notadestroyer/vLLM Benchmark

Run vLLM performance benchmarks and post results as PR comments

View on GitHub

Trust Signals

Scorecard Score
not yet scored
Maintenance Recency
Stale
License
None
namedescriptionrequireddefault
vllm-urlvLLM server URLnohttp://localhost:8000
presetBenchmark preset: quick, standard, or thoroughnoquick
context-lengthsComma-separated context lengths (overrides preset)no
concurrencyComma-separated concurrency levels (overrides preset)no
output-tokensMax output tokens per requestno500
prompt-typePrompt type: classic, deterministic, madlib, random, allnoclassic
compare-withPath to previous benchmark JSON for regression detectionno
fail-on-regressionFail the workflow if major regressions detectednofalse
python-versionPython version to useno3.12
namedescription
scoreComposite vLLM benchmark score (0-10000)
gradeBenchmark grade (S/A/B/C/D/F)
peak-throughputPeak throughput in tokens/sec
results-jsonPath to results JSON file