[Docs] Update documentation description of GroupEmbedding.

Signed-off-by: JunqiHu <[email protected]>

Author: Mesilenceki

Diff for: docs/docs_en/Group-Embedding.md

@@ -22,16 +22,20 @@ GroupEmbedding provides two levels of API.The one is `tf.nn.group_embedding_look
 def group_embedding_lookup_sparse(params,
                                   sp_ids,
                                   combiners,
-                                  partition_strategy="mod",
                                   sp_weights=None,
+                                  partition_strategy="mod",
+                                  is_sequence=False,
+                                  params_num_per_group=sys.maxsize,
                                   name=None):
 ```
 
 - `params` : List, This parameter could receive one or more EmbeddingVariables or native Tensorflow Variable.
 - `sp_ids` : List | Tuple , SparseTensor sp_ids ​​is the ID used for EmbeddingLookup, the length must be consistent with params.
 - `combiners` : List | Tuple，The pooling method of embedding values.Currently support `mean` and `sum`.
-- `partition_strategy` : str，Currently not supported.
 - `sp_weights` : List | Typle the weight of sp_ids values.
+- `partition_strategy` : str，Currently not supported.
+- `is_sequence` : bool, Op would return Tensor shape of [B, T, D] if True
+- `params_num_per_group` : int, This parameter indicates the number of Variables inside each Op. The default setting is the maximum value. The default value is suitable for GPU scenarios; when using the CPU, it is recommended to set the smaller the better.
 - `name` : str group name
 
 **group_embedding_lookup**

Diff for: docs/docs_zh/Group-Embedding.md

@@ -26,16 +26,20 @@ Group Embedding功能支持同时对多个EmbeddingVariable 聚合查询，将
 def group_embedding_lookup_sparse(params,
                                   sp_ids,
                                   combiners,
-                                  partition_strategy="mod",
                                   sp_weights=None,
+                                  partition_strategy="mod",
+                                  is_sequence=False,
+                                  params_num_per_group=sys.maxsize,
                                   name=None):
 ```
 
 - `params` : List, 该参数可以接收一个或者多个EmbeddingVariable或者是原生Tensorflow Variable
 - `sp_ids` : List | Tuple , SparseTensor ，values是用于查找的ID 长度必须和params保持一致
 - `combiners` : List | Tuple 查找完得到的embedding tensor聚合的方式，支持 `mean` 和 `sum`
-- `partition_strategy` : str 目前暂时不支持
 - `sp_weights` : List | Typle sp_ids 的 values 的权重。
+- `partition_strategy` : str 目前暂时不支持
+- `is_sequence` : bool 如果设置为True，则返回的embedding形状为（B, T, D）。
+- `params_num_per_group` : int 该参数表示每个Op内部Variable的个数，默认设置为最大值，默认值适用于GPU的场景；当使用CPU的时候建议设置越小越好
 - `name` : str group的名称
 
 **group_embedding_lookup**
@@ -55,7 +59,7 @@ def group_embedding_lookup(params,
 **group_embedding_column_scope**
 
 ```python
-def group_embedding_column_scope(name=None):
+def group_embedding_column_scope(name=None, params_num_per_group=sys.maxsize):
 ```
 
 - `name` ： scope的名称

Diff for: tensorflow/python/ops/embedding_ops.py

@@ -1608,10 +1608,11 @@ def group_embedding_lookup_sparse(params,
       is_sequence: bool
                 return list of `Tensor` of shape `[batch_size, D]` when is False
                 return list of `Tensor` of shape `[batch_size, T, D]` when is True
-      sub_group_size: int
-                A string specifying the grouping strategy of group embedding op.["gpu", "cpu"]
-                are supported. Setting "gpu" will group all embeddings to maximize the GPU utilization.
-                "cpu" will split embeddings so as to maximize intra parallelism.
+      params_num_per_group: int
+                The number of params in GroupEmbedding op.Function will schedule len(params) // params_num_per_group + 1
+                GroupEmbedding Op. Default setting would launch one Op containing all params which is suitable for GPU scenarios
+                to maximize the GPU utilization.On the contrast, you could set value to 1 when Op
+                is placed on CPU so as to maximize inter parallelism.
       name: The operations name
     Returns
     -------