Huang also said that NVIDIA has partnered with India to make this extensive Hindi language model
Expressing the challenges in developing this model, Huang said, “Hindi dialect changes every 50 Kms"
LLMs use machine learning to perform natural language processing (NLP) tasks
US-based AI chip giant NVDIA’s CEO Jensen Huang today (October 24) said that Hindi is the hardest language model to be developed across the world.
Huang was speaking at the NVIDIA AI Summit in Mumbai.
Lauding India’s effort and achievements in developing the Hindi language model (LLMs), he said, “If India can figure out how to make a large Hindi language model then you can build it in any part of the world.”
Huang also said that NVIDIA has partnered with India to make this extensive Hindi language model.
Expressing the challenges in developing this model, he said, “Hindi dialect changes every 50 Kms.”
“Once you have cracked LLMs in India, you can build it in any part of the world,” said Huang
NVIDIA has partnered with Krutrim AI and Gnani among others to develop Hindi LLMs (Large Language Models). LLMs use machine learning to perform natural language processing (NLP) tasks.
Huang further said that service partners Wipro, Infosys and TCS are helping them to take AI models to the rest of the world.
Speaking about computing infrastructure in the country, he said,” Yotta, Tata Communications and our other partners are joining us to build fundamental computing infrastructure in India.”
Earlier this year, NVIDIA began offering its latest chips to its Indian partners, such as Tata Communications and Jio Platforms to support India in building AI-cloud infrastructure.
The chip behemoth has notably released a small language model for Hindi available as an NVIDIA NIM microservice. The model Nemotron-4-Mini-Hindi-4B can be easily deployed on any NVIDIA GPU-accelerated system for optimized performance, as per NVIDIA.
During the Global AI Summit in July, NVIDIA executive Mohit Sewak stressed upon the need for multimodal LLMs in India. He said that it would require large-scale, multilingual datasets and synthetic data generation.
Sewak also mentioned that the largest model he knows can deal with 100 languages and India has around 23 official languages with around 10, 500 unique dialects.
