gonzalez-agirre's picture
Update README.md
7353ca5
metadata
language: ca
license: apache-2.0
tags:
  - catalan
  - masked-lm
  - distilroberta
widget:
  - text: El Català és una llengua molt <mask>.
  - text: Salvador Dalí va viure a <mask>.
  - text: La Costa Brava  les millors <mask> d'Espanya.
  - text: El cacaolat és un batut de <mask>.
  - text: <mask> és la capital de la Garrotxa.
  - text: Vaig al <mask> a buscar bolets.
  - text: Antoni Gaudí vas ser un <mask> molt important per la ciutat.
  - text: Catalunya és una referència en <mask> a nivell europeu.
pipeline_tag: fill-mask

DistilRoBERTa-base-ca-v2

Table of Contents

Click to expand

Model description

This model is a distilled version of projecte-aina/roberta-base-ca-v2. It follows the same training procedure as DistilBERT, using the implementation of Knowledge Distillation from the paper's official repository.

The resulting architecture consists of 6 layers, 768 dimensional embeddings and 12 attention heads. This adds up to a total of 82M parameters, which is considerably less than the 125M of standard RoBERTa-base models. This makes the model lighter and faster than the original, at the cost of slightly lower performance.

We encourage users of this model to check out the projecte-aina/roberta-base-ca-v2 model card to learn more details about the teacher model.

Intended uses and limitations

This model is ready-to-use only for masked language modeling (MLM) to perform the Fill-Mask task. However, it is intended to be fine-tuned on non-generative downstream tasks such as Question Answering, Text Classification or Named Entity Recognition.

How to use

Usage example where the model is passed to a fill-mask pipeline to predict the masked word (<mask>) from a given text.

from pprint import pprint
from transformers import pipeline
pipe = pipeline("fill-mask", model="projecte-aina/distilroberta-base-ca-v2")
text = "El <mask> és el meu dia preferit de la setmana."
pprint(pipe(text))
[{'score': 0.2531125545501709,
  'sequence': ' El dilluns és el meu dia preferit de la setmana.',
  'token': 2885,
  'token_str': ' dilluns'},
 {'score': 0.13626143336296082,
  'sequence': ' El divendres és el meu dia preferit de la setmana.',
  'token': 2539,
  'token_str': ' divendres'},
 {'score': 0.11026635020971298,
  'sequence': ' El dijous és el meu dia preferit de la setmana.',
  'token': 2868,
  'token_str': ' dijous'},
 {'score': 0.10040736198425293,
  'sequence': ' El dissabte és el meu dia preferit de la setmana.',
  'token': 2480,
  'token_str': ' dissabte'},
 {'score': 0.09762872755527496,
  'sequence': ' El diumenge és el meu dia preferit de la setmana.',
  'token': 2587,
  'token_str': ' diumenge'}]

Limitations and bias

At the time of submission, no measures have been taken to estimate the bias embedded in the model. However, we are well aware that our models may be biased since the corpora have been collected using crawling techniques on multiple web sources. We intend to conduct research in these areas in the future, and if completed, this model card will be updated.

Training

Training data

The training corpus consists of several corpora gathered from web crawling and public corpora, as shown in the table below:

Corpus Size (GB)
Catalan Crawling 13.00
RacoCatalá 8.10
Catalan Oscar 4.00
CaWaC 3.60
Cat. General Crawling 2.50
Wikipedia 1.10
DOGC 0.78
Padicat 0.63
ACN 0.42
Nació Digital 0.42
Cat. Government Crawling 0.24
Vilaweb 0.06
Catalan Open Subtitles 0.02
Tweets 0.02

Training procedure

This model has been trained using a technique known as Knowledge Distillation, which is used to shrink networks to a reasonable size while minimizing the loss in performance.

It basically consists in distilling a large language model (the teacher) into a more lightweight, energy-efficient, and production-friendly model (the student).

So, in a “teacher-student learning” setup, a relatively small student model is trained to mimic the behavior of a larger teacher model. As a result, the student has lower inference time and the ability to run in commodity hardware.

Evaluation

CLUB benchmark

This model has been fine-tuned on the downstream tasks of the Catalan Language Understanding Evaluation benchmark (CLUB), which includes the following datasets:

Dataset Task Total Train Dev Test
AnCora NER 13,581 10,628 1,427 1,526
AnCora POS 16,678 13,123 1,709 1,846
STS-ca STS 3,073 2,073 500 500
TeCla TC 137,775 110,203 13,786 13,786
TE-ca RTE 21,163 16,930 2,116 2,117
CatalanQA QA 21,427 17,135 2,157 2,135
XQuAD-ca QA - - - 1,189

Evaluation results

This is how it compares to its teacher when fine-tuned on the aforementioned downstream tasks:

Model \ Task NER (F1) POS (F1) STS-ca (Comb.) TeCla (Acc.) TEca (Acc.) CatalanQA (F1/EM) XQuAD-ca 1 (F1/EM)
RoBERTa-base-ca-v2 89.29 98.96 79.07 74.26 83.14 89.50/76.63 73.64/55.42
DistilRoBERTa-base-ca 87.88 98.83 77.26 73.20 76.00 84.07/70.77 62.93/45.08

1 : Trained on CatalanQA, tested on XQuAD-ca.

Additional information

Authors

Language Technologies Unit at Barcelona Supercomputing Center (langtech@bsc.es).

Contact information

For further information, send an email to aina@bsc.es.

Copyright

Copyright by the Language Technologies Unit at Barcelona Supercomputing Center.

Licensing information

This work is licensed under a Apache License, Version 2.0.

Funding

This work was funded by the Departament de la Vicepresidència i de Polítiques Digitals i Territori de la Generalitat de Catalunya within the framework of Projecte AINA.

Citation information

There is no publication for this specific model, but you can cite the paper where the teacher model was presented:

@inproceedings{armengol-estape-etal-2021-multilingual,
    title = "Are Multilingual Models the Best Choice for Moderately Under-resourced Languages? {A} Comprehensive Assessment for {C}atalan",
    author = "Armengol-Estap{\'e}, Jordi  and
      Carrino, Casimiro Pio  and
      Rodriguez-Penagos, Carlos  and
      de Gibert Bonet, Ona  and
      Armentano-Oller, Carme  and
      Gonzalez-Agirre, Aitor  and
      Melero, Maite  and
      Villegas, Marta",
    booktitle = "Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021",
    month = aug,
    year = "2021",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.findings-acl.437",
    doi = "10.18653/v1/2021.findings-acl.437",
    pages = "4933--4946",
}

Disclaimer

Click to expand

The models published in this repository are intended for a generalist purpose and are available to third parties. These models may have bias and/or any other undesirable distortions.

When third parties, deploy or provide systems and/or services to other parties using any of these models (or using systems based on these models) or become users of the models, they should note that it is their responsibility to mitigate the risks arising from their use and, in any event, to comply with applicable regulations, including regulations regarding the use of Artificial Intelligence.

In no event shall the owner and creator of the models (BSC) be liable for any results arising from the use made by third parties of these models.