NVIDIA Scales Indic Language Pretraining to 2.8 Trillion Tokens
October 15, 2026
NVIDIA released an open data generation pipeline and a pretraining dataset totaling 2.8 trillion tokens for Indic languages. The methodology focuses on scaling high-quality multilingual data to improve model performance across diverse linguistic contexts.
HOW THIS AFFECTS YOU
●
builderYou can improve model performance for users in Indic-speaking regions using this data.
●
researcherYou can utilize this large-scale Indic dataset for multilingual pretraining research.