IndicTrans2 model is built on a transformer-based sequence-to-sequence architecture, the model leverages script unification and shared lexical features to enhance translation quality across diverse scripts. It supports extended context lengths up to 2048 tokens and incorporates efficient attention mechanisms like FlashAttention for improved performance. This model is particularly suited for applications requiring high-quality English-to-Indic language translations in both formal and informal contexts.
MIT
Sector Agnostic
Jay Gala and Pranjal A Chitale and A K Raghavan and Varun Gumma and Sumanth Doddapaneni and Aswanth Kumar M and Janki Atul Nawale and Anupama Sujatha and Ratish Puduppully and Vivek Raghavan and Pratyush Kumar and Mitesh M Khapra and Raj Dabre and Anoop Kunchukuttan
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
Open
02/05/25 11:00:51
N.A.
0
© 2026 - Copyright AIKosh. All rights reserved.