Hello @Residentx10! Let me take a stab at addressing your questions. It does look like you are referencing documentation for two different versions of Model Manager: one built on the SAS 9.4 platform and the other on Viya 4. If you would like to learn more about SAS Viya, I suggest starting here. I am not familiar with Model Manager on SAS 9.4, but I can speak to Model Manager on Viya. First, out-of-the-box, Model Manager on SAS Viya can only automatically retrain models developed in SAS Model Studio (i.e., Visual Data Mining and Machine Learning). This is because Model Manager doesn't automatically retrain the model itself - it requests Model Studio to automatically retrain the model pipeline and register the champion model. This is because model retraining requires the model training code, whereas the code stored within Model Manager is for scoring new transactions using the trained model. Model training and model inferencing/serving/scoring are two different processes. And while the scoring code is enough to use the model, it is not enough to automatically retrain the model. I have seen people leverage another tool, SAS Workflow Manager, which is shipped with Model Manager, to create their own automated retraining process for Python models by providing the required training code and some guardrails for retraining. So while it may not be out-of-the-box, you can add customizations to support automatic retraining. Some organizations prefer automated retraining, but others prefer to retrain manually to retain finer control over the process. I see the later much more often and that route is well supported in Model Manager.
Second, to score models written in specific languages, Model Manager on Viya uses an environment corresponding to that language (i.e. models written in Python run in a Python environment and models written in R run in a R environment). This allows us to work with open source models without requiring users to recode them in another language. If you notice in the documentation image you included for Model Manager on SAS 9.4, you'll see a note about Data Step set next to R. This references R models rewritten to run in SAS Data Step, not purely native R. There hasn't been a lot of requests for C, Java, and Matlab model execution, so we have focused our resources on Python and R. If you are interested in C, Java, or Matlab model execution, I encourage you to submit a feature request to document your need.
I also am not aware of any tools for CPU/GPU sizing or estimation, but compared to model training, model scoring is a not a computationally intensive process. With Model Manager on SAS Viya, we offer containerized model deployments so end-users have finer control over the inferencing architecture and can take advantage of scaling to meet traffic needs.
