Updates KPIs and the latency-components chart; multi-run comparison charts stay unchanged.
Requests / sec
__GUIDELLM_KPI_RPS__
Output tokens / sec
__GUIDELLM_KPI_OUT_TPS__
Total tokens / sec
__GUIDELLM_KPI_TOTAL_TPS__
Request latency
P95
__GUIDELLM_KPI_LAT_P95__
ms
P99
__GUIDELLM_KPI_LAT_P99__
ms
TTFT
P95
__GUIDELLM_KPI_TTFT_P95__
ms
P99
__GUIDELLM_KPI_TTFT_P99__
ms
ITL
P95
__GUIDELLM_KPI_ITL_P95__
ms
P99
__GUIDELLM_KPI_ITL_P99__
ms
Error rate
__GUIDELLM_KPI_ERROR__
Run comparison
__GUIDELLM_STATIC_TABLE__
Charts, tabs, and the interactive details below require JavaScript.
Request latency by benchmark
End-to-end P95/P99 as load increases — lower is better.
Request outcomes
Share of successful, incomplete, and errored requests.
Latency components
P95 and P99 latency components for the selected run.
Run comparison
Filter parent vs subagent requests in turn curves.
Latency & prompt size by turn
Latency on the left axis; prompt median tokens on the right (dashed).
Per-turn table
Request size
WebSocket round-trip