IT5 Cased Small Efficient EL32 for Informal-to-formal Style Transfer 🧐

Shout-out to Stefan Schweter for contributing the pre-trained efficient model!

This repository contains the checkpoint for the IT5 Cased Small Efficient EL32 model fine-tuned on Informal-to-formal style transfer on the Italian subset of the XFORMAL dataset as part of the experiments of the paper IT5: Large-scale Text-to-text Pretraining for Italian Language Understanding and Generation by Gabriele Sarti and Malvina Nissim.

Efficient IT5 models differ from the standard ones by adopting a different vocabulary that enables cased text generation and an optimized model architecture to improve performances while reducing parameter count. The Small-EL32 replaces the original encoder from the T5 Small architecture with a 32-layer deep encoder, showing improved performances over the base model.

A comprehensive overview of other released materials is provided in the gsarti/it5 repository. Refer to the paper for additional details concerning the reported scores and the evaluation approach.

Using the model

Model checkpoints are available for usage in Tensorflow, Pytorch and JAX. They can be used directly with pipelines as:

from transformers import pipelines

i2f = pipeline("text2text-generation", model='it5/it5-efficient-small-el32-informal-to-formal')
i2f("nn capisco xke tt i ragazzi lo fanno")
>>> [{"generated_text": "non comprendo perché tutti i ragazzi agiscono così"}]

or loaded using autoclasses:

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

tokenizer = AutoTokenizer.from_pretrained("it5/it5-efficient-small-el32-informal-to-formal")
model = AutoModelForSeq2SeqLM.from_pretrained("it5/it5-efficient-small-el32-informal-to-formal")

If you use this model in your research, please cite our work as:

@article{sarti-nissim-2022-it5,
    title={{IT5}: Large-scale Text-to-text Pretraining for Italian Language Understanding and Generation},
    author={Sarti, Gabriele and Nissim, Malvina},
    journal={ArXiv preprint 2203.03759},
    url={https://arxiv.org/abs/2203.03759},
    year={2022},
	month={mar}
}

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 10.0

Framework versions

• Transformers 4.15.0
• Pytorch 1.10.0+cu102
• Datasets 1.17.0
• Tokenizers 0.10.3