Fine-tuned NLLB-200 model for low-resource Kangri language
Supports long input, chunked translation, and optional preprocessing.