RoBERTweetTurkCovid

RoBERTweetTurkCovid (uncased)

Pretrained model on Turkish language using a masked language modeling (MLM) objective. The model is uncased. The pretrained corpus is a Turkish tweets collection related to COVID-19.

Model architecture is similar to RoBERTa-base (12 layers, 12 heads, and 768 hidden size). Tokenization algorithm is WordPiece. Vocabulary size is 30k.

The details of pretraining can be found at this paper:

@InProceedings{clef-checkthat:2022:task1:oguzhan,
author = {Cagri Toraman and Oguzhan Ozcelik and Furkan Şahinuç and Umitcan Sahin},
title = "{ARC-NLP at CheckThat! 2022:} Contradiction for Harmful Tweet Detection",
year = {2022},
booktitle = "Working Notes of {CLEF} 2022 - Conference and Labs of the Evaluation Forum",
editor = {Faggioli, Guglielmo andd Ferro, Nicola and Hanbury, Allan and Potthast, Martin},
series = {CLEF~'2022},
address = {Bologna, Italy},
}

The following code can be used for model loading and tokenization, example max length (768) can be changed:

	model = AutoModel.from_pretrained([model_path])
	#for sequence classification:
	#model = AutoModelForSequenceClassification.from_pretrained([model_path], num_labels=[num_classes])

	tokenizer = PreTrainedTokenizerFast(tokenizer_file=[file_path])
	tokenizer.mask_token = "[MASK]"
	tokenizer.cls_token = "[CLS]"
	tokenizer.sep_token = "[SEP]"
	tokenizer.pad_token = "[PAD]"
	tokenizer.unk_token = "[UNK]"
	tokenizer.bos_token = "[CLS]"
	tokenizer.eos_token = "[SEP]"
	tokenizer.model_max_length = 768

BibTeX entry and citation info

@InProceedings{clef-checkthat:2022:task1:oguzhan,
author = {Cagri Toraman and Oguzhan Ozcelik and Furkan Şahinuç and Umitcan Sahin},
title = "{ARC-NLP at CheckThat! 2022:} Contradiction for Harmful Tweet Detection",
year = {2022},
booktitle = "Working Notes of {CLEF} 2022 - Conference and Labs of the Evaluation Forum",
editor = {Faggioli, Guglielmo andd Ferro, Nicola and Hanbury, Allan and Potthast, Martin},
series = {CLEF~'2022},
address = {Bologna, Italy},
}