lunyoro-nllb_lun2en

Fine-tuned NLLB-200 model for Lunyoro/Rutooro β†’ English translation.

Lunyoro-Rutooro is a Bantu language spoken by the Bunyoro-Kitara and Tooro kingdoms in western Uganda.

Model Details

  • Base model: facebook/nllb-200-distilled-600M
  • Fine-tuned on: ~53,948 English-Lunyoro sentence pairs
  • Training: 10 epochs, AdamW optimizer, cosine LR schedule
  • Hardware: NVIDIA GPU (CUDA)
  • Source language code: run_Latn
  • Target language code: eng_Latn

Dataset

The training data was compiled from:

  • Crowd-sourced word and sentence submissions
  • Runyoro-Rutooro dictionary entries (Excel)
  • Parallel sentence corpora
  • Back-translation augmentation (~53,948 total pairs, quality-filtered)

Usage

from transformers import NllbTokenizer, AutoModelForSeq2SeqLM

model_name = "keithtwesigye/lunyoro-nllb_lun2en"
tokenizer = NllbTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

tokenizer.src_lang = "run_Latn"
text = "Oraire ota?"
inputs = tokenizer(text, return_tensors="pt", truncation=True)
output = model.generate(
    **inputs,
    forced_bos_token_id=tokenizer.convert_tokens_to_ids("eng_Latn"),
    num_beams=4,
    max_length=256
)
print(tokenizer.decode(output[0], skip_special_tokens=True))

Related Models

Model Description
lunyoro-en2lun MarianMT English β†’ Lunyoro
lunyoro-lun2en MarianMT Lunyoro β†’ English
lunyoro-nllb_en2lun NLLB-200 English β†’ Lunyoro
lunyoro-nllb_lun2en NLLB-200 Lunyoro β†’ English

Full Application

Source code: chriskagenda/TRANSLATOR

Downloads last month
174
Safetensors
Model size
0.6B params
Tensor type
F32
Β·
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Spaces using keithtwesigye/lunyoro-nllb-lun2en 2