Feature: Add inference recipes

## Describe the feature you'd like
It is nice that there are so many pre-training and fine-tuning examples in this project. It would be great to add examples of how to deploy these models and use them for offline/batch job and online/service inference.

## How would this feature be used?
Once I pre-train or fine-tune a model, I would like to serve it on SageMaker HyperPod, and use it either by submitting curl requests, or through a simple grad.io or other UI.

Start by reviewing the repository's existing pre-training and fine-tuning examples to understand how recipes are organized. Add examples covering deployment on SageMaker HyperPod for offline or batch jobs and online or service inference, with curl requests or a simple UI as the usage path.

Written by the indexing model from the issue text.