Simple end to end workflow from PyTorch model building to DJL inference

PyTorch is a fast growing and very popular open source Machine Learning framework. Its imperative design combined with “numpy” like workflow makes it a compelling first choice for beginners and professionals alike. However, serving these model in production is not straightforward and things are particularly difficult if the goal is to serve them natively in Java.

Amazon’s Deep Java Library (DJL) aims to solve this particular pain point by providing high level APIs that can run inference on PyTorch models with very little code. My recent test drive with DJL tells me that it could be a very powerful tool but the existing example set and community guidance (aka stackoverflow help :) ) could be a little intimidating for new folks, specially those who come from python background and are unfamiliar with the java style. This simple demo, hopefully, makes things easier for them. All the scripts are also available at my git repo here.

#deep-learning #pytorch #machine-learning #java #aws #pytorch model in deep java library