AI4Bharat

Indian-language models and research resources

AI4Bharat is an initiative of IIT Madras that develops language datasets, models and tools. The Nilekani Centre inaugurated in July 2022 added an institutional base for Indian-language technology research and training. It was a new centre within the initiative, not the first existence of all AI4Bharat work. [1]

IndicTrans2 illustrates its model contribution. The team released English-to-Indic and Indic-to-English models covering all 22 scheduled Indian languages in May 2023, then described direct Indic-to-Indic and smaller distilled variants in December. These are downloadable translation systems, rather than a claim that every Indian language is covered. [2]

The December model announcement states MIT licensing and links the released models. Its December 3 publication and the repository's December 1 release entry are different records; neither is substituted for an unverified exact day of the original May release. [2]