| configs: | |
| - config_name: default | |
| data_files: 'data/*.parquet' | |
| # Beetle-Data/ca-100M-pretok | |
| Pretokenized chunks (`input_ids` = 513-token packed sequences, | |
| no cross-document bleeding). Sharded incrementally; the marker | |
| `data/_finalized.json` is committed once all parts are uploaded. | |