未验证 提交 fd19ee01 编写于 作者: X xsrobin 提交者: GitHub

add FLAGS_communicator_merge_sparse_grad (#932)

上级 9b698ce9
......@@ -106,6 +106,25 @@ Note
This flag has strong relationship with trainer thread num. The default value should be the same with thread num.
communicator_merge_sparse_grad
*******************************
(since 1.5.0)
merge sparse gradient before sending.
Values accepted
---------------
Bool. The default value is True.
Example
-------
FLAGS_communicator_merge_sparse_grad=True will merge sparse gradient before sending.
Note
-------
Merging sparse gradient would be time-consuming. If the sparse gradient has many duplicated ids, it will save memory and communication could be much faster. Otherwise it will not save memory.
communicator_min_send_grad_num_before_recv
*******************************************
(since 1.5.0)
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册