Model optimization, a tree hiding a forest?

08/02/2024 | 13h30 - 14h00 | Demo Stage 1

Information

Model Optimization: a tree hiding a forest?" is a presentation that explores the optimization of AI models, focusing on data selection, training, and production considerations. The presentation covers various optimization methods, their impact on model size, quality, and latency, and the importance of defining latency and throughput requirements. Real-life examples of latency-optimized and throughput-optimized systems are provided, along with a solution for concurrent usage. The presentation concludes with a summary of optimization factors and potential tools for optimized inference.