Munich MLOps Community Meetup #5

Hello fellow MLOps Engineers and ML Enthusiasts

We're announcing the date of our next meetup event - we're back on Thursday, November 16th at 18:30, at TUM.ai, Student Initiative Space.

Event agenda:

18:30 - 18:50 - Beers & Networking

18:50 - 19:00 - Opening talk from the host of the event

19:00 - 19:40 - Why RAG is hack and how to make it better? - Muhtasham Oblokulov, Machine Learning Engineer @ MunichRe

19:40 - 20:20 - Fast-Track to Distributed Training: Let’s Discuss SoTa Parallelism Strategies - Ekin Karabulut, Data Scientist @ Run:ai

20:20 - Onwards - Beer, pizza & networking

More about the talks:

1. Talk #1: Why RAG is hack and how to make it better?
2. Talk #2: Fast-Track to Distributed Training: Let’s Discuss SoTa Parallelism Strategies
   Exploring distributed training can be tricky for data scientists. The landscape is crowded with tool guides and options, making it challenging to find a straightforward path - specific to your usecase. In this talk, we'll cut through the confusion and discuss when distributed training is needed and what state-of-the-art is. We'll break down practical strategies, explaining why it's necessary as datasets and models get bigger. We'll explore techniques like data parallelism that make experiments faster and handle large batches for better generalization. Additionally, we'll chat about different model parallelism techniques designed for enormous models that can't fit into a single node. To wrap things up, we'll take a look at the current strategies offered by various tools and libraries. This insight will help you navigate the choices and find the right approach for your use case.

About our speakers:

1. Muhtasham Oblokulov (LinkedIn)
   Muhtasham is prompt engineer at Munich Re and LLM hacker at Munich NLP.
2. Ekin Karabulut (LinkedIn)
   Ekin is a data scientist, and GPU enthusiast based in Munich. During her Master’s studies, she dove into the privacy implications of federated learning with deep neural networks while also gaining practical experience as a data scientist. Through her journey, she focused on distributed training techniques and observed inefficiencies in GPU usage both in research and industry settings. As a Data Scientist at Run:ai she thus established the “AI Infrastructure Club” Discord server - to discuss all things GPU-related, exchange ideas, and support each other in making the most out of GPUs. Currently, she is diving deeper into the efficient usage of large models in production.

Looking forward to your RSVPs and to meeting you there!

Keep on hacking!🤩🤩🤩