The dataset is currently empty. Upload or create new data files. Then, you will be able to explore them in the Dataset Viewer.
Leibniz Result Publication Checkout
This dataset stores the published, program-neutral Leibniz result records. The three tiers below are the source of truth; they reference one another by id and content digest rather than embedding each other.
submissions/: program-neutral submission records. The submission is the program graph; a learned submission additionally carries fitted parameters and a training-provenance record, while a deterministic submission carries neither. Submissions are benchmark-agnostic — the same submission is evaluated against many benchmarks.benchmarks/: descriptive benchmark (universe) metadata records.evaluations/: evaluation records, one per (submission, benchmark) pair — the validated-bit Score, the capability-map data backing it, the measured energy cost, optional certification diagnostics, and lineage back to the submission, the benchmark, and the measurement evidence.
Local runtime state produced by the benchmark workflow — training/ (in-progress
estimates), models/ (checkpoint artifacts), views/ (generated console views),
and programs/ (queued program sources) — is not part of the published record
set.
- Downloads last month
- 1