WINDWAY DATADocumentation développeur

WINDWAY DATA / DOCS / v1.0.0

Jeux de données

Deux échantillons publics approuvés de 100 enregistrements et cinq familles chacun.

Référence technique · anglais

Release contents

Each archive contains data/train.jsonl, data/validation.jsonl, data/test.jsonl, SCHEMA.json, README.md, LICENSE, SOURCE_LICENSE.json, PROVENANCE.md, VERIFICATION.md, MANIFEST.json, CHECKSUMS.sha256, verification_results.json and shipped Python validator files. Each sample has five task families with 20 records each.

Hugging Face configuration

The mirror repository names refer to the historical 25K corpus. The sample is specifically the windway_sample_v1 configuration, under windway_sample/v1.0.0/. Its 100 records are not the default full-corpus configuration. Splits are grouped; inspect group_id when designing leakage controls. Preserve the supplied splits.

Release evidence

Rechercher dans la documentation