From 113cd6b31c97081150e933534db3bda9253d8278 Mon Sep 17 00:00:00 2001
From: ying <lcy.seso@gmail.com>
Date: Tue, 23 Jan 2018 13:07:05 +0800
Subject: [PATCH] add multi-head scaled_dot_product attention.

---
 python/paddle/v2/fluid/nets.py | 50 ++++++++++++++++++++++------------
 1 file changed, 33 insertions(+), 17 deletions(-)

diff --git a/python/paddle/v2/fluid/nets.py b/python/paddle/v2/fluid/nets.py
index 3390fa59461..dfae9c9391a 100644
--- a/python/paddle/v2/fluid/nets.py
+++ b/python/paddle/v2/fluid/nets.py
@@ -11,14 +11,14 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
+import pdb
 import layers
 
 __all__ = [
     "simple_img_conv_pool",
     "sequence_conv_pool",
     "glu",
-    "dot_product_attention",
+    "scaled_dot_product_attention",
 ]
 
 
@@ -179,7 +179,7 @@ def scaled_dot_product_attention(queries,
 
         .. math::
 
-            Attention(Q, K, V)= softmax(QK^\mathrm{T})V
+            Attention(Q, K, V)= softmax(QK^\mathrm{T})V
 
     Refer to `Attention Is All You Need
     <https://arxiv.org/pdf/1706.03762.pdf>`_.
@@ -195,8 +195,8 @@ def scaled_dot_product_attention(queries,
                           LoDTensor.
 
     Returns:
-        tuple: The Tensor variables representing the output and attention
-               scores.
+        Variable: The context Tensor computed by multi-head scaled dot product
+                  attention.
 
     Examples:
         .. code-block:: python
@@ -239,26 +239,42 @@ def scaled_dot_product_attention(queries,
         Returns:
           a Tensor with shape [..., n, m/n]
         """
+        if num_heads == 1: return x
+
         hidden_size = x.shape[-1]
-        #
+        # reshape the 3-D input: [batch_size, max_sequence_length, hidden_dim]
+        # into a 4-D output:
+        # [batch_size, max_sequence_length, num_heads, hidden_size_per_head].
         reshaped = layers.reshape(
-            x=x, shape=x.shape[:-1] + [num_heads, hidden_size // num_heads])
-        pass
-
-    def __combine_heads():
-        pass
-
-    q = __split_heads(quries, num_heads)
+            x=x,
+            shape=list(x.shape[:-1]) + [num_heads, hidden_size // num_heads])
+        # permuate the original dimensions into:
+        # [batch_size, num_heads, max_sequence_len, hidden_size_per_head]
+        return layers.transpose(x=reshaped, perm=[0, 2, 1, 3])
+
+    def __combine_heads(x):
+        if len(x.shape) == 3: return
+        if len(x.shape) != 4:
+            raise ValueError("Input(x) should be a 4-D Tensor.")
+
+        trans_x = layers.transpose(
+            x, perm=[x.shape[0], x.shape[2], x.shape[1], x.shape[3]])
+        return layers.reshape(x=layers.reshape(
+            x=trans_x,
+            shape=[trans_x.shape[0], trans_x[1], trans_x[2] * trans_x[3]]))
+
+    q = __split_heads(queries, num_heads)
     k = __split_heads(keys, num_heads)
     v = __split_heads(values, num_heads)
 
     key_dim_per_head = keys.shape[-1] // num_heads
-    scale = key_dim_per_head**-0.5
+    scaled_q = layers.scale(x=q, scale=key_dim_per_head**-0.5)
+    product = layers.matmul(x=k, y=scaled_q, transpose_y=True)
 
-    product = layers.matmul(x=k, y=q, transpose_y=True)
     attn_scores = layers.reshape(
         x=layers.reshape(
             x=product, shape=[-1, product.shape[-1]], act="softmax"),
         shape=product.shape)
-    context = layers.matmul(attn_scores, values)
-    return context, attn_scores
+    ctx_multiheads = layers.matmul(attn_scores, values)
+    context = __combine_heads(ctx_multiheads)
+    return context
-- 
GitLab