Hyderabad: Soon, one will be able to search for Telugu culture, literature, regions or even get any text translated into native dialects as the first Telugu Large Language Model (LLM) is all set to be launched in Aug.
The International Institute of Information Technology, Hyderabad (IIITH), and Swecha, which have launched Viswam.ai to create AI solutions for the Global South, are planning to release the basic version at the ‘AI Days’ conference in April and the first full version in Aug.
“Unlike the English language, Telugu, or for that matter any regional language, is digitally starved. We managed to create a large dataset for the language. We digitised 8 crore pages of books and 4,000 hours of speech data, which include various accents of Telugu. Using this, we are planning to launch the first Telugu LLM in Aug,” said Ramesh Loganathan, professor of Practice, Co-Innovations, at IIITH.
He said that using the model one can look for any data pertaining to Telugu or food, art, temples, forts and occupations in Telugu states.
“It, however, will not be able provide information like OpenAI because of language constraints,” he added.
On the second day of the ‘AI Days’ conference, the team is planning to organise a hackathon and launch the basic version of the Telugu LLM. After the conference, it is planning to get one lakh interns on board to collect data in the form of interviews to create much stronger datasets so that the first Telugu LLM can be launched by Aug.
Y Kiran Chandra, centre head, Viswam.ai, said the model will be completely open so that anyone can use it or even make changes to the source code to create their own models. “The foundation was laid for this when we first started digitising Chandamama Kathalu in 2023. Now, we have rich datasets available, which makes it possible to launch the first Telugu AI,” he added.
