Skip to content

Latest commit

 

History

History
65 lines (52 loc) · 2.85 KB

File metadata and controls

65 lines (52 loc) · 2.85 KB

Chart values reference

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)
gatewayImage bifrost the proxy backend (bifrost / litellm / portkey)
gatewayTag latest --model-tag
runnerTag latest --agent-tag / --benchmark-tag
benchmarkVersion "" --benchmark-version
agentVersion "" --agent-version
litellmVersion "" --litellm-version
maxBudget "" --max-budget
reasoningEffort "" --agent-reasoning-effort

Knobs a benchmark may override

Field Default
timeout "300"
deadlineGrace 600
activeDeadlineSeconds "" (derives from timeout + deadlineGrace)
runnerArgs /entrypoint.sh; rc=$?; /usr/local/bin/reap-sidecars; exit $rc
resources.requests { cpu: 500m, memory: 512Mi }
resources.limits { cpu: 2, memory: 2Gi }
outputVolume {} (→ 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.

Field Default Example use
initContainers [] a wait-for-<service> gate
runnerExtraEnv [] DESKTOP_URL for osworld
gatewayExtraEnv []
runnerExtraVolumeMounts []
extraVolumes []
extraManifests [] extra Deployments/Services (full manifests)