BART-IT: Italian pretraining for BART sequence to sequence model
BART-IT is a sequence-to-sequence model, based on the BART architecture that is specifically tailored to the Italian language. The model is pre-trained on a large corpus of Italian text, and can be fine-tuned on a variety of tasks.
Model description
The model is a base-sized BART model, with a vocabulary size of 52,000 tokens. It has 140M parameters and can be used for any task that requires a sequence-to-sequence model. It is trained from scratch on a large corpus of Italian text, and can be fine-tuned on a variety of tasks.
Pre-training
The code used to pre-train BART-IT together with additional information on model parameters can be found here.
Fine-tuning
The model in this repository is a pre-trained model without any fine-tuning. In order to use the model for a specific task, you can fine-tune it on a specific dataset.
The model has been fine-tuned for the abstractive summarization task on 3 different Italian datasets:
Usage
In order to use the model, you can use the following code:
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
tokenizer = AutoTokenizer.from_pretrained("morenolq/bart-it")
model = AutoModelForSeq2SeqLM.from_pretrained("morenolq/bart-it")
input_ids = tokenizer.encode("Il modello BART-IT è stato pre-addestrato su un corpus di testo italiano", return_tensors="pt")
outputs = model.generate(input_ids, max_length=40, num_beams=4, early_stopping=True)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
Citation
If you find this model useful for your research, please cite the following paper:
@Article{BARTIT,
AUTHOR = {La Quatra, Moreno and Cagliero, Luca},
TITLE = {BART-IT: An Efficient Sequence-to-Sequence Model for Italian Text Summarization},
JOURNAL = {Future Internet},
VOLUME = {15},
YEAR = {2023},
NUMBER = {1},
ARTICLE-NUMBER = {15},
URL = {https://www.mdpi.com/1999-5903/15/1/15},
ISSN = {1999-5903},
DOI = {10.3390/fi15010015}
}