Aitra Meter v0.3.0

Aitra Meter v0.3.0

Welcome to Aitra Meter new release!. This is latest release for Aitra Meter based on main branch

Key Features

• Core measurement agent (per-node DaemonSet)
• Energy providers : NVML (default), AMD , Zeus , DCGM
• Inference providers: vLLM , generic-prometheus
• Aggregation service
• Metrics based on Prometheus exposition
• Storage: pluggable backend interfaces such as SQLite
• Alerting
• Integrations: OpenTelemetry / OTLP, OpenCost, KEDA

For more details on features, refer here

Release Images

Component	Image Tag
Measurement Agent	aitra-meter/measurement-agent:v0.3.0
Aggregation Service	aitra-meter/aggregation-service:v0.3.0

What's Changed

• fix(ci): make main build, test, and lint green by @stevenphtan in #62
• feat(energy): add DCGM EnergyProvider reading dcgm-exporter by @stevenphtan in #59
• feat(metrics): AI efficiency metric family — model token/energy primitives (part of #40) by @stevenphtan in #60
• feat(alerting): cost-budget + TTFT alerts, runbooks, helm cost budgets (closes #46) by @stevenphtan in #61
• build(deps): bump actions/checkout from 6 to 7 in the all-actions group by @dependabot[bot] in #57
• fix(release): add missing Dockerfiles, unpin image tags, bump chart to 0.8.0 by @stevenphtan in #74
• chore(codeowners): make @skdwriting the default reviewer by @stevenphtan in #72
• docs(positioning): reposition README as Prometheus exporter for AI inference efficiency (closes #47) by @xiaoxixixingxing in #66
• docs(proposals): draft OTel GenAI semconv energy proposal (part of #38) by @xiaoxixixingxing in #65
• feat(opencost): cost-per-token recording rule, derived panel column, guide expansion (closes #44) by @xiaoxixixingxing in #64
• feat(grafana): audience-specific dashboards with Helm auto-provisioning (closes #42) by @xiaoxixixingxing in #68
• chore(helm): make Grafana the default surface, standalone dashboard opt-in by @stevenphtan in #81
• docs(adr): ADR-0007 — per-GPU energy attribution (sum only for TP>1) by @stevenphtan in #90
• feat(dashboard): Vercel deployment behind an authenticated tunnel by @xiaoxixixingxing in #73
• build(deps): bump the all-go group across 1 directory with 12 updates by @dependabot[bot] in #63
• build(deps): bump golang from 1.25-bookworm to 1.26-bookworm in /build/measurement-agent by @dependabot[bot] in #85
• build(deps): bump golang from 1.25-bookworm to 1.26-bookworm in /build/aggregation-service by @dependabot[bot] in #86
• build(deps): bump actions/setup-go from 6 to 7 in the all-actions group by @dependabot[bot] in #87
• fix(generic-prometheus): sum multi-series metrics; unblocks SGLang by @stevenphtan in #77
• fix(k8s): separate the GPU scheduling label from the hardware tier label by @stevenphtan in #80
• feat: per-GPU/per-model energy attribution + KubeCon demo kit (closes #89) by @xiaoxixixingxing in #91
• docs: add FEATURES.md cumulative feature inventory by @stevenphtan in #93
• Update README with build and test instructions and demo information by @sushanthakumar in #94

New Contributors

• @sushanthakumar made their first contribution in #94

Full Changelog: v0.2.2...v0.3.0

What's Changed

• fix(ci): make main build, test, and lint green by @stevenphtan in #62
• feat(energy): add DCGM EnergyProvider reading dcgm-exporter by @stevenphtan in #59
• feat(metrics): AI efficiency metric family — model token/energy primitives (part of #40) by @stevenphtan in #60
• feat(alerting): cost-budget + TTFT alerts, runbooks, helm cost budgets (closes #46) by @stevenphtan in #61
• build(deps): bump actions/checkout from 6 to 7 in the all-actions group by @dependabot[bot] in #57
• fix(release): add missing Dockerfiles, unpin image tags, bump chart to 0.8.0 by @stevenphtan in #74
• chore(codeowners): make @skdwriting the default reviewer by @stevenphtan in #72
• docs(positioning): reposition README as Prometheus exporter for AI inference efficiency (closes #47) by @xiaoxixixingxing in #66
• docs(proposals): draft OTel GenAI semconv energy proposal (part of #38) by @xiaoxixixingxing in #65
• feat(opencost): cost-per-token recording rule, derived panel column, guide expansion (closes #44) by @xiaoxixixingxing in #64
• feat(grafana): audience-specific dashboards with Helm auto-provisioning (closes #42) by @xiaoxixixingxing in #68
• chore(helm): make Grafana the default surface, standalone dashboard opt-in by @stevenphtan in #81
• docs(adr): ADR-0007 — per-GPU energy attribution (sum only for TP>1) by @stevenphtan in #90
• feat(dashboard): Vercel deployment behind an authenticated tunnel by @xiaoxixixingxing in #73
• build(deps): bump the all-go group across 1 directory with 12 updates by @dependabot[bot] in #63
• build(deps): bump golang from 1.25-bookworm to 1.26-bookworm in /build/measurement-agent by @dependabot[bot] in #85
• build(deps): bump golang from 1.25-bookworm to 1.26-bookworm in /build/aggregation-service by @dependabot[bot] in #86
• build(deps): bump actions/setup-go from 6 to 7 in the all-actions group by @dependabot[bot] in #87
• fix(generic-prometheus): sum multi-series metrics; unblocks SGLang by @stevenphtan in #77
• fix(k8s): separate the GPU scheduling label from the hardware tier label by @stevenphtan in #80
• feat: per-GPU/per-model energy attribution + KubeCon demo kit (closes #89) by @xiaoxixixingxing in #91
• docs: add FEATURES.md cumulative feature inventory by @stevenphtan in #93
• Update README with build and test instructions and demo information by @sushanthakumar in #94

New Contributors

• @sushanthakumar made their first contribution in #94

Full Changelog: v0.2.2...v0.3.0