AISim - E2E Accuracy Overview

Loading evidence…

Exclude:
Loading summary…

MAPE vs silicon — AISim over successful replay points; AIC (legacy CLI) over successful baseline predictions

End-to-end accuracy metrics grouped by model and workload. Activate a column heading to sort model groups. Expand a workload, then select a GPU for details.
AISim MAPE
( / ) 		AIC (legacy CLI) MAPE
( / )
Loading accuracy matrix…
About this comparison · methodology & provenance

Compare AISim with AIC (legacy CLI). We show both CLI results against measured silicon to help users build confidence that AISim is comparable to AIC (legacy CLI) during migration.

AIC (legacy CLI) results are temporary: we will remove this comparison series when we deprecate AIC (legacy CLI). Check successful-point counts alongside errors; the two CLIs can have different prediction coverage.

Loading branch evidence…

Loading snapshot dimensions…

Read the scope first

These numbers apply only to exact measured configurations in this snapshot.

  • Client-observed TTFT and TPOT are reported separately.
  • Multi-node configurations are included when available; use the exclusion filter to hide them.
  • No universal pass/fail threshold is applied.

Reproducible snapshot identity

Loading provenance…
Download JSON