Pragmatic and Standardized MLOps - Maria Vechtomova

Описание

Summary:

In this YouTube video, the presenter provides an overview of a data community event and shares links to upcoming workshops and interviews. The video then delves into deploying machine learning models across departments, using Kubernetes and AWS, and highlights the Marvelous MLOps blog as a resource for ML knowledge. The creator discusses posting memes, newsletters, and LinkedIn content while defining envelopes as a data science deployment tool. The video also covers the overwhelming landscape of MLOps tools and the importance of focusing on organizational structure and utilizing existing tools. Model registry and monitoring options, including Artifactory and S3 buckets, are discussed, along with a maturity assessment with 60 questions that start with traceability and reproducibility. The importance of guidelines and team structure for successful MLOps in large organizations is emphasized, along with reusable CI/CD pipelines with a cookie-cutter repository for data scientists to simplify deployment and packaging. The video also covers implementing machine learning models and addressing deployment errors with a team of three senior profiles and discusses the necessary skills for implementing ML tools and the importance of standardized monitoring. The presenter also discusses an internal use case for LLMS, the potential for the LLM Ops course, and challenges with non-English languages. The video concludes with a discussion on data science teams and their collaboration with food retailers in different countries, a review of data science and machine learning courses for software engineers, the importance of data engineering, and resource recommendations.

Key Takeaways:
- The video is about a data community event and upcoming workshops and interviews
- The presenter talks about deploying machine learning models across departments using Kubernetes and AWS, and highlights the Marvelous MLOps blog as a resource for ML knowledge
- The video covers MLOps tools, model registry and monitoring, and the importance of organizational structure and utilizing existing tools
- The presenter emphasizes the importance of guidelines and team structure for successful MLOps in large organizations, along with reusable CI/CD pipelines with a cookie cutter repository for data scientists to simplify deployment and packaging
- The video also covers implementing machine learning models and addressing deployment errors with a team of three senior profiles, and discusses necessary skills for implementing ML tools and the importance of standardized monitoring
- The presenter also discusses an internal use case for LLMS, potential for LLM Ops course, and challenges with non-English languages
- The video concludes with a discussion on data science teams and their collaboration with food retailers in different countries, a review of data science and machine learning courses for software engineers, and the importance of data engineering, along with resource recommendations.

We talked about:

00:00 DataTalks.Club intro
03:07 Maria's background
09:45 Marvelous MLOps
11:10 Maria's definition of MLOps
14:45 Alternate team setups without a central MLOps team
17:55 Pragmatic vs non-pragmatic MLOps
18:41 Must-have ML tools (categories)
22:23 Maturity assessment
24:01 What to start with in MLOps
28:42 Standardized MLOps
35:21 Convincing DevOps to implement
39:29 Understanding what the tools are used for instead of knowing all the tools
42:53 Maria's next project plans
45:44 Is LLM Ops a thing?
49:42 What Ahold Delhaize does
54:05 Resource recommendations to learn more about MLOps
57:14 The importance of data engineering knowledge for ML engineers

Links:

- LinkedIn: https://www.linkedin.com/company/marvelous-mlops/
- Website: https://marvelousmlops.substack.com/

Free MLOps course: https://github.com/DataTalksClub/mlops-zoomcamp
Join DataTalks.Club: https://datatalks.club/slack.html
Our events: https://datatalks.club/events.html