Refine dot_product_attention

81abcdea · ranqiu · 2e1cd331 · 81abcdea
隐藏空白更改
内联并排

Showing with 2 addition and 4 deletion

python/paddle/trainer_config_helpers/networks.py python/paddle/trainer_config_helpers/networks.py +2 -4

未找到文件。
--- a/python/paddle/trainer_config_helpers/networks.py
+++ b/python/paddle/trainer_config_helpers/networks.py
@@ -1476,10 +1476,8 @@ def dot_product_attention(encoded_sequence,
        expand_as=encoded_sequence,
        name='%s_expand' % name)
-    m = linear_comb_layer(
+    m = dot_prod_layer(
-        weights=expanded,
+        input1=expanded, input2=encoded_sequence, name='%s_dot-product' % name)
-        vectors=encoded_sequence,
-        name='%s_dot-product' % name)
    attention_weight = fc_layer(
        input=m,