Received 67 proposals to build AI models trained on Indian datasets: MoS IT

The government has received proposals from 67 companies till February 15 to build state-of-the-art foundational AI models trained on Indian datasets, the ministry of electronics and information technology informed the Lok Sabha on Wednesday.

 

Of the 67 proposals, 22 are focused on large language models and large multimodal models, while the other 45 are proposals to develop domain-specific small language models, the minister of state for electronics and information technology, Jitin Prasada, told the Lok Sabha in his written response to a question.

 

“The majority of small language models (SLMs) target key sectors such as healthcare, education, and financial services. Along with funding support, a wide range of graphics processing units (GPUs) have been requested by teams submitting these proposals,” Prasada said in his reply.

 

 

To support companies that have proposed to build these foundational models on India-specific datasets, the government launched the India AI datasets platform, AIKosh, on March 6. It is designed to act as a unified data platform, integrating datasets from all existing data platforms, onboarding non-government data contributors, and providing AI-centric features.

 

At present, the AIKosh platform offers a vast collection of 300 datasets, sourced from 12 different sectors, and also includes more than 150 AI models provided by five different entities, Prasada said on Wednesday in response to another question.

 

“Additionally, the platform provides over 10 toolkits comprising development utilities for project integration, as well as tutorials to help users understand and navigate its features,” he said.