64')
-            pre_scores = fluid.layers.data(
-                name='pre_scores', shape=[1], lod_level=2, dtype='float32')
-            probs = fluid.layers.data(
-                name='probs', shape=[10000], dtype='float32')
+            pre_ids = fluid.data(
+                name='pre_id', shape=[None, 1], lod_level=2, dtype='int64')
+            pre_scores = fluid.data(
+                name='pre_scores', shape=[None, 1], lod_level=2, dtype='float32')
+            probs = fluid.data(
+                name='probs', shape=[None, 10000], dtype='float32')
             topk_scores, topk_indices = fluid.layers.topk(probs, k=beam_size)
             accu_scores = fluid.layers.elementwise_add(
                 x=fluid.layers.log(x=topk_scores),
@@ -5772,28 +5794,46 @@ def beam_search(pre_ids,
 
 def beam_search_decode(ids, scores, beam_size, end_id, name=None):
     """
-    Beam Search Decode Layer. This layer constructs the full hypotheses for
-    each source sentence by walking back along the LoDTensorArray :attr:`ids`
-    whose lods can be used to restore the path in the beam search tree.
+    This operator is used after beam search has completed. It constructs the
+    full predicted sequences for each sample by walking back along the search
+    paths stored in lod of ``ids`` . The result sequences are stored in a
+    LoDTensor, which uses the following way to parse:
+
+    .. code-block:: text
+
+        If lod = [[0, 3, 6], [0, 12, 24, 40, 54, 67, 82]]
+
+        The first level of lod stands for: There are 2 samples each having 3
+        (beam width) predicted sequence.
+
+        The second level of lod stands for: The lengths of the first sample's
+        3 predicted sequences are 12, 12, 16; The lengths of the second sample's
+        3 predicted sequences are 14, 13, 15.
+
+
     Please see the following demo for a fully beam search usage example:
         fluid/tests/book/test_machine_translation.py
 
     Args:
-        ids(Variable): The LodTensorArray variable containing the selected ids
-            of all steps.
-        scores(Variable): The LodTensorArray variable containing the selected
-            scores of all steps.
+        ids(Variable): The LoDTensorArray variable containing the selected ids
+            of all steps. Each LoDTensor in it has int64 data type and 2 level
+            lod which can be used to get the search paths.
+        scores(Variable): The LodTensorArray variable containing the accumulated
+            scores corresponding to selected ids of all steps. It has the same size
+            as ``ids`` . Each LoDTensor in it has the same shape and lod as the
+            counterpart in ``ids`` , and has a float32 data type.
         beam_size(int): The beam width used in beam search.
         end_id(int): The id of end token.
-        name(str|None): A name for this layer(optional). If set None, the layer
-                        will be named automatically.
+        name(str, optional): For detailed information, please refer 
+            to :ref:`api_guide_Name`. Usually name is no need to set and 
+            None by default.
 
     Returns:
-        Variable: The LodTensor pair containing the generated id sequences \
-            and the corresponding scores. The shapes and lods of the two \
-            LodTensor are same. The lod level is 2 and the two levels \
-            separately indicate how many hypotheses each source sentence has \
-            and how many ids each hypothesis has.
+        tuple: The tuple contains two LodTensor variables. The two LodTensor, \
+            containing the full sequences of ids and the correspongding accumulated \
+            scores, have the same shape flattened to 1D and have the same 2 level \
+            lod. The lod can be used to get how many predicted sequences each sample \
+            has and how many ids each predicted sequence has.
 
     Examples:
         .. code-block:: python
@@ -5832,71 +5872,67 @@ def lstm_unit(x_t,
               param_attr=None,
               bias_attr=None,
               name=None):
-    """Lstm unit layer. The equation of a lstm step is:
-
-        .. math::
-
-            i_t & = \sigma(W_{x_i}x_{t} + W_{h_i}h_{t-1} + b_i)
-
-            f_t & = \sigma(W_{x_f}x_{t} + W_{h_f}h_{t-1} + b_f)
-
-            c_t & = f_tc_{t-1} + i_t tanh (W_{x_c}x_t + W_{h_c}h_{t-1} + b_c)
-
-            o_t & = \sigma(W_{x_o}x_{t} + W_{h_o}h_{t-1} + b_o)
-
-            h_t & = o_t tanh(c_t)
+    """
+    Long-Short Term Memory (LSTM) RNN cell. This operator performs LSTM calculations for
+    one time step, whose implementation is based on calculations described in `RECURRENT
+    NEURAL NETWORK REGULARIZATION <http://arxiv.org/abs/1409.2329>`_  .
 
-    The inputs of lstm unit include :math:`x_t`, :math:`h_{t-1}` and
-    :math:`c_{t-1}`. The 2nd dimensions of :math:`h_{t-1}` and :math:`c_{t-1}`
-    should be same. The implementation separates the linear transformation and
-    non-linear transformation apart. Here, we take :math:`i_t` as an example.
-    The linear transformation is applied by calling a `fc` layer and the
-    equation is:
+    We add forget_bias to the biases of the forget gate in order to
+    reduce the scale of forgetting. The formula is as follows:
     
     .. math::
 
-            L_{i_t} = W_{x_i}x_{t} + W_{h_i}h_{t-1} + b_i
+        i_{t} & = \sigma(W_{x_{i}}x_{t} + W_{h_{i}}h_{t-1} + b_{i})
 
-    The non-linear transformation is applied by calling `lstm_unit_op` and the
-    equation is:
+        f_{t} & = \sigma(W_{x_{f}}x_{t} + W_{h_{f}}h_{t-1} + b_{f} + forget\\_bias)
 
-        .. math::
+        c_{t} & = f_{t}c_{t-1} + i_{t} tanh (W_{x_{c}}x_{t} + W_{h_{c}}h_{t-1} + b_{c})
 
-            i_t = \sigma(L_{i_t})
+        o_{t} & = \sigma(W_{x_{o}}x_{t} + W_{h_{o}}h_{t-1} + b_{o})
 
-    This layer has two outputs including :math:`h_t` and :math:`c_t`.
+        h_{t} & = o_{t} tanh (c_{t})
+
+    :math:`x_{t}` stands for ``x_t`` , corresponding to the input of current time step;
+    :math:`h_{t-1}` and :math:`c_{t-1}` correspond to ``hidden_t_prev`` and ``cell_t_prev`` ,
+    representing the output of from previous time step.
+    :math:`i_{t}, f_{t}, c_{t}, o_{t}, h_{t}` are input gate, forget gate, cell, output gate
+    and hidden calculation.
 
     Args:
-        x_t (Variable): The input value of current step, a 2-D tensor with shape
-            M x N, M for batch size and N for input size.
-        hidden_t_prev (Variable): The hidden value of lstm unit, a 2-D tensor
-            with shape M x S, M for batch size and S for size of lstm unit.
-        cell_t_prev (Variable): The cell value of lstm unit, a 2-D tensor with
-            shape M x S, M for batch size and S for size of lstm unit.
-        forget_bias (float): The forget bias of lstm unit.
-        param_attr(ParamAttr|None): The parameter attribute for the learnable
-                               hidden-hidden weights.
-                               If it is set to None or one attribute of ParamAttr,
-                               lstm_unit will create ParamAttr as param_attr.
-                               If the Initializer of the param_attr is not set, the
-                               parameter is initialized with Xavier. Default: None.
-        bias_attr (ParamAttr|None): The bias attribute for the learnable bias
-                              weights. If it is set to False, no bias will be added
-                              to the output units. If it is set to None or one attribute of ParamAttr,
-                              lstm_unit will create ParamAttr as bias_attr.
-                              If the Initializer of the bias_attr is not set,
-                              the bias is initialized zero. Default: None.
-        name(str|None): A name for this layer(optional). If set None, the layer
-                       will be named automatically.
+        x_t(Variable): A 2D Tensor representing the input of current time step.
+            Its shape should be :math:`[N, M]` , where :math:`N` stands for batch
+            size, :math:`M` for the feature size of input. The data type should
+            be float32 or float64.
+        hidden_t_prev(Variable): A 2D Tensor representing the hidden value from
+            previous step. Its shape should be :math:`[N, D]` , where :math:`N`
+            stands for batch size, :math:`D` for the hidden size. The data type
+            should be same as ``x_t`` .
+        cell_t_prev(Variable): A 2D Tensor representing the cell value from
+            previous step. It has the same shape and data type with ``hidden_t_prev`` .
+        forget_bias (float, optional): :math:`forget\\_bias` added to the biases
+            of the forget gate. Default 0.
+        param_attr(ParamAttr, optional):  To specify the weight parameter property.
+            Default: None, which means the default weight parameter property is used.
+            See usage for details in :ref:`api_fluid_ParamAttr` .
+        bias_attr (ParamAttr, optional): To specify the bias parameter property.
+            Default: None, which means the default bias parameter property is used.
+            See usage for details in :ref:`api_fluid_ParamAttr` .
+        name(str, optional): For detailed information, please refer 
+            to :ref:`api_guide_Name`. Usually name is no need to set and 
+            None by default.
 
     Returns:
-        tuple: The hidden value and cell value of lstm unit.
+        tuple: The tuple contains two Tensor variables with the same shape and \
+            data type with ``hidden_t_prev`` , representing the hidden value and \
+            cell value which correspond to :math:`h_{t}` and :math:`c_{t}` in \
+            the formula.
 
     Raises:
-        ValueError: The ranks of **x_t**, **hidden_t_prev** and **cell_t_prev**
-                    not be 2 or the 1st dimensions of **x_t**, **hidden_t_prev**
-                    and **cell_t_prev** not be the same or the 2nd dimensions of
-                    **hidden_t_prev** and **cell_t_prev** not be the same.
+        ValueError: Rank of x_t must be 2.
+        ValueError: Rank of hidden_t_prev must be 2.
+        ValueError: Rank of cell_t_prev must be 2.
+        ValueError: The 1st dimensions of x_t, hidden_t_prev and cell_t_prev must be the same.
+        ValueError: The 2nd dimensions of hidden_t_prev and cell_t_prev must be the same.
 
     Examples:
 
@@ -5905,12 +5941,12 @@ def lstm_unit(x_t,
             import paddle.fluid as fluid
 
             dict_dim, emb_dim, hidden_dim = 128, 64, 512
-            data = fluid.layers.data(name='step_data', shape=[1], dtype='int32')
-            x = fluid.layers.embedding(input=data, size=[dict_dim, emb_dim])
-            pre_hidden = fluid.layers.data(
-                name='pre_hidden', shape=[hidden_dim], dtype='float32')
-            pre_cell = fluid.layers.data(
-                name='pre_cell', shape=[hidden_dim], dtype='float32')
+            data = fluid.data(name='step_data', shape=[None], dtype='int64')
+            x = fluid.embedding(input=data, size=[dict_dim, emb_dim])
+            pre_hidden = fluid.data(
+                name='pre_hidden', shape=[None, hidden_dim], dtype='float32')
+            pre_cell = fluid.data(
+                name='pre_cell', shape=[None, hidden_dim], dtype='float32')
             hidden = fluid.layers.lstm_unit(
                 x_t=x,
                 hidden_t_prev=pre_hidden,
@@ -15035,12 +15071,13 @@ def teacher_student_sigmoid_loss(input,
 
 def add_position_encoding(input, alpha, beta, name=None):
     """
-    **Add Position Encoding Layer**
+    This operator performs weighted sum of input feature at each position
+    (position in the sequence) and the corresponding position encoding.
 
-    This layer accepts an input 3D-Tensor of shape [N x M x P], and returns an
-    output Tensor of shape [N x M x P] with positional encoding value.
+    For more details of position encoding, please refer to `Attention Is All You 
+    Need <http://arxiv.org/pdf/1706.03762.pdf>`_ .
 
-    Refer to `Attention Is All You Need <http://arxiv.org/pdf/1706.03762.pdf>`_ .
+    The formula is as follows:
 
     .. math::
         PE(pos, 2i) &= \\sin{(pos / 10000^{2i / P})}   \\\\
@@ -15048,28 +15085,36 @@ def add_position_encoding(input, alpha, beta, name=None):
         Out(:, pos, i) &= \\alpha * input(:, pos, i) + \\beta * PE(pos, i)
 
     Where:
-      - :math:`PE(pos, 2i)` : the increment for the number at even position
-      - :math:`PE(pos, 2i + 1)` : the increment for the number at odd position
+      - :math:`PE(pos, 2i)` : the value at even index `2i` for encoding of position `pos`.
+      - :math:`PE(pos, 2i + 1)` : the value at odd index `2i+1` for encoding of position `pos`
 
     Args:
-        input (Variable): 3-D input tensor with shape [N x M x P]
-        alpha (float): multiple of Input Tensor
-        beta (float): multiple of Positional Encoding Tensor
-        name (string): the name of position encoding layer
+        input(Variable): A Tensor or LoDTensor (lod level is 1). If it is a
+            Tensor, the shape should be `[N, M, P]`, where `N` stands for
+            batch size, `M` for sequence length, `P` for the size of feature
+            dimension. If it is a LoDTensor, the shape should be `[N, P]`,
+            where `N` stands for the total sequence lengths in this mini-batch,
+            `P` for the size of feature. The data type should be float32 or float64.
+        alpha(float): Indicate the weight coefficient for `input` when performing
+            weighted sum.
+        beta(float): Indicate the weight coefficient for position encoding when
+            performing weighted sum.
+        name(str, optional): For detailed information, please refer 
+            to :ref:`api_guide_Name`. Usually name is no need to set and 
+            None by default.
 
     Returns:
-        Variable: A 3-D Tensor of shape [N x M x P] with positional encoding.
+        Variable: A Tensor or LoDTensor. It has the same shape, data type and lod as `input`.
 
     Examples:
         .. code-block:: python
 
           import paddle.fluid as fluid
 
-          tensor = fluid.layers.data(
+          tensor = fluid.data(
               name='tensor',
-              shape=[32, 64, 512],
-              dtype='float32',
-              append_batch_size=False)
+              shape=[None, 64, 512],
+              dtype='float32')
           position_tensor = fluid.layers.add_position_encoding(
               input=tensor, alpha=1.0, beta=1.0)
 

python/paddle/fluid/nets.py

@@ -363,67 +363,67 @@ def scaled_dot_product_attention(queries,
                                  num_heads=1,
                                  dropout_rate=0.):
     """
-    The dot-product attention.
-
+    This interface Multi-Head Attention using scaled dot product.
     Attention mechanism can be seen as mapping a query and a set of key-value
-    pairs to an output. The output is computed as a weighted sum of the values,
-    where the weight assigned to each value is computed by a compatibility
-    function (dot-product here) of the query with the corresponding key.
-
-    The dot-product attention can be implemented through (batch) matrix
-    multipication as follows:
+    pairs to an output. Multi-Head Attention performs attention using multi-head
+    parallel, and the inputs of attention would be transformed by linear projection.
+    The formula is as follows:
 
     .. math::
 
-            Attention(Q, K, V)= softmax(QK^\mathrm{T})V
+        MultiHead(Q, K, V ) & = Concat(head_1, ..., head_h)
+
+        where \  head_i & = Attention(QW_i^Q , KW_i^K , VW_i^V )
 
-    Refer to `Attention Is All You Need
-    <https://arxiv.org/pdf/1706.03762.pdf>`_.
+        Attention(Q, K, V) & = softmax (\\frac{QK^\mathrm{T}}{\sqrt{d_k}}) V
+
+    For more details, please refer to `Attention Is All You Need
+    <https://arxiv.org/pdf/1706.03762.pdf>`_ .
+
+    Note that the implementation is adapted to batch, and all matrix multiplication
+    in :math:`Attention(Q, K, V)` is batched matrix multiplication. Refer to
+    :ref:`api_fluid_layers_matmul` .
 
     Args:
-        queries (Variable): The input variable which should be a 3-D Tensor.
-        keys (Variable): The input variable which should be a 3-D Tensor.
-        values (Variable): The input variable which should be a 3-D Tensor.
-        num_heads (int): Head number to compute the scaled dot product
-            attention. Default: 1.
-        dropout_rate (float): The dropout rate to drop the attention weight.
-            Default: 0.0.
+        queries (Variable): A 3-D Tensor with shape :math:`[N, L_q, d_k \\times h]` ,
+            where :math:`N` stands for batch size, :math:`L_q` for the sequence length
+            of query, :math:`d_k \\times h` for the feature size of query, :math:`h` for
+            head number. The data type should be float32 or float64.
+        keys (Variable): A 3-D Tensor with shape :math:`[N, L_k, d_k \\times h]` ,
+            where :math:`N` stands for batch size, :math:`L_k` for the sequence length
+            of key, :math:`d_k \\times h` for the feature size of key, :math:`h` for head
+            number. The data type should be the same as ``queries`` .
+        values (Variable): A 3-D Tensor with shape :math:`[N, L_k, d_v \\times h]` ,
+            where :math:`N` stands for batch size, :math:`L_k` for the sequence length
+            of key, :math:`d_v \\times h` for the feature size of value, :math:`h` for head
+            number. The data type should be the same as ``queries`` .
+        num_heads (int, optional): Indicate the number of head. If the numher
+            is 1, linear projection would not be performed on inputs. Default: 1.
+        dropout_rate (float, optional): The rate to drop the attention weight.
+            Default: 0.0, which means no dropout.
 
     Returns:
-        Variable: A 3-D Tensor computed by multi-head scaled dot product\
-            attention.
+        Variable: A 3-D Tensor with shape :math:`[N, L_q, d_v \\times h]` , \
+            where :math:`N` stands for batch size, :math:`L_q` for the sequence \
+            length of query, :math:`d_v \\times h` for the feature size of value. \
+            It has the same data type with inputs, representing the output of \
+            Multi-Head Attention.
 
     Raises:
-        ValueError: If input queries, keys, values are not 3-D Tensors.
-
-    NOTES:
-        1. When num_heads > 1, three linear projections are learned respectively
-           to map input queries, keys and values into queries', keys' and values'.
-           queries', keys' and values' have the same shapes with queries, keys
-           and values.
-        2. When num_heads == 1, scaled_dot_product_attention has no learnable
-           parameters.
+        ValueError: Inputs quries, keys and values should all be 3-D tensors.
+        ValueError: The hidden size of queries and keys should be the same.
+        ValueError: The max sequence length in query batch and in key batch should be the same.
+        ValueError: he hidden size of keys must be divisible by the number of attention heads.
+        ValueError: he hidden size of values must be divisible by the number of attention heads.
 
     Examples:
         .. code-block:: python
 
             import paddle.fluid as fluid
 
-            queries = fluid.layers.data(name="queries",
-                                        shape=[3, 5, 9],
-                                        dtype="float32",
-                                        append_batch_size=False)
-            queries.stop_gradient = False
-            keys = fluid.layers.data(name="keys",
-                                     shape=[3, 6, 9],
-                                     dtype="float32",
-                                     append_batch_size=False)
-            keys.stop_gradient = False
-            values = fluid.layers.data(name="values",
-                                       shape=[3, 6, 10],
-                                       dtype="float32",
-                                       append_batch_size=False)
-            values.stop_gradient = False
+            queries = fluid.data(name="queries", shape=[3, 5, 9], dtype="float32")
+            keys = fluid.data(name="keys", shape=[3, 6, 9], dtype="float32")
+            values = fluid.data(name="values", shape=[3, 6, 10], dtype="float32")
             contexts = fluid.nets.scaled_dot_product_attention(queries, keys, values)
             contexts.shape  # [3, 5, 10]
     """