Edit model card


This repository contains the IndicBART checkpoint finetuned on the 11 languages of IndicParaphrase dataset. For finetuning details, see the paper.

  • Supported languages: Assamese, Bengali, Gujarati, Hindi, Marathi, Odiya, Punjabi, Kannada, Malayalam, Tamil, and Telugu. Not all of these languages are supported by mBART50 and mT5.
  • The model is much smaller than the mBART and mT5(-base) models, so less computationally expensive for decoding.
  • Trained on large Indic language corpora (5.53 million sentences).
  • All languages, have been represented in Devanagari script to encourage transfer learning among the related languages.

Using this model in transformers

from transformers import MBartForConditionalGeneration, AutoModelForSeq2SeqLM
from transformers import AlbertTokenizer, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("ai4bharat/MultiIndicParaphraseGeneration", do_lower_case=False, use_fast=False, keep_accents=True)
# Or use tokenizer = AlbertTokenizer.from_pretrained("ai4bharat/MultiIndicParaphraseGeneration", do_lower_case=False, use_fast=False, keep_accents=True)
model = AutoModelForSeq2SeqLM.from_pretrained("ai4bharat/MultiIndicParaphraseGeneration")
# Or use model = MBartForConditionalGeneration.from_pretrained("ai4bharat/MultiIndicParaphraseGeneration")

# Some initial mapping
bos_id = tokenizer._convert_token_to_id_with_added_voc("<s>")
eos_id = tokenizer._convert_token_to_id_with_added_voc("</s>")
pad_id = tokenizer._convert_token_to_id_with_added_voc("<pad>")

# To get lang_id use any of ['<2as>', '<2bn>', '<2en>', '<2gu>', '<2hi>', '<2kn>', '<2ml>', '<2mr>', '<2or>', '<2pa>', '<2ta>', '<2te>']
# First tokenize the input. The format below is how IndicBART was trained so the input should be "Sentence </s> <2xx>" where xx is the language code. Similarly, the output should be "<2yy> Sentence </s>".
inp = tokenizer("दिल्ली यूनिवर्सिटी देश की प्रसिद्ध यूनिवर्सिटी में से एक है. </s> <2hi>", add_special_tokens=False, return_tensors="pt", padding=True).input_ids 

# For generation. Pardon the messiness. Note the decoder_start_token_id.

model_output=model.generate(inp, use_cache=True,no_repeat_ngram_size=3,encoder_no_repeat_ngram_size=3, num_beams=4, max_length=20, min_length=1, early_stopping=True, pad_token_id=pad_id, bos_token_id=bos_id, eos_token_id=eos_id, decoder_start_token_id=tokenizer._convert_token_to_id_with_added_voc("<2hi>"))

# Decode to get output strings
decoded_output=tokenizer.decode(model_output[0], skip_special_tokens=True, clean_up_tokenization_spaces=False)
print(decoded_output) # दिल्ली विश्वविद्यालय देश की प्रमुख विश्वविद्यालयों में शामिल है।

# Note that if your output language is not Hindi or Marathi, you should convert its script from Devanagari to the desired language using the Indic NLP Library.


If you wish to use any language written in a non-Devanagari script, then you should first convert it to Devanagari using the Indic NLP Library. After you get the output, you should convert it back into the original script.


Scores on the IndicParaphrase test sets are as follows:

Language BLEU / Self-BLEU / iBLEU
as 1.66 / 2.06 / 0.54
bn 11.57 / 1.69 / 7.59
gu 22.10 / 2.76 / 14.64
hi 27.29 / 2.87 / 18.24
kn 15.40 / 2.98 / 9.89
ml 10.57 / 1.70 / 6.89
mr 20.38 / 2.20 / 13.61
or 19.26 / 2.10 / 12.85
pa 14.87 / 1.35 / 10.00
ta 18.52 / 2.88 / 12.10
te 16.70 / 3.34 / 10.69


If you use this model, please cite the following paper:

  title={IndicNLG Suite: Multilingual Datasets for Diverse NLG Tasks in Indic Languages},
  author={Aman Kumar and Himani Shrotriya and Prachi Sahu and Raj Dabre and Ratish Puduppully and Anoop Kunchukuttan and Amogh Mishra and Mitesh M. Khapra and Pratyush Kumar},
  url = "https://arxiv.org/abs/2203.05437"
Downloads last month
Hosted inference API
This model can be loaded on the Inference API on-demand.

Dataset used to train ai4bharat/MultiIndicParaphraseGeneration