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.