Airavata: Introducing Hindi Instruction-tuned LLM

Jay P. Gala; Thanmay Jayakumar; Jaavid Aktar Husain; Kumar M. Aswanth; Mohammed Safi Ur Rahman Khan; Diptesh Kanojia; Ratish Puduppully; Mitesh M. Khapra; Raj Dabre; V. Rudra Murthy; Anoop Kunchukuttan

doi:10.48550/ARXIV.2401.15006

Airavata: Introducing Hindi Instruction-tuned LLM

Jay P. Gala
, Thanmay Jayakumar
, Jaavid Aktar Husain
, Kumar M. Aswanth
, Mohammed Safi Ur Rahman Khan
, Diptesh Kanojia
, Ratish Puduppully
, Mitesh M. Khapra
, Raj Dabre
, V. Rudra Murthy
, Anoop Kunchukuttan

Indian Institute of Technology Madras
Indian Institute of Information Technology, Design and Manufacturing (Kancheepuram)
University of Surrey
Agency for Science, Technology and Research (A*Star)
National Institute Of Information And Communications Technology, Japan
IBM Research India
Microsoft India

Research output: Other contribution › Research

Abstract

We announce the initial release of "Airavata," an instruction-tuned LLM for Hindi. Airavata was created by fine-tuning OpenHathi with diverse, instruction-tuning Hindi datasets to make it better suited for assistive tasks. Along with the model, we also share the IndicInstruct dataset, which is a collection of diverse instruction-tuning datasets to enable further research for Indic LLMs. Additionally, we present evaluation benchmarks and a framework for assessing LLM performance across tasks in Hindi. Currently, Airavata supports Hindi, but we plan to expand this to all 22 scheduled Indic languages. You can access all artifacts at this https URL.

Original language	English
Publication date	2024
Volume	abs/2401.15006
Number of pages	20
DOIs	https://doi.org/10.48550/ARXIV.2401.15006
Publication status	Published - 2024
Externally published	Yes

Keywords

Airavata
Hindi-language LLM
IndicInstruct dataset
instruction-tuning
LLM evaluation benchmarks

Access to Document

10.48550/ARXIV.2401.15006Licence: CC BY

https://dblp.org/rec/journals/corr/abs-2401-15006

Cite this

@misc{4d725104ace1493eb6b1576532877a8e,

title = "Airavata: Introducing Hindi Instruction-tuned LLM",

abstract = " We announce the initial release of {"}Airavata,{"} an instruction-tuned LLM for Hindi. Airavata was created by fine-tuning OpenHathi with diverse, instruction-tuning Hindi datasets to make it better suited for assistive tasks. Along with the model, we also share the IndicInstruct dataset, which is a collection of diverse instruction-tuning datasets to enable further research for Indic LLMs. Additionally, we present evaluation benchmarks and a framework for assessing LLM performance across tasks in Hindi. Currently, Airavata supports Hindi, but we plan to expand this to all 22 scheduled Indic languages. You can access all artifacts at this https URL. ",

keywords = "Airavata, Hindi-language LLM, IndicInstruct dataset, instruction-tuning, LLM evaluation benchmarks",

author = "Gala, \{Jay P.\} and Thanmay Jayakumar and Husain, \{Jaavid Aktar\} and Aswanth, \{Kumar M.\} and Khan, \{Mohammed Safi Ur Rahman\} and Diptesh Kanojia and Ratish Puduppully and Khapra, \{Mitesh M.\} and Raj Dabre and Murthy, \{V. Rudra\} and Anoop Kunchukuttan",

note = "DBLP's bibliographic metadata records provided through http://dblp.org/search/publ/api are distributed under a Creative Commons CC0 1.0 Universal Public Domain Dedication. Although the bibliographic metadata records are provided consistent with CC0 1.0 Dedication, the content described by the metadata records is not. Content may be subject to copyright, rights of privacy, rights of publicity and other restrictions.",

year = "2024",

doi = "10.48550/ARXIV.2401.15006",

language = "English",

volume = "abs/2401.15006",

type = "Other",

}

TY - GEN

T1 - Airavata: Introducing Hindi Instruction-tuned LLM

AU - Gala, Jay P.

AU - Jayakumar, Thanmay

AU - Husain, Jaavid Aktar

AU - Aswanth, Kumar M.

AU - Khan, Mohammed Safi Ur Rahman

AU - Kanojia, Diptesh

AU - Puduppully, Ratish

AU - Khapra, Mitesh M.

AU - Dabre, Raj

AU - Murthy, V. Rudra

AU - Kunchukuttan, Anoop

N1 - DBLP's bibliographic metadata records provided through http://dblp.org/search/publ/api are distributed under a Creative Commons CC0 1.0 Universal Public Domain Dedication. Although the bibliographic metadata records are provided consistent with CC0 1.0 Dedication, the content described by the metadata records is not. Content may be subject to copyright, rights of privacy, rights of publicity and other restrictions.

PY - 2024

Y1 - 2024

N2 - We announce the initial release of "Airavata," an instruction-tuned LLM for Hindi. Airavata was created by fine-tuning OpenHathi with diverse, instruction-tuning Hindi datasets to make it better suited for assistive tasks. Along with the model, we also share the IndicInstruct dataset, which is a collection of diverse instruction-tuning datasets to enable further research for Indic LLMs. Additionally, we present evaluation benchmarks and a framework for assessing LLM performance across tasks in Hindi. Currently, Airavata supports Hindi, but we plan to expand this to all 22 scheduled Indic languages. You can access all artifacts at this https URL.

AB - We announce the initial release of "Airavata," an instruction-tuned LLM for Hindi. Airavata was created by fine-tuning OpenHathi with diverse, instruction-tuning Hindi datasets to make it better suited for assistive tasks. Along with the model, we also share the IndicInstruct dataset, which is a collection of diverse instruction-tuning datasets to enable further research for Indic LLMs. Additionally, we present evaluation benchmarks and a framework for assessing LLM performance across tasks in Hindi. Currently, Airavata supports Hindi, but we plan to expand this to all 22 scheduled Indic languages. You can access all artifacts at this https URL.

KW - Airavata

KW - Hindi-language LLM

KW - IndicInstruct dataset

KW - instruction-tuning

KW - LLM evaluation benchmarks

U2 - 10.48550/ARXIV.2401.15006

DO - 10.48550/ARXIV.2401.15006

M3 - Other contribution

VL - abs/2401.15006

ER -

Airavata: Introducing Hindi Instruction-tuned LLM

Abstract

Keywords

Access to Document

Fingerprint

Cite this