sarvam-1

About Model

Sarvam-1 is a 2-billion parameter language model designed for Indian languages, optimized for token efficiency and high-quality training data. It outperforms larger models in several benchmarks, providing superior performance on Indic language tasks with enhanced computational efficiency. Built with a custom tokenizer and 4 trillion tokens from diverse sources, Sarvam-1 is ideal for applications like translation and edge device deployment.