You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Reference · for operators · derives from containers/benchmarks/_chart/values.yaml. That file is authoritative — these are its fields with defaults at the time of writing.
The shared chart containers/benchmarks/_chart renders the otelcol + gateway + runner Job.
The benchmark is named via --set benchmark=<x> (required); a benchmark with
bespoke topology overrides what differs through an optional presets/<x>.yaml
in the chart. Per-run axes arrive via --set (or the CLI). See
The Helm chart.
Required
Field
Meaning
benchmark
Benchmark name, set via --set benchmark=<x>. Selects the optional presets/<x>.yaml overlay.
Per-run axes — set at deploy via --set / the CLI
Field
Default
CLI flag
agent
claude-code
--agent
task
"0"
--task-id
registry
ghcr.io/exgentic
--registry
model
— (required)
--model — the <provider>/<model> handle the gateway routes (the runner's clean label is derived from it)
{} (→ ephemeral emptyDir); set a source to persist /output, e.g. --set outputVolume.hostPath.path=/eval-output or a persistentVolumeClaim
Platform composition — layer via --overlay (extra -f)
Field
Default
Notes
serviceAccountName
""
OpenShift sets anyuid-sa
sweepId
""
sweep bookkeeping
Composition hooks — for benchmarks with bespoke topology
Standard benchmarks leave these empty; the bespoke few (osworld, tau-bench,
visualwebarena, webarena) use them to add sidecars, Deployments, and Services.