White Paper

Benchmarking the Indus Language Model on Intel® AI Hardware

Benchmarking the Indus Language Model on Intel® AI Hardware

Pages 5 Pages

Project Indus is an open-source Hindi language model, developed by Tech Mahindra’s Makers Lab and optimized with Intel AI hardware. It targets India’s 600M+ Hindi speakers and 37 dialects, aiming to bridge the country’s digital language gap. With 1.2B parameters and trained on 100GB of curated data, Indus is one of the most cost-effective LLMs built ($150K). Benchmarking focused on throughput, latency, and concurrency using Intel® Xeon® 5th Gen CPUs and AMX/AVX-512 optimizations. Results show high scalability and efficiency, providing actionable insights for developers deploying Indic LLMs in production.

Join for free to read