Why it matters

LLMPerf targets LLM serving perf. Understanding shapes benchmark choice.

Advertisement

The architecture

Concurrent clients + payload.

Measures TTFT, ITL, throughput.

Compare providers + engines.

LLMPerf benchmarkConcurrent clientsloadMeasureTTFT + ITL + throughputCompareproviders / enginesAnyscale + open source; useful for provider comparison
LLMPerf.
Advertisement

How it works end to end

github.com/ray-project/llmperf.

OpenAI-compatible API.

Provider + local + vLLM benchmarks.