Run Sonnet on Google Cloud ML about sonnet HOT 5 CLOSED

FYI.
Google Cloud ML Engine already has an optimized and GPU-enabled TensorFlow version installed. you could remove the tensorflow line from setup.py to make it work,or the GPU not being used if you download the CPU-only tensorflow package from PyPI and shadowing the existing TensorFlow installation.

from sonnet.

@Shuolongbj Hi, i make Sonnet work on google Cloud after a while, i did it by compiling the library on my machine, upload the wheel file to Google Cloud Storage, and specifying the --package option when uploading the job.

After a while I came to realize that the trainign was going slow, so I started to think that the gpu was not being used. what i think, needs to be done is recompiling the sonnet library against a GPU supported Tensorflow, and it maybe work fine in that case.

Thanks for your answer. :)

from sonnet.

Hi @afhuertass,

Sonnet should be using the version of tensorflow you installed with pip.

Do you get only CPU devices when you run a session with device placement being logged?
(see here)

And do you get GPU devices when you run tensorflow ops without Sonnet?

from sonnet.

@afhuertass Can you try installing the pip package (either 'dm-sonnet' or 'dm-sonnet-gpu') ?

from sonnet.

Closed due to inactivity. Feel free to reopen the bug if you have any problem installing from pip.

from sonnet.