| # Data | |
| Place datasets here by processing stage. | |
| - `raw/`: original downloaded or manually collected data. | |
| - `cleaned/`: normalized Malayalam text. | |
| - `deduplicated/`: cleaned text after exact and fuzzy deduplication. | |
| - `tokenized/`: model-ready tokenized datasets. | |
| Do not commit large datasets directly unless they are intentionally small samples. | |