@inproceedings{singh-etal-2026-nanda,
    title = "Nanda Family: Open-Weights Generative Large Language Models for {H}indi",
    author = "Singh, Aaryamonvikram  and
      Banerjee, Debopriyo  and
      Sahnan, Dhruv  and
      Choudhury, Monojit  and
      Chauhan, Shivam  and
      Das, Rocktim Jyoti  and
      Han, Xudong  and
      Li, Haonan  and
      Jadhav, Alok Anil  and
      Agarwal, Utkarsh  and
      Choudhary, Mukund  and
      Koto, Fajri  and
      Bhat, Junaid Hamid  and
      Shukla, Awantika  and
      Ghosh, Samujjwal  and
      Kamboj, Samta  and
      Pandit, Onkar  and
      Pradhan, Lalit  and
      Pal, Rahul  and
      Sahu, Sunil Kumar  and
      Mullah, Parvez  and
      El Filali, Ali  and
      Quraishi, Zainul Abedien Ahmed  and
      Sengupta, Neha  and
      Ramakrishnan, Gokulakrishnan  and
      Joshi, Rituraj  and
      Gosal, Gurpreet  and
      Sheinin, Avraham  and
      Vassilieva, Natalia  and
      Nakov, Preslav",
    editor = "Demberg, Vera  and
      Inui, Kentaro  and
      Marquez, Llu{\'i}s",
    booktitle = "Proceedings of the 19th Conference of the {E}uropean Chapter of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
    month = mar,
    year = "2026",
    address = "Rabat, Morocco",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-eacl/2026.eacl-long.288/",
    pages = "6086--6108",
    ISBN = "979-8-89176-380-7",
    abstract = "Large language models remain predominantly English-centric, which limits their utility for underrepresented languages. We help bridge this gap for Hindi with Llama-3-Nanda-10B-Chat (aka Nanda-10B) and Llama-3.1-Nanda-87B-Chat (aka Nanda-87B), forming the Nanda family of open-weight bilingual models (https://github.com/MBZUAI-IFM/Nanda-Family). Our approach integrates: (i) a tokenizer extending Llama{'}s vocabulary with 20{\%} Hindi-specific tokens, thus halving Hindi tokenization fertility while preserving English efficiency, (ii) Hindi-first parameter-efficient continual pretraining using Llama Pro on a 65B-token corpus spanning Devanagari script, code-mixed, and Romanized Hindi, and (iii) bilingual instruction and safety alignment on a large culturally grounded dataset. The resulting Nanda models outperform open-weight LLMs of comparable size: Nanda-87B yields high generative quality, and Nanda-10B shows competitive general-purpose performance. Nanda-87B demonstrates state-of-the-art performance on summarization, translation, transliteration, and instruction following. Moreover, both models achieve state-of-the-art performance in safety and in cultural knowledge. Our results demonstrate that careful tokenizer design, data curation, and continual pretraining can yield capable and safe LLMs for resource-poor languages without compromising English performance."
}