Use the LocalMaxxing CLI (the `lmx` binary) to benchmark local and remote LLM inference (tokens/sec, TTFT, peak VRAM), run quality evals (custom suites, lm-eval harness, LM-judge, eval shards, Terminal-Bench), generate hardware metadata, run KV-cache context sweeps, and submit results to localmaxxing.com. Covers authentication, choosing local vs remote benchmark mode, controlling the benchmark prompt for spec-decode-representative numbers, dry-run validation, saved runs and profiles, and reading live enums via `lmx context`.
2026-07-23