Can Your Laptop Run AI?

LLM Benchmarks on CPU/GPU, and How to Open It to the Community at Zero Cost

Don't forget to get your ticket on Eventbrite to register for the event: https://www.eventbrite.it/e/can-your-laptop-run-ai-tickets-1985370183664

More and more enthusiasts are trying to run AI/ML models directly on their own laptops. However, there has been a lack of objective, standardized, and reproducible third-party benchmarks focused on technical performance, such as tokens per second or training time. In fact, there was a lack of them.

This talk will cover two main topics. On the one hand, it will present the results of real benchmarks on AI workloads executed on local hardware, including LLM inference (3B to 14B) and tabular ML models, comparing CPU and GPU performance, latency, and model scaling.
On the other hand, a significant part of the talk will focus on the making-of of the project, which was built entirely using open tools and at zero cost. Moving from an idea, to an individual project, to a solution that anyone can contribute to required a fairly articulated architecture and a mix of tools: not only Python code, but also automation with GitHub Actions, managed DuckDB through MotherDuck, Streamlit Cloud, and much more.

Speaker: Alberto Danese is Head of Data Science & Advanced Analytics at Nexi, where he leads the development of machine learning products and data-driven solutions for the digital payments sector. He has more than 15 years of experience in data science and engineering and is a Kaggle Competitions Grandmaster. He often shares experiences and projects related to AI and data science through technical and managerial talks, as well as through his blog, All About Data.