Reference¶
Tables to look things up in. For step-by-step instructions, see the Submission Workflow instead.
Artifacts you author¶
| Page | Covers |
|---|---|
| Submission package layout | The run folder, the assembled bundle, and what is shared vs per-point |
point.yaml |
Per-measurement-point disclosure fields |
system_desc.json |
Hardware and software description fields, with a template |
Measurement¶
| Page | Covers |
|---|---|
| Metrics and regions | Metric definitions, the region algorithm, pre-computed boundary tables |
Tools¶
| Page | Covers |
|---|---|
| Benchmark runner CLI | inference-endpoint — runs the benchmark |
| Submission CLI | endpoints-submission-cli — registers runs, creates submissions |
| Submission checker | submission-checker — validates a bundle |
Process¶
| Page | Covers |
|---|---|
| Compliance checks | Every check, cross-walked from checker rule ID to rules clause |
| Submission states | Status values and what moves between them |
| Glossary | Terms and acronyms |
These pages are hand-written
Each CLI page names the upstream document it tracks. Generating them from --help output is the
right end state; until then, the upstream repository is authoritative and these pages can lag.
Last verified against: mlcommons/endpoints_policies@v1.0_rules_dev (a7ec3cc), 2026-09-19.