MLOps

This is the repository for the labs/tutorials of a course about Machine Learning Operations (MLOps), ZHAW.

NOTE: Labs might still be in the process of being updated, please refresh right before the corresponding lab is touched in the lecture!

Table of Contents

Lab No.	Contents	Tools	Link
1	Deep Learning Recap	PyTorch, CNNs, Transformers	Lab 01
2	Prompt engineering a chatbot	🤗Transformers, Streamlit	Lab 02
3	Experiment management and hyperparameter tuning	MLflow, Ray Tune	Lab 03
4	CI/CD and testing for code and models	GitHub Actions, PyTest, DeepChecks (CML, SkyPilot)	Lab 04
5	From notebooks to pipelines to batch processing	Ploomber, Airflow	Lab 05
6	Data-centric MLOps	Git LFS, DVC, Albumentations, 🤗Diffusers	Lab 06
7	Deploying and protecting machine learning models	MLServer, Alibi Detect	Lab 07
	Creating a (shared) VM in GCP	Google Cloud	GCP
	Vertex AI tutorial	Google cloud Vertex AI	VertexAI
	Example showing how to use python packages with Jupyter Notebooks	Jupyter	Sample

Setup

We use conda environments to manage the lab dependencies. Every lab has its own conda environment. Install Anaconda, Miniconda or Mamba for your platform.

We further recommend Windows Users to use the Windows Subsystem for Linux (WSL) as some software we are using does not support Windows.

Finally, it is a good idea to install Docker Desktop as some of the labs contain parts that benefit from Docker (it is not a requirement, though).

Creating environments with conda

Once you have conda installed, you can create an environment from a env.yaml file using the following command:

conda env create -f env.yaml

Then, to enter the environment:

conda activate <environment name>

So, for the first lab this command becomes:

conda activate mlops-lab-01

Running Notebooks

Most labs will make use of Jupyter Notebooks, which you can either run locally or on Google Colab. There are of course many other ways to run them, and you are free to use whichever tool and setup you want, but we cannot guarantee compatibility.

Below we show you two ways of running the lab notebooks that are known to work.

We recommend beginning each lab by running the notebooks locally, and only switching to Colab for longer computations. Images sometimes do not render correctly in Colab, and some features might not work as expected.

Running locally

To run notebooks locally, proceed as follows:

1. Open a terminal and navigate to the lab directory (e.g. cd lab01).
2. Activate the conda environment for this lab (see above e.g. e.g. conda activate mlops-lab-01).
3. Run jupyter lab. This will result in output similar to the following:

    To access the server, open this file in a browser:
        file:///some/long/path/here/jpserver-74325-open.html
    Or copy and paste one of these URLs:
        http://localhost:8888/lab?token=token
        http://127.0.0.1:8888/lab?token=token

4. Click on or copy the link and past it in your browser.
5. Now, click on the jupyter notebook of your liking.

Running in Google Colab

1. Clone or download this repository.
2. Navigate your browser to colab.research.google.com.
3. In the menu, select File > Upload notebook
4. Select the notebook you want to open.
5. Once your notebook is open, add a code cell at the very top. You can do this, by hovering your cursor over the top of the notebook until the Code and Text buttons appear as shown in the image below:
6. In the code cell, add the following:

!git clone https://github_pat_[PAT]@github.com/MLOps-ZHAW/MLOps_Labs
%cd MLOps_Labs/[path/to/notebook]

where [PAT] is a Github Access Token (looks like gitgub_pat_....., your lecturer will tell you), and you replace [path/to/notebook] with the path to the jupyter notebook you just opened. For instance, if you opened lab01a_01_tensor_tutorial.ipynb, the statement would become %cd MLOps_Labs/lab01.

7. Connect to a GPU by changing the runtime type: