emuduki/ateso-english-mt

Ateso-English Machine Translation Model

Fine-tuned from Helsinki-NLP/opus-mt-en-mul on the Ateso-English Parallel Corpus.

Language pair: Ateso (teo) -> English (en)

Training data: emuduki/ateso-english-parallel-corpus

Usage

from transformers import MarianMTModel, MarianTokenizer

model_name = "emuduki/ateso-english-mt"
tokenizer  = MarianTokenizer.from_pretrained(model_name)
model      = MarianMTModel.from_pretrained(model_name)

def translate(text):
    inputs = tokenizer([text], return_tensors="pt", padding=True)
    output = model.generate(**inputs, num_beams=4)
    return tokenizer.decode(output[0], skip_special_tokens=True)

print(translate("Ibere Aiyong eong icomu iguru"))
# Expected: In the beginning God created the heavens

Training details

• Base model: Helsinki-NLP/opus-mt-en-mul
• Fine-tuning: 5 epochs, batch size 64 (effective), fp16
• Hardware: Google Colab T4 GPU
• Dataset: first open parallel corpus for Ateso (teo)

Built on 2026-05-31.