Tech Mahindra, a prominent IT services provider, has unveiled a Hindi-first large language model (LLM) tailored for educational purposes. This innovative model, developed in partnership with NVIDIA under the ‘Project Indus’ initiative, aims to democratize education by offering foundational knowledge in disciplines such as physics in natural Hindi.
The newly introduced model, equipped with an advanced framework boasting 8 billion parameters, seeks to foster inclusivity and accessibility in the education sector. It supports the creation of autonomous AI agents capable of conversing fluently in Hindi, a significant advancement from its predecessor with 1.2 billion parameters.
To power its enhanced capabilities, the education-centric variant leverages NVIDIA’s cutting-edge AI frameworks, including the NeMo framework and NVIDIA NIM microservices. Overcoming the challenge of limited Hindi-language datasets, the model has been trained on 500 million synthetic tokens generated by NVIDIA’s NeMo Data Designer.
Tech Mahindra’s CIO, Nikhil Malhotra, emphasized the pivotal role of AI in driving national digital infrastructure and fostering inclusive growth. He highlighted the necessity for domain-specific language models tailored for linguistically diverse countries like India, particularly in educational contexts. Through Project Indus, the collaboration with NVIDIA addresses this gap by delivering a Hindi-first, indigenous AI model, facilitating scalable and pertinent AI-driven learning experiences and citizen-centric services in India.
Source: Inc42 Media