Skip to content

vllm bench sweep

Overview

Benchmark for a parameter sweep.

Subcommands

Command Description
serve Run vLLM server benchmark under multiple settings.
serve_workload Explore the latency-throughput tradeoff for different workload levels.
startup Benchmark vLLM startup time over parameter combinations.
plot Plot performance curves from parameter sweep results.
plot_pareto Plot Pareto frontier between tokens/s/user and tokens/s/GPU from parameter sweep results.