PyTorch GPU Scoring With Sagemaker, Dask, and Coiled
This is a quick screencast showing PyTorch model inferencing across 20 GPU workers via Dask and Coiled. The code is from a wonderful example by Saturn Cloud and Stephanie Kirmer which I'm running on a Coiled cluster from an Amazon Web Services (AWS) Sagemaker notebook.
Running this job across 5 T4 GPU machines took ~ 2mins. Maybe that's fine, maybe that's not. So I just asked Coiled for 10 machines and the same job ran in 1 minute. Not bad. In the video I scale up to 20 GPU machines just for the heck of it.
And while GPUs are expensive, renting them for short periods of time from Coiled is not. Overall I've spent $1.50 playing around with 20 GPU workers for 45 mins.
Смотрите видео PyTorch GPU Scoring With Sagemaker, Dask, and Coiled онлайн без регистрации, длительностью часов минут секунд в хорошем качестве. Это видео добавил пользователь Gus Cavanaugh 04 Октябрь 2021, не забудьте поделиться им ссылкой с друзьями и знакомыми, на нашем сайте его посмотрели 110 раз и оно понравилось 2 людям.