aleksahet/BERTic-squad-sr-lat

BERTić-SQuAD-sr-lat

BERTić-SQuAD-sr-lat is a Question Answering neural network for Serbian. It is obtained by fine-tuning BERTić on a synthetically generated Serbian QA dataset based on SQuAD v1.1.

Usage

from transformers import pipeline

model_name = 'aleksahet/BERTic-squad-sr-lat'

pipe = pipeline('question-answering', model=model_name, tokenizer=model_name)

sample = {
  'question': 'Kojim sportom se bavi Novak Đoković?',
  'context': 'Novak Đoković (Beograd, 22. maj 1987) srpski je teniser. Mnogi teniski kritičari, bivši igrači i saigrači smatraju Đokovića jednim od najboljih tenisera u istoriji.'
}

res = pipe(sample)

Performance

The model vas evaluated on XQuAD samples translated to Serbian. We report the following results:

• Exact Match: 73.91%
• F1: 82.97%