energy monitoring

A practical guide to measuring and reducing AI energy on Linux with Bash: install powertop, turbostat, nvtop, numactl, gnuplot; baseline with powertop/turbostat; log per-job energy using a ~100-line ai-energy-run.sh (CPU via RAPL, optional GPU watts); optimize with thread tuning, NUMA pinning, power caps, batch/precision tweaks; and enforce energy budgets in CI using a guard script that fails over-target runs.