I am working on retraining the already trained ResNet with Siamese Network with Triplet Loss and those who don't know, it is very high memory + resource intensive task. So I want to know which service I should use for training the network.
The reason I want to use AWS is that someone told me that there is a library Boto which you can use to save your data. My data would be around 25+ GBs.
I want min 32GB RAM, 128GB SSD, GPU or TPU (please suggest which one to use). Also, how can I get all my data in the permanent storage from AWS s3 links? I have a script to download but it takes too much time to download all those and if the memory is temporary, I'll lose it. Like in Google Colab. Cost is not an issue but the service and permanent memory management is.
