kathe2026-indictrans2-en-kas

KATHE 2026 — English → Kashmiri (kas_Arab)

Weights used for the KATHE 2026 shared task submission.

This is a redistributed copy of ai4bharat/indictrans2-en-indic-1B by AI4Bharat, published here so that the task organizers can load it without needing access to the gated upstream repository.

All credit for the underlying model belongs to AI4Bharat. Released under the MIT licence, matching upstream.

Usage

This model requires IndicTransToolkit for pre/postprocessing and trust_remote_code=True. See the inference scripts in the accompanying GitHub repository — load_model.py, infer_single.py, infer_batch.py.

from load_model import load_model, translate
tok, model, ip, device = load_model()
print(translate(["I am going to school today."], tok, model, ip, device))

Citation

@article{gala2023indictrans2,
  title   = {IndicTrans2: Towards High-Quality and Accessible Machine Translation
             Models for all 22 Scheduled Indian Languages},
  author  = {Gala, Jay and Chitale, Pranjal A and Raghavan, Raghavan AK and others},
  journal = {Transactions on Machine Learning Research},
  year    = {2023}
}