IndicTrans2 English → Prakrit (1B)

This model is a fine-tuned version of IndicTrans2 English→Indic 1B, adapted for the English → Prakrit machine translation task.

The model was developed as part of the research presented in English-to-Prakrit Machine Translation via Multilingual Transfer Learning.

Base Model

  • Base Model: ai4bharat/indictrans2-en-indic-1B
  • Architecture: IndicTrans2 (Encoder–Decoder Transformer)

Usage

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer

model = AutoModelForSeq2SeqLM.from_pretrained(
    "D3v1s0m/indictrans2-en-prakrit-1B",
    trust_remote_code=True,
)

tokenizer = AutoTokenizer.from_pretrained(
    "D3v1s0m/indictrans2-en-prakrit-1B",
    trust_remote_code=True,
)

Training

The model was obtained by fine-tuning the IndicTrans2 English→Indic 1B checkpoint on an English–Prakrit parallel corpus using supervised fine-tuning with the Hugging Face Transformers library.

Evaluation

The model was evaluated using SacreBLEU.

Evaluation scores will be updated as additional experiments are completed.

Limitations

  • Intended for English → Prakrit translation.
  • Performance depends on the quality and domain of the input text.
  • The model has been fine-tuned on a specific English–Prakrit parallel corpus and may not generalize equally well to all Prakrit dialects or historical variants.

Citation

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

@misc{choksi2026englishtoprakritmachinetranslationmultilingual,
      title={English-to-Prakrit Machine Translation via Multilingual Transfer Learning},
      author={Om Choksi and Smit Kareliya and Shrikant Malviya and Pruthwik Mishra},
      year={2026},
      eprint={2606.06038},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2606.06038},
}

Paper: https://arxiv.org/abs/2606.06038

Acknowledgements

This model builds upon the following open-source resources:

We gratefully acknowledge the AI4Bharat team for developing IndicTrans2 and making the pretrained model publicly available.

License

This model is a fine-tuned derivative of ai4bharat/indictrans2-en-indic-1B.

Please refer to the upstream model repository for licensing information regarding the original model, tokenizer, and accompanying source files included with this repository.

Downloads last month
27
Safetensors
Model size
1B params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for D3v1s0m/indictrans2-en-prakrit-1B

Finetuned
(5)
this model

Paper for D3v1s0m/indictrans2-en-prakrit-1B