Open FerdiMT opened 5 years ago
mielgosez
commented
5 years ago In my case it takes 30 mins per epoch (I was planning to do 13, but now I will try with 6). Regarding the resources, you will find a list of solutions (PaaS, IaaS) that we can consider, we can discuss what is the best alternative for our team:
Axel-Bravo
commented
5 years ago @mielgosez Not yet useful to check the ressources, as they are paying . Better try alternatives as a full free year . Google seems to be $300 free credit, worth the effort to try it 😄 !
mielgosez
commented
5 years ago I missed that train. I signed up GCP for a training that I never completed, and I am no longer entitled with those benefits. Sooo, please learn from my mistakes and enjoy your 1-year trial :)
FerdiMT
commented
5 years ago Can't you create an account with another creditcard number and mail? :D
Axel-Bravo
commented
5 years ago Any advancements on your side @FerdiMT 🤓 ?
Hello All:
So I don't know in your case, but for me, developing this last challenge in Colab has become basically a matter of getting the NN to run in the morning and hope colab do not collapse and can upload the results by the end of the day (and no more than 15 epochs). Therefore, if we want to do trial-error on the networks we are doing and try different architectures, I think is very important that we find a way to get more processing power. With this I don't have the intention to put pressure on Axel or the group to get some instances (I completely understand we need a bit of credibility/consisency to ask for it), but among us to figure out if there is any way we could. For instance, I've seen google cloud (https://cloud.google.com/) gives you basically 1 year for free (I have to check it closely though), but maybe we can have something like that.
If this is not an issue for you, then of course I can adapt ;) Just sharing pain points.