Quantization, Performance, and Why Rust Keeps AI from Corroding

Lawrence Freeman will be joining us again for another talk on all things Rust. Heres a taster for what to expect this month; Floating-point numbers are convenient but costly, imprecise, and prone to subtle errors that degrade performance and correctness - especially in AI systems where they can slow models and increase power use.

This 90-minute hands-on tutorial explores how numbers are represented in memory, covering precision, rounding, overflow, and how quantization trades accuracy for speed and efficiency. Using Rust, we’ll build a small quantization framework step by step, from fundamentals to Post-Training Quantization (PTQ) for AI inference.

Focusing on AI, we’ll show how these techniques improve inference performance, reduce model size, and make systems more efficient - while Rust’s safety and explicitness help prevent numeric corrosion in production code.

Agenda (approx)
18:00 - 🍕 Arrivals, networking, pizza/drinks
18:30 - 👋 Event starts, introductions
18:35 - 🗨️ Talk starts
20:00 - 🍹Event finishes, head to a nearby pub (optional)