TL;DR:
- Indian companies like Ola, Sarvam AI, CoRover.ai, and Zoho are making significant strides in the LLM arena.
- Krutrim Si Designs by Ola offers multilingual AI models with impressive capabilities in 22 Indian languages.
- Sarvam AI’s OpenHathi-Hi-v0.1 promises top-tier performance for Indic languages.
- CoRover.ai’s BharatGPT empowers developers with multilingual virtual assistant capabilities.
- Zoho is developing AI models for specific domain challenges.
- Tech Mahindra’s Indus Project aims to establish a foundational LLM for Indian languages.
- Reliance-NVIDIA and Tata-NVIDIA partnerships focus on AI infrastructure and upskilling.
- AI4Bharat champions open-source language AI with models like IndicBart and IndicBERT.
- Project Vaani collects speech data to build a comprehensive language model for India.
Main AI News:
In a recent visit to India, OpenAI CEO Sam Altman raised eyebrows by casting doubt on the possibility of India producing a groundbreaking Large Language Model (LLM). However, the landscape of the LLM arena in India tells a different story. The country is witnessing a surge of innovative developments as Indian companies silently forge ahead, each with ambitious aspirations and unique strengths.
Ola’s Bold Entry into the LLM Realm
One notable player in this field is Bhavish Aggarwal, co-founder of Ola, who has entered the Indian LLM arena with Krutrim Si Designs, a new AI venture. The company’s inaugural offering is the Krutrim family of multilingual AI models, with “Krutrim” meaning “artificial” in Sanskrit.
Krutrim comes in two sizes: the base model, which has been trained on a staggering 2 trillion tokens and unique datasets, and the larger multi-model Krutrim Pro, set to launch next quarter, designed for advanced problem-solving and task execution. Tokens, the essential building blocks for LLMs in processing and representing text data, play a pivotal role in Krutrim’s capabilities.
What sets Krutrim apart is its remarkable multilingual prowess, with the ability to comprehend 22 Indian languages and generate content in 10, including Marathi, Hindi, Telugu, Kannada, and Odiya. The claim even stands that Krutrim outperforms OpenAI’s GPT-4 in handling Indic languages.
Sarvam AI’s OpenHathi: A Game-Changer for Hindi
Sarvam AI, an AI startup, made significant waves with the release of OpenHathi-Hi-v0.1, the first Hindi LLM in the OpenHathi series. Built on Meta AI’s Llama2-7B architecture, OpenHathi-Hi-v0.1 promises performance on par with GPT-3.5 for Indic languages.
The uniqueness of Sarvam AI’s approach lies in a 48,000-token extension to Llama2-7B’s tokenizer and a two-phase training process. The first phase incorporates embedding alignment, aligning randomly initialized Hindi embeddings. The second phase, bilingual language modeling, focuses on cross-lingual token attention, ensuring real-world applicability.
In a remarkable move, Sarvam AI collaborated with KissanAI to fine-tune its base model using conversational data collected from interactions between a GPT-powered bot and farmers in various languages.
CoRover.ai’s BharatGPT: Bridging the Multilingual Gap
CoRover.ai, based in Bengaluru, unveiled BharatGPT, its indigenous LLM, earlier this month. Developed in collaboration with Bhashini, a National Language Translation Mission (NLTM) under the Ministry of Electronics and Information Technology (MeitY), BharatGPT supports over 12 Indian languages.
CoRover.ai specializes in AI virtual assistants, including chatbots, voice bots, and video bots, serving numerous organizations such as IRCTC, LIC, IGL, KSRTC, NPCI, and the Government of India. They boast a user base exceeding 1 billion, and BharatGPT empowers developers and business users to effortlessly create multilingual text and voice-enabled virtual assistants.
Zoho’s Ascent in the LLM Race
Indian software-as-a-service (SaaS) giant Zoho declared its intentions to develop its LLM, mirroring the likes of OpenAI’s GPT and Google’s PaLM 2. Sridhar Vembu, Zoho’s founder, revealed during the CNBC-TV18 and Moneycontrol Global AI Conclave that the company is working on smaller AI models ranging from 7 billion to 20 billion parameters to address specific domain challenges for its clients.
Zoho recently introduced a suite of 13 generative AI extensions and integrations for its applications, all powered by ChatGPT.
Tech Mahindra’s Indus Project: Rising to the Challenge
Tech Mahindra’s CEO, CP Gurnani, wasted no time in responding to Altman’s skepticism, stating, “Challenge accepted.” The company promptly unveiled the Indus Project, aimed at creating a foundational LLM for Indian languages. The initial phase focuses on crafting an LLM for the Hindi language, encompassing 40 dialects, with plans to launch it early next year.
Tech Mahindra envisions building an Open Source Large Language AI model to cater to the needs of 25 percent of the world’s population.
Reliance-NVIDIA and Tata-NVIDIA: Powerhouse Partnerships
Global chip design leader NVIDIA and Reliance Industries have joined forces to develop a foundational LLM tailored to India’s diverse languages, tailored for generative AI applications. Reliance Jio will oversee the execution, leveraging its expertise in mobile telephony, 5G spectrum, fiber networks, and related domains.
In another strategic partnership, NVIDIA collaborates with the Tata Group to work on generative AI applications and provide AI upskilling to over six lakh employees at TCS. This collaboration extends to implementing AI across various domains, including design, styling, engineering, simulation testing, and autonomous vehicle capabilities. Additionally, NVIDIA will support Tata Communications in constructing AI infrastructure.
AI4Bharat: Championing Open-Source Language AI
Backed by Nandan Nilekani and based at IIT Madras, AI4Bharat focuses on developing open-source language AI for Indian languages. They offer datasets, models, and applications, with notable models like IndicBart and IndicBERT. These models are designed for Indic languages, and despite their relatively fewer parameters, they excel across various tasks.
Project Vaani: A Collaborative Language Model Initiative
The Indian Institute of Sciences (IISc), AI and Robotics Technology Park (ARTPARK), and Google have united to create Project Vaani. This initiative aims to build a comprehensive language model by collecting and transcribing open-source, anonymized speech data from all 773 districts in India. Project Vaani emphasizes diversity across linguistic, educational, urban-rural, age, and gender spectrums across three distinct phases, with the initial phase targeting 80 districts across 10 states.
Conclusion:
The burgeoning developments in India’s Large Language Model (LLM) ecosystem indicate a significant shift in the AI market. Indian companies are poised to challenge established players with their diverse language capabilities and innovative approaches. This surge of LLM initiatives demonstrates India’s growing influence in the global AI revolution, offering businesses new opportunities for multilingual applications and AI-driven solutions.
