Examining model artifacts and exporting models

A model artifact contains one or several files that are produced by a training job and that are required for model deployment. The number and nature of these files depend on the algorithm that was trained. As we've seen many times, the model artifact is stored as a model.tar.gz file, at the S3 output location defined in the estimator.

Let's look at different examples, where we reuse artifacts from the jobs we previously trained.

Examining and exporting built-in models

Almost all built-in algorithms are implemented with Apache MXNet, and their artifacts reflect this. For more information on MXNet, please visit https://mxnet.apache.org/.

Let's see how we can load these models directly. Another option would be to use Multi Model Server (MMS) (https://github.com/awslabs/multi-model-server), but we'll proceed as follows:

1. Let's start from the artifact for the Linear Learner model we trained in...