9 lines (7 loc) · 388 Bytes

Changelog

0.2.0

Breaking: datasets are now parquet-only for ingestion and core dataset exports.
Removed JSONL-based ingestion readers and legacy auto-detection paths.
detect-language path loading now accepts parquet files/directories only.
dedup --output-deduped now accepts .parquet only.
Evaluation bundles, CSV/JSON reports, and pair artifacts remain supported.