This is a variant of Zabantu pre-trained on a monolingual dataset of Sepedi(nso) sentences on a transformer network with 120 million traininable parameters.

Usage Example(s)

python

from transformers import pipeline

# Initialize the pipeline for masked language model
unmasker = pipeline('fill-mask', model='dsfsi/zabantu-nso-120m')

# The Sepedi sentence with a masked token
sample_sentences = ["mopresidente wa <mask> wa afrika-borwa",   # original token: maloba
"bašomedi ba polase ya dinamune ya zebediela citrus ba hlomile magato a <mask> malebana le go se sepetšwe botse ga dilo ka polaseng eo."  # original token: boipelaetšo
]

# Perform the fill-mask task
results = unmasker(sentence)

# Display the results
for result in results:
    print(f"Predicted word: {result['token_str']} - Score: {result['score']}")
    print(f"Full sentence: {result['sequence']}\n")
    print("=" * 80)