Bengaluru-based startup Sarvam AI unveiled two indigenous Large Language Models (LLMs) during the AI Impact Summit 2026, highlighting India’s commitment to Sovereign AI. LLMs are advanced AI systems trained on vast text datasets, enabling understanding and generation of human language. These models, comprising 35B and 105B parameters, utilize a unique tokenization process for efficient language representation, employ a self-attention mechanism for contextual relevance, and leverage stacked transformer layers for deep language understanding.
Key training methods include pre-training on extensive raw data, fine-tuning on specialized datasets, and Reinforcement Learning from Human Feedback (RLHF) for improved accuracy. Features of these LLMs encompass generative capabilities for original content creation, multilingualism, and zero-shot reasoning to tackle novel problems. As AI technology evolves, these LLMs position India as a leader in the global AI landscape, paving the way for innovations across various sectors.
Source link
