Datasets:

---

nvidia

/

aisim-data

Like 1

Follow

NVIDIA 68.6k

Summary

• Add independent op-latency, e2e, and fpm YAML contracts.
• Add strict validation, generated catalog checks, fixtures, and CI.
• Import one pinned, unofficial InferenceX E2E point with normalized Parquet data, source artifacts, effective configuration, overrides, compressed server logs, attribution, and reconciliation.
• Record the exact AISimulate Replay gap as unsupported rather than substituting an analytical estimate.

Validation

• python scripts/validate.py
• python scripts/build_catalog.py --check
• pytest (4 passed)

Evidence limitations

• The source artifact has no request-level table, client log, immutable image digest, pinned model revision, or request error accounting; each is explicitly marked unavailable.
• The source workflow failed overall, while the imported operating point completed successfully.