Stop treating language models as black boxes. Join us to build your own LLM from the ground up—understanding every layer, every attention head, and every training step that brings an AI to life.
What You'll Learn:
- 🏗️ Transformer architecture: decoder blocks, attention mechanisms, and the building blocks of modern LLMs
- 🔄 From tokens to text: embeddings, RoPE (Rotary Positional Embeddings), and how models understand language
- 🎯 Training your own Gemma: implementing a complete training pipeline with PyTorch
- ⚡ Modern training practices: mixed precision, gradient accumulation, learning rate scheduling
- 🛠️ Beyond APIs: why building from scratch teaches you what no API documentation can
Format - Talk + Demo:
- 📊 Architectural deep-dive with visual explanations
- 🧪 Training demonstrations and inference examples
- 📦 Open-source stack you can take home and experiment with
Who Should Attend: Developers and ML practitioners who want to understand what's really happening inside language models. Whether you're tired of treating LLMs as magic or ready to customise models for specific needs, you'll leave with a complete understanding of the foundations.
Prerequisites: Python and basic ML concepts. PyTorch familiarity is helpful but we'll cover the essentials.
Plus, once the talk is over you can stay a bit longer for a networking aperitivo where you can chat, meet new people, and just have a good time. ✌️
