In its mission to construct open supply LLMs for Indian researchers and builders, BharatGen, the federal government backed AI initiative, has launched a 2.9 billion parameter bilingual LLM, referred to as Param 1.
The newly launched LLM, dubbed ‘BharatGen Param 1 Indic Scale’, is a pre-trained base mannequin constructed fully from scratch and contains a staggering 25% Indic information—a stark distinction to the mere 0.01% Indic information sometimes utilized in fashions like Meta’s Llama.
You’ll be able to take a look at the mannequin on AIKosha.
“Pre-training is a gigantic enterprise and sometimes an insurmountable barrier for a lot of. That’s why we’ve taken on this problem—to supply a sturdy basis that you could simply fine-tune on your particular functions,” BharatGen mentioned in a press release.
Builders can now fine-tune the mannequin by way of AIKosha to construct various functions starting from Indic chatbots to India-specific copilots and data methods. “With our 2.9 billion parameter base mannequin, we’re unlocking new prospects for innovation and progress throughout the nation. We hope this sovereign LLM mannequin checkpoint serves as a basis for India-specific options, enabling builders to fine-tune and form the subsequent era of AI functions for Bharat,” Prof Ganesh Ramakrishnan, head of BharatGen, advised AIM.
Alongside the LLM, the group additionally launched 20 new speech fashions (throughout 19 Indian language variations)—concentrating on voice-first interfaces and speech-based innovation for Indian customers on AIKosha—the AI innovation repository by MeitY, Authorities of India.
This contains 9 fashions below A2TTS-v0.5: Speaker Adaptive TTS. These enable builders to generate speech that matches a supplied speaker’s voice, out there in Marathi, Bengali, Hindi, Gujarati, Tamil, Kannada, Punjabi, Telugu, and Malayalam.
There are 5 fashions below Speaker-Conditioned TTS (pflow) which provide high-fidelity text-to-speech fashions for Marathi, Tamil, Hindi, Telugu, and Bengali.
Then there are different 5 fashions below Voicebox TTS Fashions, that are adaptable voice synthesis for Hindi, Tamil, Marathi, Telugu, and Bengali.
BharatGen says that these fashions have been constructed from the bottom up with information collected straight for 5 Indian languages, addressing a significant hole in high-quality, publicly out there speech fashions for Indic languages.
AIKosha, launched by Union Minister Ashwini Vaishnaw, is India’s official AI repository and the brand new residence for these fashions. The repository goals to centralise India’s AI property and gas collaborative innovation.
“These foundational fashions are engineered to supercharge India’s AI analysis and innovation ecosystem,” BharatGen famous, inviting the neighborhood to “construct an AI that genuinely speaks to, and for, India.”
Together with Ramakrishnan, the contributors of the mannequin embrace Kundeshwar Pundalik, Durga S, Prateek Chanda, Vedant Goswami, Atul Kumar Singh, Saral Sureka, Panditi Bhagawan, Ajay Nagpal, Smita Gautam, Pankaj Singh, Rishi Bal, and Prof Rohit Saluja.
Whereas earlier talking with AIM, Ramakrishnan mentioned that in contrast to personal entities, BharatGen operates with a transparent mission—‘GenAI for Bharat, by Bharat’. With an funding of below ₹235 crores, which is near $27 million, leveraging cost-efficient computing and attracting prime expertise from graduates from IITs.
Learn: This Govt-Funded ₹235 Cr AI Initiative is India’s Real Answer to DeepSeek
The BharatGen consortium includes IIT Bombay, IIT Kanpur, IIT Mandi, IIT Madras, IIT Hyderabad, IIIT Hyderabad, and IIM Indore.
Vaishnaw had earlier said that India would have its personal foundational AI fashions inside 7-8 months, and BharatGen is a key a part of that imaginative and prescient. “Sure, we’re very a lot on monitor. The Minister has been briefed, and we’re aligned with the timeline,” Ramakrishnan had confirmed earlier. Param-1 is a transparent signal of that roadmap.
“Our purpose isn’t just to construct AI fashions however to supply assets that startups and system integrators can leverage,” mentioned Ramakrishnan.
Final month, MeitY additionally chosen Sarvam AI below the IndiaAI Mission to develop India’s sovereign LLM as a part of the trouble to create indigenous AI capabilities. The group had proposed the event of a 70-billion parameter multimodal AI mannequin that helps each Indian languages and English, and work on it has already begun.
The submit BharatGen Launches Param-1 India’s Foundational LLM Constructed from Scratch appeared first on Analytics India Journal.
