[
https://issues.apache.org/jira/browse/SINGA-388?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16588970#comment-16588970
]
ASF subversion and git services commented on SINGA-388:
-------------------------------------------------------
Commit a44a01c0f291cfca8a688570e3c752b1ef6ec829 in incubator-singa's branch
refs/heads/master from xuewanqi
[ https://git-wip-us.apache.org/repos/asf?p=incubator-singa.git;h=a44a01c ]
SINGA-388 Develop some RNN layers by calling tiny operations like matmul,
addbias
- Add unit test case for both vallina rnn and lstm.
The unit test cases include gradients shape check as well as value check
compared with numerical
calculation results.
- Add device_check() to valina_rnn and lstm, this function can check the device
of inputs and
paramerters. If the devices of them are not the same, the funciton can
transfer them on a same
device.
- fix some bugs in test cases and source codes.
> Develop some RNN layers by calling tiny operations like matmul, addbias.
> ------------------------------------------------------------------------
>
> Key: SINGA-388
> URL: https://issues.apache.org/jira/browse/SINGA-388
> Project: Singa
> Issue Type: New Feature
> Reporter: Xue Wanqi
> Priority: Major
>
> RNN layer can be divided as many smaller operations like matmul, addbias. in
> the ticket, some RNN layers will be developed by calling these tiny
> operations.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
