Nvidia rolls out Hindi-language AI model in India: What is it and how will it work?

Nvidia, a global leader in chip technology, has introduced a new lightweight artificial intelligence (AI) model designed for India's widely spoken Hindi language. The model, named Nemotron-4-Mini-Hindi-4B, has 4 billion parameters and is aimed at helping Indian firms create their own AI solutions. This move highlights Nvidia’s focus on tapping into India's growing demand for AI technologies.

Nvidia explained that the AI model was developed using both real and synthetic Hindi data, along with a balanced amount of English data. 

Tech Mahindra is the first to use the model to create Indus 2.0, an AI system focusing on Hindi and its various dialects.

In India, only about 10% of the 1.4 billion population speaks English, with 22 languages officially recognised in the country. Companies, from large corporations to startups, are increasingly building AI models based on India's linguistic diversity to better connect with consumers. These models are useful for services like AI-driven customer support and content translation.

Unlike large language models like OpenAI's GPT-4, smaller models like Nemotron-4-Mini-Hindi-4B are cheaper to develop and use, making them more accessible for businesses with limited resources.

Nvidia has been active in India for nearly two decades, with offices and engineering centres in key cities like Bengaluru and Hyderabad, as global chipmakers invest in India’s expanding semiconductor industry.