Krutrim-2 Instruct

About Model

Krutrim-2 is a transformer-based Large Language Model (LLM) developed by OLA Krutrim Labs, designed for Indic-language generation, multilingual AI tasks, and complex reasoning. The model has been trained on web data, Indian-context text, synthetic data, and books, ensuring robust linguistic representation. Fine-tuned with Direct Preference Optimization (DPO), it improves alignment, safety, and reasoning accuracy. The model outperforms comparable models (5-10x its size) on Indic tasks and delivers top-3 performance on 5 out of 7 BharatBench tasks, making it highly relevant for Indian NLP applications. Featuring 40 layers, a 5,120 hidden dimension, 32 attention heads, and a vocabulary of 131,072 tokens, it supports context lengths up to 128K tokens. Krutrim-2 aims to set a benchmark for Indic AI research, multilingual applications, and advanced NLP capabilities.