Metrum Insights Documentation
Metrum Insights helps teams run reproducible AI system benchmarks across models, modalities, serving frameworks, hardware, and deployment environments.
Start here
- Quickstart: get your first benchmark running in under 15 minutes.
- User Guide: every core workflow, end to end.
- Feature Reference: per-feature configuration lookup.
- Admin Guide: identity, RBAC, billing, backup, GDPR, cloud provisioning.
- API Reference: REST and RPC surface for programmatic access.
- Hardware Compatibility: supported accelerators per workload type.
- Sizer Usage Guide: pick the right hardware before you deploy.
- Release Notes - v4.0: what's new, what's changed, and what's coming.
The benchmark workflow
- Define what you want to learn: model, modality, precision, framework, input/output shape, concurrency, hardware, and success metrics.
- Model the benchmark in Metrum Insights: project, workloads, scenarios, hardware instances, command templates, and run settings.
- Onboard hardware or provision cloud capacity.
- Execute the benchmark and collect job status, outputs, telemetry, logs, SUT data, and cost evidence.
- Package the results into reports that can be compared and reproduced.