multi gpu support for lstm layer

[zachluo]

The implement of lstm layer in caffe is to use a sub-net. But it seems to fail to use multi gpu when training lstm. It outputs:
Check failed: Caffe::root_solver() || root_net_ root_net_ needs to be set for all non-root solvers
Any one meet this issue?

[cypof]

Does it work with the nccl version?

#4563

On Oct 17, 2016 8:41 AM, "zachluo" notifications@github.com wrote:

The implement of lstm layer in caffe is to use a sub-net. But it seems to
fail to use multi gpu when training lstm. It outputs:
Check failed: Caffe::root_solver() || root_net_ root_net_ needs to be set
for all non-root solvers
Any one meet this issue?

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#4851, or mute the thread
https://github.com/notifications/unsubscribe-auth/AA4RXtHfw9GecVF6nlL8eSq8IFp1SWXMks5q05cagaJpZM4KYxvm
.

[junshi15]

this PR addresses the problem.
#4806

[zachluo]

as @junshi15 says, it works !

[shelhamer]

Note this is solved by further releases in the switch to NCCL #4563, which is included in 1.0.