Skip to content
This repository was archived by the owner on Nov 17, 2023. It is now read-only.
This repository was archived by the owner on Nov 17, 2023. It is now read-only.

Gluon GPU memory efficiency #7582

Description

@hnhuang

For bugs or installation issues, please provide the following information.
The more information you provide, the more likely people will be able to help you.

Environment info

Operating System:
Ubuntu 16.04
Package used (Python/R/Scala/Julia):
Python
MXNet version:
0.11
Python version and distribution:
2.7

I am running a vgg+fcn model on keras + tensorflow and gluon. I can set batch_size = 32 on keras + tensorflow, but I can only set batch_size = 20 on gluon.

The last block of my model on gluon is:

block 7

self.net = nn.Sequential(prefix='net')
self.net.add(nn.Conv3D(256, kernel_size = 3, padding = 1, activation='relu'))
self.net.add(nn.Conv3D(4096, kernel_size = 1, padding = 0, activation='relu'))
self.net.add(nn.Conv3D(4096, kernel_size = 1, padding = 0, activation='relu'))
self.net.add(nn.Conv3D(max, kernel_size = 1, activation='sigmoid'))

The corresponding part on keras+tf is:

x = Conv3D(256, (3, 3, 3), padding='same', activation='relu', kernel_initializer='normal', name='rpn_conv1')(base_layers)
x = Conv3D(4096, (1, 1, 1), padding='same', activation='relu', kernel_initializer='normal', name='rpn_fc1')(x)
x = Conv3D(4096, (1, 1, 1), padding='same', activation='relu', kernel_initializer='normal', name='rpn_fc2')(x)
x_dist = Conv3D(max, (1, 1, 1), activation='sigmoid', kernel_initializer='uniform', name='rpn_out_class')(x)

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions