Daijiworld Media Network – Bengaluru

Bengaluru, Apr 27: Bengaluru-based Sarvam AI has become the first company to receive government support to build an indigenous large language model (LLM), Union IT Minister Ashwini Vaishnaw announced on Saturday.

Speaking at the launch of the guidelines for the Electronics Component Manufacturing Scheme (ECMS), Mr. Vaishnaw said more firms will soon be selected to work on sovereign LLM projects. The government is keen to develop AI models suited to India’s cultural and linguistic diversity, addressing concerns about the western bias seen in existing foundational models.

Founded in mid-2023, Sarvam AI will have access to 400 GPUs for six months from the pool of over 14,000 GPUs acquired under the IndiaAI Mission. GPUs are critical for developing and deploying AI models, which require intense computational resources to process billions of parameters and trillions of tokens.

In a statement, Sarvam AI said its model would be "fluent in Indian languages, designed for voice interactions, capable of reasoning, and ready for secure, population-scale deployment."

Highlighting progress on other fronts, Vaishnaw noted that 350 datasets have been uploaded to AI Kosh, a government portal aimed at supporting AI development in India. He added that tools and technologies requested from the IITs and other institutes under the India AI Safety Institute initiative are nearing completion.

Encouraging domestic electronics manufacturers, Vaishnaw also stressed the importance of setting up in-house design teams and achieving "six sigma" quality standards. Under the ECMS, companies manufacturing passive components such as printed circuit boards and sub-assemblies will be eligible for investment- and revenue-linked incentives.