Hello community, I am trying to set up an EC2 server with an OpenNMT AMI but cannot find any in the AMI-list. What can I do? I tryed different regions.
Kind regards, Kai
Hello community, I am trying to set up an EC2 server with an OpenNMT AMI but cannot find any in the AMI-list. What can I do? I tryed different regions.
Kind regards, Kai
Hello,
You can use a base AMI with Docker and use the OpenNMT Docker images:
@Kai, could you please let me know, if you have managed to get OpenNMT running on AWS and, if yes, with which AMI? I have tried with the GPU instance g2.2xlarge, but training took way too long.
Thanks!
Cheers,
Martin
Hey Martin
Yes, I got it to run using this AMI: ami-0233214e13e500f77
My current throughput is about 2500 token/s, which I think is not bad at all, considering that training on CPU processed 100 token/s.
Kind regards, Kai
Hi Kai,
Thanks a lot for the reply. In my call to train.py I missed the GPUID parameter, so the GPU was never actually being used. Problem solved. My throughput is now in the 4k to 5k tok/s range on a P2 instance.