#16: Compressing Foundation Models as Easy as Image Compression? by M. Genzel

Join our free BLISS Speaker Series Summer 2025!

We are excited to feature Dr. Martin Genzel, Senior Research Engineer at Merantix Momentum, who will discuss "Can Compressing Foundation Models be as Easy as Image Compression?", lasting approximately 45 minutes. After the talk, seize the opportunity to connect with fellow AI enthusiasts to share ideas and questions while enjoying free drinks. Door close by 7.15pm, so please come early! Also, "attend"ing (RSVP) here on Meetup is strictly necessary to be guaranteed entry.
Please note that Meetup has recently been quite keen on promoting its Plus program. However, you are not obligated to purchase it, as both our events and the platform remain free.

Abstract: The widespread adoption of Foundation Models, especially LLMs, is often hindered by their substantial size and computational demands, especially in resource-limited settings. While post-training compression offers a promising avenue to mitigate these challenges, the process can feel like a "black box" for the user, requiring significant expertise and trial-and-error to find the right balance between model size and performance. This talk introduces Any Compression via Iterative Pruning (ACIP), a novel algorithmic approach designed with the user in mind. ACIP allows for intuitive and direct control over the compression-performance trade-off, akin to compressing an image. It leverages a single gradient descent run of iterative pruning to establish a global parameter ranking, from which models of any target size can be immediately materialized. ACIP demonstrates strong predictive performance on downstream tasks without costly fine-tuning. Across various open-weight LLMs, it achieves state-of-the-art compression results compared to existing factorization-based methods. Moreover, it seamlessly complements common quantization techniques for even greater compression.

We are BLISS e.V., the AI organization in Berlin that connects like-minded individuals who share great interest and passion for the field of machine learning. This summer 2025, we will host an exciting speaker series on site in Berlin, featuring excellent researchers from Merantix Momentum, Meta AI, Inria, Microsoft AI4Science, Google DeepMind, and University of Oxford.
Website: https://bliss.berlin
Youtube: https://www.youtube.com/@bliss.ev.berlin

Disclaimer: By attending this event you agree to be photographed.