Optimise LLM Inference Throughput from First Principles (Part I)
ML Systems