It is a sequence-to-sequence pre-trained model based on IndicBART, fine-tuned on the IndicSentenceSummarization dataset, supporting 11 Indian languages: Hindi, Marathi, Punjabi, Tamil, Telugu, Bengali, Gujarati etc. Smaller than mBART and mT5, it is computationally more efficient for decoding. Trained on a vast corpus of 5.53 million sentences, each language is represented in its own script, removing the need for script mapping to/from Devanagari, making it ideal for summarization tasks across various Indian languages.
MIT
Sector Agnostic
Aman Kumar, Himani Shrotriya, Prachi Sahu, Raj Dabre, Ratish Puduppully, Anoop Kunchukuttan, Amogh Mishra, Mitesh M. Khapra, Pratyush Kumar
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
Open
21/02/25 13:20:56
N.A.
0
© 2026 - Copyright AIKosh. All rights reserved.