Hanooman joins the Indic AI race alongside Ola’s Krutrim, SaravamAI’s OpenHathi, and IIT-Madras’s Airavata model.
Seetha Mahalaxmi Healthcare (SML), in collaboration with the IIT Bombay-led BharatGPT ecosystem, has announced the launch of ‘Hanooman’ a series of Indic large language models trained on 22 Indian languages. Vishnu Vardhan, the founder of SML, revealed that they have developed an enterprise-ready family of foundational language models ranging up to 40 billion parameters in size.
The first four models in the series, with 1.5 billion, 7 billion, 13 billion, and 40 billion parameters, will be released next month and will be open-sourced. Currently, the models can respond in 11 Indian languages, including Hindi, Tamil, Telugu, Malayalam, and Marathi, with aspirations to extend their capabilities to all 22 Indian languages.
Hanooman is set to have multimodal AI capabilities for generating text-to-text, text-to-speech, text-to-video, and vice versa content. SML is already in discussions with BFSI enterprises, healthcare organizations, and mobile app providers to offer model-as-a-service or create specialized models by fine-tuning the Hanooman series. The first among these fine-tuned versions is the healthcare model VizzhyGPT, trained on large data of multiomics, clinical research, and EMR data.
Vardhan highlighted that one of the major challenges in building Indian LLMs is sourcing quality datasets in Indian languages. The first hurdle is to improve the quality of these datasets, whether in text, audio, or video format, as organisations currently use synthetic datasets created by translating from other languages, which may increase the chances of inaccuracies or hallucination.
The announcement was made on the sidelines of the Nasscom Technology and Leadership Forum in Mumbai. The BharatGPT ecosystem is a research consortium led by IIT Bombay, involving seven other IITs, and is supported by the Department of Science and Technology, SML, and Reliance Jio. Professor Ganesh Ramakrishnan of IIT-B mentioned that their engagement with Reliance Jio involves industry-specific downstream applications in areas such as telecom and retail, requiring the building of smaller customised ML models.
