Namaste, Vanakam, Sat Sri Akar — these are just three forms of greeting in India, which has 22 constitutionally recognized languages and more than 1,500 more recorded in the census. Approximately 10% of residents speak English, the most common language on the internet.
As the world’s most populous country moves forward with its rapid digitalization journey, Indian companies and local startups are developing multilingual AI that will make technology accessible to more Indians in their first language. The model is currently being developed. This is a case study in the development of sovereign AI, a national AI infrastructure that is built on local datasets and reflects regionally specific dialects, cultures, and customs.
These projects will power customer service AI agents for businesses to quickly translate content, expand access to information, and more easily serve more than 1.4 billion diverse people. , building language models for Indian languages and English.
To support these efforts, NVIDIA released a small language model for Hindi, India’s most popular language with over 500 million speakers. This model, called Nemotron-4-Mini-Hindi-4B, is now available as an NVIDIA NIM microservice and can be easily deployed on NVIDIA GPU acceleration systems to optimize performance.
Tech Mahindra, an Indian IT services and consulting company, has used the Nemotron Hindi NIM microservice for the first time to develop an AI model called Indus 2.0 that focuses on Hindi and its dozens of dialects. Indus 2.0 leverages Tech Mahindra’s high-quality fine-tuning data to further improve model accuracy and expand opportunities for clients in banking, education, healthcare, and other industries to offer localized services.
Tech Mahindra will introduce Indus 2.0 at the NVIDIA AI Summit to be held in Mumbai from October 23-25. The company is also developing TeNo, a sovereign large-scale language model (LLM) platform, using NVIDIA NeMo.
NVIDIA NIM makes AI adoption in Hindi as easy as Ek, Do, Teen
The Nemotron Hindi model has 4 billion parameters and is derived from Nemotron-4 15B, a 15 billion parameter multilingual language model developed by NVIDIA. The model combines real-world Hindi data, synthetic Hindi data, and an equivalent amount of English data using NVIDIA NeMo, an end-to-end cloud-native framework and suite of microservices for generative development. , pruned, extracted and trained. love.
This dataset was created with NVIDIA NeMo Curator, which improves the accuracy of generative AI models by processing high-quality multimodal data at scale for training and customization. NeMo Curator uses the NVIDIA RAPIDS library to accelerate data processing pipelines on multi-node GPU systems, reducing processing time and total cost of ownership. It also provides prebuilt pipelines and building blocks for synthetic data generation, data filtering, classification, and deduplication to process high-quality data.
After fine-tuning with NeMo, the final model leads multiple accuracy benchmarks for AI models with up to 8 billion parameters. Packaged as a NIM microservice, it can be easily leveraged to support use cases across industries such as education, retail, and healthcare.
It is available as part of the NVIDIA AI Enterprise software platform, giving businesses access to additional resources such as technical support and enterprise-grade security to streamline AI development for production environments.
A group of companies that serve multilingual people
Innovators, leading enterprises, and global system integrators across India use NVIDIA NeMo to build customized language models.
Companies participating in the NVIDIA Inception program for cutting-edge startups are using NeMo to develop AI models in several Indian languages.
Sarvam AI provides voice-to-text, text-to-speech, translation, and data analysis models to enterprise customers. The company has developed Sarvam 1, India’s first indigenous multilingual LLM. It was trained from scratch on domestic AI infrastructure powered by NVIDIA H100 Tensor Core GPUs.
Sarvam 1 is developed using NVIDIA AI Enterprise software, including NeMo Curator and NeMo Framework, and supports English and 10 major Indian languages, including Bengali, Marathi, Tamil, and Telugu. Masu.
Sarvam AI also uses NVIDIA NIM microservices, NVIDIA Riva for conversational AI, NVIDIA TensorRT-LLM software, and NVIDIA Triton Inference Server to optimize and deploy conversational AI agents with sub-second latency .
Another Inception startup, Gnani.ai, is a multilingual text-to-speech LLM that powers an AI customer service assistant that handles approximately 10 million real-time voice interactions every day at more than 150 banking, insurance, and financial services companies in India and the US. built. The model supports 14 languages and was trained on over 14 million hours of conversational audio data using NVIDIA Hopper GPUs and the NeMo Framework.
Gnani.ai uses TensorRT-LLM, Triton Inference Server, and Riva NIM microservices to optimize AI for virtual customer service assistants and voice analytics.
Large companies using NeMo to build LLMs include:
Flipkart, India’s leading e-commerce company majority-owned by Walmart, has launched an open-source toolkit that allows developers to add programmable guardrails to LLM to enhance the safety of conversational AI systems. Integrates with NeMo Guardrails. Krutrim, a company in the Ola Group, which includes India’s top ride booking platform, is developing a multilingual India Foundation Model using Mistral NeMo 12B, a state-of-the-art LLM developed by Mistral AI and NVIDIA. Zoho Corporation, a Chennai-based global technology company, uses NVIDIA TensorRT-LLM and NVIDIA Triton Inference Server to optimize and deliver language models for over 700,000 customers. The company uses NeMo running on NVIDIA Hopper GPUs to pre-train narrow, small, medium, and large models for more than 100 business applications from scratch.
World-class system integrators in India are also offering NVIDIA NeMo-accelerated solutions to their customers.
Infosys will work on specific tools and solutions using the NVIDIA AI stack. The company’s Center of Excellence also develops small-scale, AI-powered language models that are delivered as a service to customers. Tata Consultancy Services has developed AI solutions based on NVIDIA NIM agent blueprints for the telecom, retail, manufacturing, automotive, and financial services industries. TCS’ products include domain-specific, NeMo-powered Contains language models. Wipro uses NVIDIA AI Enterprise software such as NIM Agent Blueprints and NeMo to help businesses easily develop custom conversational AI solutions, including digital humans to support customer service interactions.
Wipro and TCS use NeMo Curator’s synthetic data generation pipeline to generate data in languages other than English and customize LLM for their clients.
To learn more about NVIDIA’s collaborations with Indian companies and developers, watch a replay of the company’s founder and CEO Jensen Huang’s fireside chat at the NVIDIA AI Summit.
