Sarvam AI Launches Multilingual AI Model Sarvam 1, Boosting AI Efficiency
Hyderabad: Bengaluru-based startup Sarvam AI has recently unveiled the Sarvam 1 LLM, a groundbreaking open-source AI model. This innovative model has been meticulously trained on 11 different languages, namely Bengali, English, Gujarati, Hindi, Kannada, Marathi, Malayalam, Oriya, Punjabi, Tamil, and Telugu.
Sarvam 1, boasting a 2-billion-parameter structure, has been trained on an impressive 4 trillion tokens utilizing a custom tokeniser deployed on Nvidia H100 Tensor Core GPUs. One of its standout features is its exceptional efficiency, claimed to be up to four times more effective compared to other AI models focused on Indian languages.
Enhancing Multilingual Capabilities
To cater to various multilingual tasks, the Sarvam-2T training corpus incorporates datasets covering 20% of content in languages such as Hindi, English, and programming languages. Leveraging synthetic data generation techniques, Sarvam AI has effectively tackled the challenge of limited high-quality training data for Indian languages.
Developers can leverage the base model of Sarvam AI, accessible on Hugging Face, to develop their own AI applications tailored for Indic languages. A significant milestone includes the launch of the nation's first Hindi LLM, Open Hathi, in December 2023, built on Meta AI's Llama models.
In August 2024, the startup rolled out its foundational AI model, Sarvam 2B, reinforcing its position as a key player in advancing AI for Indic languages.
Recognition and Acknowledgment
Meta showcased the efforts of Sarvam AI at its recent 'Build with AI Summit' in Bengaluru, applauding the startup's contributions towards advancing AI for Indic languages and creating Hindi LLMs despite operating under resource constraints.
Further discussions at the summit included insights from Meta's VP and Chief AI Scientist, Yann LeCun, accentuating India's flourishing developer ecosystem and the vast talent pool within the country. The event underscored India's potential to spearhead AI adoption and integration across various sectors, showcasing initiatives like the Skill AI Chatbot for AI-driven education across multiple languages.
Meta also shed light on the impact of open-source Llama models, empowering platforms like Arivihan, India's pioneering fully automated online learning platform.
AI Hackathon and Impact Grants
Prior to the summit, Meta organized an AI Hackathon aimed at fostering innovation through the utilization of Meta Llama's open platform. The event witnessed over 1,500 registrations, with notable winners including CurePharmaAI, CivicFix, and evAIssment. An all-women team, SheBuilds, received special recognition.
Participants were also encouraged to submit proposals for the Llama Impact Grants, offering potential funding of up to $100K regionally and $500K globally. These grants play a vital role in supporting startups in their endeavors to develop AI solutions.
In 2023, the Indian non-profit Wadhwani AI garnered global acclaim for leveraging Llama 3 to enhance its AI-enabled oral reading fluency assessment, creating personalized practice modules for students.
For all the latest updates, stay connected with us!
TAGGED: World Stroke Day Marks Driving Actions Against Stroke, Ayurveda Day 2024: Celebrating Ancient Wisdom And Modern Innovations For Global Health, PM Modi and Spanish PM Pedro Sanchez Pause Vadodara Roadshow To Meet Differently-Abled Artist Diya Gosai, Tragic Electric Shock In Chhattisgarh Claims Three Lives, 2 Men Jailed In UK For Illegal Import Of Pharma Drugs From India.
For media partnership and Ad-sales inquiries, please reach out to us via Email: [email protected] Copyright © 2024 Ushodaya Enterprises Pvt. Ltd., All Rights Reserved.
