Quick View of AWS Services For ML: EC2 vs ECS vs Lambda/Sagemaker
So as the DS/ML trying to figure out which cloud services to use - I offer another quick breakdown.
Previously, I used terms like IAAS, PAAS, and FAAS - and now I just feel a little icky.
Here I take the POV of a DS who wants to deploy something and (try) to ditch the buzzwords:
-- "Hey, Cloud: here's code, please deploy it"
-- "Hey, Cloud: here's my code, but i want additional control, so I built a container image, e.g., Docker"
-- "Hey, Cloud: here's my code, but I need even more control than docker, so please give me a VM and I'll configure everything myself"
At Continual, when talking to users of the various cloud ML platforms, one of the most successful patterns we see is the use of Docker. Especially when using a higher level PAAS/FAAS service, they'll often accept docker images in addition to just passing source code. This can help get around system dependency issue.s
Using AWS as an example, the DS team might standardize on Sagemaker. When one of the advanced users hits something difficult, she often has an easier time building a local docker image and having Sagemaker run her container rather than fighting with the sundry Sagemaker services to resolve whatever system dependency issue was causing the original problem (just kidding, we know it's CUDA)
Смотрите видео Quick View of AWS Services For ML: EC2 vs ECS vs Lambda/Sagemaker онлайн без регистрации, длительностью часов минут секунд в хорошем качестве. Это видео добавил пользователь Gus Cavanaugh 04 Январь 2023, не забудьте поделиться им ссылкой с друзьями и знакомыми, на нашем сайте его посмотрели 846 раз и оно понравилось 13 людям.