model metrics

A practical, Linux-first guide to turn AI performance from guesswork to measurable gains: install sysstat/htop/nvtop/glances/hyperfine/numactl/Netdata, pick KPIs, baseline with one-liners, automate sampling via a tiny Bash logger, visualize trends, and fix bottlenecks (CPU/I/O/data pipeline) by tuning batch size, NUMA, power, and concurrency—then verify throughput/latency improvements with real-world scripts and examples.