Bengaluru-based startup Sarvam AI was selected by Government of India to build country’s first sovereign Large Language Model (LLM) under the IndiaAI Mission. Sarvam was chosen from over 400 applicants for this strategic initiative aimed at establishing India’s AI autonomy.
Key Features of Initiative
Aim: To create an indigenous AI foundational model capable of reasoning, voice-first design, and fluency across multiple Indian languages.
Indigenously Built: The model will be built, deployed, and optimized entirely within India using domestic infrastructure, promoting strategic autonomy and strengthening India’s AI leadership.
Compute Allocation:
- GoI has allocated 4,096 Nvidia H100 GPUs for 6 months (via empanelled partners: Jio, CtrlS, Yotta, Tata Communications).
Model Variants: Sarvam will develop three model variants:
- Sarvam-Large – for advanced reasoning & generation.
- Sarvam-Small – for real-time interactive applications.
- Sarvam-Edge – for compact, on-device usage.
Collaboration:
- Sarvam AI is working with AI4Bharat at IIT Madras, a leader in Indian language AI research.
- The model will support fluency in 22 Indian languages + English.
Deployment Timeline:
- The model is expected to be ready for population-scale deployment in 6 months.
Challenges Identified
Data Diversity & Availability:
- Difficulty in accessing large, diverse datasets covering India’s linguistic diversity and dialects.
- Non-English Indian languages have complex grammar, syntax, and contextual nuances.
Bias & Fairness:
- Need to address gender, caste, religion, and societal biases while ensuring fairness in outputs.
Content Curation:
- Data cleansing, copyright, and licensing challenges pose hurdles in building large-scale datasets.
Interoperability:
- Ensuring smooth integration across devices, applications, and platforms will be technically demanding.
Data Deficit:
- India needs vast multimodal datasets from private (Jio, Airtel, Zomato, etc.) and public sectors (health, education, agriculture, etc.).
- AIKosh (IndiaAI Datasets Platform) has been launched to address this gap, but more efforts are needed.
About IndiaAI Mission
Launched in: 2023
By: Ministry of Electronics & IT + Nasscom (National Association of Software and Service Companies)
Budget: ₹10,372 crore (~$1.25 billion) announced in March 2024.
Mission Goals:
- “Making AI in India” – Develop domestic AI models.
- “Making AI Work for India” – Apply AI solutions for Indian sectors.
Key Pillars of IndiaAI Mission:
Common Compute Facility:
- Provides GPU access to startups/researchers.
- Reduces dependency on costly foreign computing resources.
AIKosh (IndiaAI Datasets Platform):
- Develops India-specific datasets to train AI.
- Reduces over-reliance on Western-trained models.
- Supports linguistic and cultural inclusivity.
AI Safety Institute of India (Upcoming): Focuses on AI risk assessment, safety guidelines, and ensuring secure AI tools.
IndiaAI Innovation Centre: Builds domain-specific foundational AI models.
AI Application Development Initiative: Supports AI-based solutions for commercial & social sectors.
Future Skills Initiative: Establishes AI labs in smaller cities to develop talent.
Startup Financing: Provides funding to AI startups to promote innovation.
Significance
- India is among the top countries globally in AI advancement (Stanford AI Index 2024).
- India leads globally in AI skill penetration (UNESCO State of Education Report 2022).
- India houses 2,975 Global Capability Centers (GCCs) employing 1.9 million professionals (Zinnov report).
- The mission will democratize access to AI computing resources, foster domestic innovation, and position India as an AI powerhouse.