Skip to first unread message
康洋
Sep 19, 2016, 2:03:39 AM9/19/16
to Caffe Users
As I know, caffe store all data blobs_ for efficiently computing gradient. Is it possible for caffe to store only some partial important data blob and retain the computing chain of each blob which can be used in recompute these blob when needed? The purpose here is to reduce the GPU memory use.
Reply all
Reply to author
Forward
0 new messages