Macedonian-ASR/mt5-restore-capitalization-macedonian

Fine-tuned mt5-base model for restoring capitalization and punctuation for Macedonian language

The model is fine-tuned on a subset of the Macedonian portion of Wikipedia.

Authors:

1. Dejan Porjazovski
2. Ilina Jakimovska
3. Ordan Chukaliev
4. Nikola Stikov

This collaboration is part of the activities of the Center for Advanced Interdisciplinary Research (CAIR) at UKIM.

Usage

pip install transformers

from transformers import T5Tokenizer, T5ForConditionalGeneration

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

recap_model_name = "Macedonian-ASR/mt5-restore-capitalization-macedonian"
recap_tokenizer = T5Tokenizer.from_pretrained(recap_model_name)
recap_model = T5ForConditionalGeneration.from_pretrained(recap_model_name)
recap_model.to(device)

sentence = "скопје е главен град на македонија"
inputs = recap_tokenizer(["restore capitalization and punctuation: " + sentence], return_tensors="pt", padding=True).to(device)
outputs = recap_model.generate(**inputs, max_length=768, num_beams=5, early_stopping=True).squeeze(0)
recap_result = recap_tokenizer.decode(outputs, skip_special_tokens=True)
print(recap_result)
-> "Скопје е главен град на Македонија."