I‘ve opened a pull request on the runtime repository. I want to test my code on the server. Could someone merge it? In addition, is it possible to increase the running time limit of a single epsilon to 1 hour? Thanks.
Hi @hecci,
We’ve merged the PR and updated the competition container. You should be able to run code that imports cupy
now. We also increased the running time to 1 hour as requested. Good luck!
Robert
Hi, it seems that 1 hour is not enough. Could you increase the running time limit to 3 hours? Thanks.
Thanks for merging the PR. I noticed the total running time for 3 epsilons is increased to 1 hours, but my code requires up to 1 hour per epsilon. Is there any chance that the total running time can be increased to 3 hours? Thank you.
@hecci - Sure thing, the runtime limit has been increased to 3h total. Good luck.
Will there be a time limit on final submissions? Or only submissions to the prescreened leaderboard?
@rmckenna As long as the final submissions will finish in a reasonable amount of time (like 4-6 hours max) then you don’t have to worry about it. You can assume the input files for final scoring will be the same order of magnitude in size.
What are the specs of the machine it will be run on? What if it takes more like 24 hours? Will such submissions be rejected?
You can assume an AWS p2.8xlarge or comparable host.
Offhand 24 hours seems excessive – is that necessary for your method?
Yeah some of the methods I am considering require that much time (per epsilon). For that host would we be able to utilize all cores or will you be running other in parallel on those resources?
@rmckenna We’ve conferred with the organizers, and the maximum time per epsilon in final scoring will be 4 hours (so 8 hours total).
For that host would we be able to utilize all cores or will you be running other in parallel on those resources?
You can use all cores.
Thank you for the clarification. Can we utilize the 8 hours total in different ways than an even split? E.g., we may want to use 6 hours for Epsilon=10.0 and 2 hours for epsilon=1.0. Would that be acceptable?
@rmckenna Sure, that will be fine.