Tutorial proposal: Massively cutting server costs for model training …

…with spot instances on Azure (#1718)

* Add proposal for course automation task

* Rework proposal

* Slight modification

* Update README.md

* Add essay proposal

* Add Khalid to proposal

* Delete README.md

Co-authored-by: Khashayar Etemadi <khaes@kth.se>
Co-authored-by: César Soto Valero <cesarsotovalero@gmail.com>

contributions/executable-tutorial/marcelj/README.md

@@ -0,0 +1,23 @@
+# Assignment Proposal
+
+## Title
+
+Massively cutting server costs for model training with spot instances on Azure.
+
+## Names and KTH ID
+  - Marcel Juschak (marcelj@kth.se)
+  - Khalid El Yaacoub (khalidey@kth.se)
+
+## Deadline
+
+Task 3
+
+## Category
+
+Executable Tutorial
+
+## Description
+
+Training a machine learning (ML) model is one of the core components of MLOps, e.g. for continuous deployment. However, training a model can require high-end hardware resources over many days which in turn leads to high monetary costs. Spot instances on cloud platforms like Azure are servers that can be rented for usually 10% to 25% of the original price. As a downside, access to the instance may be withdrawn at any point after a 30 second notice. Model training can take advantage of spot instances by checkpointing the training state so that training can be resumed from a checkpoint after termination and restart of the server. 
+
+In this tutorial we will show how to create a docker container that trains an ML model with checkpointing and resumes training after random termination + automatic restart of the Azure spot instance.