[
https://issues.apache.org/jira/browse/SINGA-309?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15964038#comment-15964038
]
ASF subversion and git services commented on SINGA-309:
-------------------------------------------------------
Commit 38da7891440f912845a27b4e77985c6173e1edbe in incubator-singa's branch
refs/heads/master from wangwei
[ https://git-wip-us.apache.org/repos/asf?p=incubator-singa.git;h=38da789 ]
SINGA-309 Update the layer setting/config dynamically
Update the layer settings and internal variables (e.g cudnn descriptors) if the
input data's batchsize is changed from previous iteration.
TODO(wangwei) update the layer settings if the input sample shape changes.
> Update the layer setting/config dynamically
> -------------------------------------------
>
> Key: SINGA-309
> URL: https://issues.apache.org/jira/browse/SINGA-309
> Project: Singa
> Issue Type: Improvement
> Reporter: wangwei
> Assignee: wangwei
>
> Layers like convolution, pooling, dense, are configured before the
> training/inference.
> Their internal variables (e.g., buffers and weight matrix) are initialized in
> the setup phase or in the first iteration of training/inference.
> Sometimes, the input data size may change, e.g., the batchsize is different
> for training and validation/test.
> This ticket is to dynamically update the layer config/setting and its
> internal variables when the input data/tensor size is changed.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
