Hi,
the deploy notebook seems to have some issues and in any case I don't see instructions for deploying after fitting the estimator.
It would be nice to have a working one :)
Hi Luca! supercool example-project you made here!
aktually I just wanted to use the latest version of pytorch and now ended up questioning the idea behind sagemaker's PyTorch-Estimator abstraction. This estimator-class wants a framework-version as argument, which is determining the docker-image that sagemaker uses? But in your requirements.txt you install another pytorch version. Sagemaker's PyTorch abstraction takes a docker-image-uri as an argument. Now I'm really confused what this PyTorch-Estimator-class is actually good for? Why not "simply" using the generic Estimator-Class and providing it with the docker-image of choice?