G42, an Abu Dhabi-based AI technology company backed by Microsoft, has introduced "Nanda," a large language model (LLM) specifically designed for Hindi speakers in India. The unveiling took place at the UAE-India Business Forum in Mumbai, attended by dignitaries from both countries.
Nanda: A Bilingual LLM for India
Nanda, named after India's second-highest peak, Nanda Devi, is a 13-billion-parameter LLM trained on a massive dataset of 2.13 trillion tokens, encompassing Hindi, English, and Hinglish languages.
"We genuinely believe that #Nanda can get integrated into the fabric of India," said Manu Kumar Jain, CEO of G42 India, in a post on X. "It meets all the sovereign requirements and can help take large-scale technology initiatives to newer heights. It will offer over half a billion Hindi language speakers the opportunity to harness the potential of generative AI.”
Jais: G42's Previous Arabic LLM
The launch of Nanda follows G42's release of Jais last year, an open-source Arabic LLM designed for Arabic natural language processing (NLP). Jais was released with multiple model sizes, including one with 70 billion parameters.
Microsoft's Investment in G42
In April 2024, Microsoft made a significant investment of $1.5 billion in G42, becoming a major backer and securing a seat on the company's board. This investment highlights Microsoft's commitment to advancing AI capabilities and expanding its reach in emerging markets.
Use Cases and Availability
G42 has not yet announced specific use cases for Nanda or its public availability. It remains to be seen whether the LLM will be made accessible to the general public or reserved for government and enterprise applications.
Nanda's development represents a significant step towards making AI technology more accessible to Hindi speakers in India, opening up new possibilities for innovation and digital inclusion.