Benchmark an LLM server under open-loop load and report goodput

Sweep request rates with realistic prompt and output lengths, report TTFT, inter-token and end-to-end latency percentiles and goodput under a stated SLO, and locate the knee.

The exercise happens on Self-directed (vLLM or another serving engine). Canonical link: https://docs.vllm.ai/en/latest/cli/bench/serve/. Link check: link responded (5 Oct 2026). Opened from this app 0 times; opening is not counted as an attempt.

Supports: Needs a GPU that serves a small open model. Prepares benchmark-design and serving-metrics discussions.