Datasets:

---

nvidia

/

aisimulate-fpm-dataset

Like 1

Follow

NVIDIA 68.9k

Moves the already-public calibration data behind ai-dynamo/aisimulate#158 out of its historical GitHub experiment archive into this dataset. This is a storage migration, not recollection or new accuracy qualification.

• Preserve all seven unique schema-7 Parquet payloads, with ten original source aliases recorded; GB300 TP4 full/bounded principal tables contain 142 rows each, and GB200 TP4 contains 126.
• Keep the GB300 tables as historical development evidence (including the original backend labels). GB200 remains quarantined and cannot become a serving baseline through migration.
• Preserve observed-only GB300 native intervals and GB200 holdout inputs, provenance, source hashes and licenses. Unsupported observation protocols remain explicitly unsupported; predictions and accuracy reports are not measurement truth.
• Add explicit schema-7 import/validation without relabeling data as schema 6.
• Source revision: dd1fa97add17d3d74f580f4c3e0566c1b5f11827. Consumer pins will use this HF PR's full commit SHA and verify file hashes.

Companion contribution: ai-dynamo/aisimulate#232 contains newer B300 TP2/TP4 and GB300 TP2 profiles. Both contributions start from the same dataset baseline; catalogs must be refreshed on the second contribution after the first merges. Please do not merge conflicting catalogs blindly.

Request changes — reviewed commit 757ff1f99f716a1d21985723ae534cef4afbb5ab.

• [P2] Preserve the producer revision separately from the archive revision. The importer sets aisim_commit to the archive commit and marks it known. The dataset contract defines this field as the producer revision. Both principal GB300 tables record original aic_revision=40c9541bb0bb593f403345efde01eafba436543c, but the migration publishes dd1fa97add17d3d74f580f4c3e0566c1b5f11827. The other component tables have different producer revisions, and GB200 records an installed-package identity rather than a Git SHA. Preserve these original identities in provenance, derive the snapshot producer status from them, and use null with unknown/mixed status where a single producer commit is unavailable. Keep the archive commit in source_revision. Regenerate the affected manifests/catalogs and add a regression check distinguishing producer and archive revisions.

Validation passed: full dataset validation, 48 tests, byte-for-byte comparison of all seven Parquet payloads against the pinned GitHub source, and all 15 file references in the AISimulate consumer manifest. GB200 remains quarantined. Native prediction and GPU accuracy were not rerun.

After the fix, merge this PR first, then regenerate #12's catalogs against merged main. The two heads conflict in five catalog files. Refresh affected consumer pins and hashes after metadata changes.

Addressed the requested changes in b35883ee5f8b4a82e24844a7e872056aff64287f.

• Producer identity is now derived from the recorded aic_revision, separately from the archive source_revision. Original Git and installed-package identities are retained; snapshots with multiple producers use null/mixed, and missing or non-Git producer evidence uses null/unknown. The importer no longer substitutes the archive commit.
• Regenerated manifests and catalogs; added regression tests for producer/archive separation, collection provenance preservation and rejecting inconsistent identities. Full dataset validation and all 52 tests pass. All 7 calibration Parquet payloads remain byte-identical. Remote readback verified SHA-256 for all 46 contributed files at this revision.

AISimulate consumer pins and metadata hashes are being refreshed to this revision. These checks establish migration integrity, not new GPU collection or independent accuracy qualification. Merge #11 first; after that merge, I will refresh #12 against the actual merged main and validate the combined catalogs before requesting its merge.

Re-review passed at b35883ee5f8b4a82e24844a7e872056aff64287f. The producer/archive identity finding is resolved: original producer identities are preserved, and snapshot status is unknown/mixed where appropriate. Independently verified all seven Parquet payloads against the pinned GitHub source; full dataset validation and all 52 tests pass. Merging #11 first. #12 must retain this import when its catalogs are refreshed against merged main. These checks establish migration integrity, not new GPU accuracy qualification.