| # Swahili-Safi Dataset | |
| A relatively clean dataset for Swahili language modeling, built by combining and cleannig several existing datasets. | |
| Sources include: | |
| ``` | |
| mc4-sw | |
| oscar-sw | |
| swahili_news | |
| IWSLT | |
| XNLI | |
| flores 101 | |
| swahili-lm | |
| gamayun-swahili-minikit | |
| broadcastnews-sw | |
| ``` | |
| In total this dataset is ~3.4 GB, with over 21 million lines. | |