plantl-gob-es/lm-spanish

Official source for spanish Language Models and resources made @ BSC-TEMU within the "Plan de las Tecnologías del Lenguaje" (Plan-TL).

Spanish Language Models 💃🏻

A repository part of the MarIA project.

Corpora 📃

Corpora	Number of documents	Number of tokens	Size (GB)
BNE	201,080,084	135,733,450,668	570GB

Models 🤖

• ✨ new ✨ Ǎguila-7B: https://huggingface.co/projecte-aina/aguila-7b

  A 7B parameters LLM that has been trained on a mixture of Spanish, Catalan and English data, adding up to a total of 26B tokens. It uses the Falcon-7b model as a starting point, a state-of-the-art English language model that was openly released just a few months ago by the Technology Innovation Institute. Read more here

• RoBERTa-base BNE: https://huggingface.co/PlanTL-GOB-ES/roberta-base-bne

• RoBERTa-large BNE: https://huggingface.co/PlanTL-GOB-ES/roberta-large-bne

  Transformer-based masked language models for the Spanish language. They are based on the RoBERTa large model and has been pre-trained using the largest Spanish corpus known to date, with a total of 570GB of clean and deduplicated text processed for this work, compiled from the web crawlings performed by the National Library of Spain (Biblioteca Nacional de España) from 2009 to 2019.

• longformer-base-4096-bne-es: https://huggingface.co/PlanTL-GOB-ES/longformer-base-4096-bne-es

  The Longformer version of the roberta-base-ca-v2 masked language model for the Catalan language. The use of these models allows us to process larger contexts (up to 4096 tokens) as input without the need of additional aggregation strategies. The pretraining process of this model started from the roberta-base-ca-v2 checkpoint and was pretrained for MLM on both short and long documents in Catalan.

• GPT2-base BNE: https://huggingface.co/PlanTL-GOB-ES/gpt2-base-bne

• GPT2-large BNE: https://huggingface.co/PlanTL-GOB-ES/gpt2-large-bne

  Transformer-based model for the Spanish language. They are based on the GPT-2 model and has been pre-trained using the largest Spanish corpus known to date, with a total of 570GB of clean and deduplicated text processed for this work, compiled from the web crawlings performed by the National Library of Spain (Biblioteca Nacional de España) from 2009 to 2019.

See results achieved on several tasks below. Vegeu els resultats obtinguts en diverses tasques més avall.

Usage example ⚗️

For the RoBERTa-base

from transformers import AutoModelForMaskedLM
from transformers import AutoTokenizer, FillMaskPipeline
from pprint import pprint
tokenizer_hf = AutoTokenizer.from_pretrained('PlanTL-GOB-ES/roberta-base-bne')
model = AutoModelForMaskedLM.from_pretrained('PlanTL-GOB-ES/roberta-base-bne')
model.eval()
pipeline = FillMaskPipeline(model, tokenizer_hf)
text = f"¡Hola <mask>!"
res_hf = pipeline(text)
pprint([r['token_str'] for r in res_hf])

For the RoBERTa-large

from transformers import AutoModelForMaskedLM
from transformers import AutoTokenizer, FillMaskPipeline
from pprint import pprint
tokenizer_hf = AutoTokenizer.from_pretrained('PlanTL-GOB-ES/roberta-large-bne')
model = AutoModelForMaskedLM.from_pretrained('PlanTL-GOB-ES/roberta-large-bne')
model.eval()
pipeline = FillMaskPipeline(model, tokenizer_hf)
text = f"¡Hola <mask>!"
res_hf = pipeline(text)
pprint([r['token_str'] for r in res_hf])

Fine-tunned models 🧗🏼‍♀️🏇🏼🤽🏼‍♀️🏌🏼‍♂️🏄🏼‍♀️

• RoBERTa-base-BNE for Capitel-POS: https://huggingface.co/PlanTL-GOB-ES/roberta-base-bne-capitel-pos
• RoBERTa-large-BNE for Capitel-POS: https://huggingface.co/PlanTL-GOB-ES/roberta-large-bne-capitel-pos
• RoBERTa-base-BNE for Capitel-NER: https://huggingface.co/PlanTL-GOB-ES/roberta-base-bne-capitel-ner
• RoBERTa-base-BNE for Capitel-NER: https://huggingface.co/PlanTL-GOB-ES/roberta-base-bne-capitel-ner-plus (very robust)
• RoBERTa-large-BNE for Capitel-NER: https://huggingface.co/PlanTL-GOB-ES/roberta-large-bne-capitel-ner
• RoBERTa-base-BNE for SQAC: https://huggingface.co/PlanTL-GOB-ES/roberta-base-bne-sqac
• RoBERTa-large-BNE for SQAC: https://huggingface.co/PlanTL-GOB-ES/roberta-large-bne-sqac

For a complete list, refer to https://huggingface.co/PlanTL-GOB-ES

Other Spanish Language Models 👩‍👧‍👦

Domain-specific language models:

• ⚖️ Legal Language Model: https://github.com/PlanTL-GOB-ES/lm-legal-es
• ⚕️ Biomedical and Clinical Language Models: https://github.com/PlanTL-GOB-ES/lm-biomedical-clinical-es

Word embeddings 🔤

• Spanish CBOW Word Embeddings in Floret, trained with the corpus from the National Library of Spain (Biblioteca Nacional de España or BNE) using floret: https://zenodo.org/record/7314098
• Biomedical Spanish CBOW Word Embeddings in Floret, trained with a biomedical Spanish corpus using floret: https://zenodo.org/record/7314041
• Spanish Skip-Gram Word Embeddings in FastText, trained with the corpus from the BNE: https://zenodo.org/record/5046525
• Spanish Legal Domain Word & Sub-Word Embeddings, trained with a Spanish Legal resources: https://zenodo.org/record/5036147

spaCy models

• es_cantemist_ner_trf, basic Spacy BioNER pipeline, with a RoBERTa-based model [bsc-bio-ehr-es] (https://huggingface.co/PlanTL-GOB-ES/bsc-bio-ehr-es) and a dataset, CANTEMIST, annotated with tumour morphology entities: https://huggingface.co/PlanTL-GOB-ES/es_cantemist_ner_trf
• es_pharmaconer_ner_trf, basic Spacy BioNER pipeline, with a RoBERTa-based model [bsc-bio-ehr-es] (https://huggingface.co/PlanTL-GOB-ES/bsc-bio-ehr-es) and a dataset, Pharmaconer, a NER dataset annotated with substances, compounds and proteins entities: https://huggingface.co/PlanTL-GOB-ES/es_pharmaconer_ner_trf

Datasets 🗂️

• Spanish Question Answering Corpus (SQAC)🦆: https://huggingface.co/datasets/PlanTL-GOB-ES/SQAC
• Spanish Semantic Text Similarity (STS-es): https://huggingface.co/datasets/PlanTL-GOB-ES/sts-es
• Professional translation into Spanish of Winograd NLI dataset (WNLI-es): https://huggingface.co/datasets/PlanTL-GOB-ES/wnli-es
• Spanish dataset of the CoNLL-2002 Shared Task (CoNLL-NERC): https://huggingface.co/datasets/PlanTL-GOB-ES/CoNLL-NERC-es
• Spanish corpus for thematic Text Classification tasks (WikiCAT_es): https://huggingface.co/datasets/PlanTL-GOB-ES/WikiCAT_esv2
• English corpus for thematic Text Classification tasks (WikiCAT_en): https://huggingface.co/datasets/PlanTL-GOB-ES/WikiCAT_en

For a complete list, refer to https://huggingface.co/PlanTL-GOB-ES

EvalES: The Spanish Evaluation Benchmark 🏆

The EvalES benchmark consists of 10 tasks: Named Entity Recognition and Classification (CoNLL-NERC and CAPITEL-NERC), Part-of-Speech Tagging (UD-POS and CAPITEL-POS ), Text Classification (MLDoc), Paraphrase Identification (PAWS-X), Semantic Textual Similarity (STS), Question Answering (SQAC), Textual Entailment (XNLI) and Massive.

Results ✅

Dataset	Metric	RoBERTa-b	RoBERTa-l	BETO*	mBERT	BERTIN**	Electricidad***
MLDoc	F1	0.9664	0.9702	0.9714🔥	0.9617	0.9668	0.9565
CoNLL-NERC	F1	0.8851🔥	0.8823	0.8759	0.8691	0.8835	0.7954
CAPITEL-NERC	F1	0.8960	0.9051🔥	0.8772	0.8810	0.8856	0.8035
PAWS-X	F1	0.9020	0.9150🔥	0.8930	0.9000	0.8965	0.9045
UD-POS	F1	0.9907🔥	0.9904	0.9900	0.9886	0.9898	0.9818
CAPITEL-POS	F1	0.9846	0.9856🔥	0.9836	0.9839	0.9847	0.9816
SQAC	F1	0.7923	0.8202🔥	0.7923	0.7562	0.7678	0.7383
STS	Combined	0.8533🔥	0.8411	0.8159	0.8164	0.7945	0.8063
XNLI	Accuracy	0.8016	0.8263🔥	0.8130	0.7876	0.7890	0.7878
Massive	Accuracy	0.8605	0.8722	0.8732🔥	0.8504	0.8500	0.8517

* A model based on BERT architecture.

** A model based on RoBERTa architecture.

*** A model based on Electra architecture.

For more information, refer to https://benchmark.plantl.bsc.es/

Demos

• Anonimizador: Anonymizer for Spanish and Catalan user generated content in conversational systems: https://anonimizador.plantl.bsc.es/
• spaCy: Demo of the different spacy models for Spanish: https://spacy.plantl.bsc.es/
• QA: Question/Answer system in the Spanish Wikipedia based on models and datasets generated under PlanTL: https://qa.plantl.bsc.es/
• Traductor: Automatic translators between Spanish and Catalan and between Spanish and Galician: https://traductor.plantl.bsc.es/
• EvalES: Collection of resources for assessing natural language comprehension systems: https://benchmark.plantl.bsc.es/

Cite 📣

@article{gutierrezfandino2022,
	author = {Asier Gutiérrez-Fandiño and Jordi Armengol-Estapé and Marc Pàmies and Joan Llop-Palao and Joaquin Silveira-Ocampo and Casimiro Pio Carrino and Carme Armentano-Oller and Carlos Rodriguez-Penagos and Aitor Gonzalez-Agirre and Marta Villegas},
	title = {MarIA: Spanish Language Models},
	journal = {Procesamiento del Lenguaje Natural},
	volume = {68},
	number = {0},
	year = {2022},
	issn = {1989-7553},
	url = {http://journal.sepln.org/sepln/ojs/ojs/index.php/pln/article/view/6405},
	pages = {39--60}
}

Contact 📧

📋 We are interested in (1) extending our corpora to make larger models (2) train/evaluate the model in other tasks.

For questions regarding this work, contact [email protected]

Disclaimer

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 of the models (SEDIA – State Secretariat for digitalization and artificial intelligence) nor the creator (BSC – Barcelona Supercomputing Center) be liable for any results arising from the use made by third parties of these models.

Los modelos publicados en este repositorio tienen una finalidad generalista y están a disposición de terceros. Estos modelos pueden tener sesgos y/u otro tipo de distorsiones indeseables.

Cuando terceros desplieguen o proporcionen sistemas y/o servicios a otras partes usando alguno de estos modelos (o utilizando sistemas basados en estos modelos) o se conviertan en usuarios de los modelos, deben tener en cuenta que es su responsabilidad mitigar los riesgos derivados de su uso y, en todo caso, cumplir con la normativa aplicable, incluyendo la normativa en materia de uso de inteligencia artificial.

En ningún caso el propietario de los modelos (SEDIA – Secretaría de Estado de Digitalización e Inteligencia Artificial) ni el creador (BSC – Barcelona Supercomputing Center) serán responsables de los resultados derivados del uso que hagan terceros de estos modelos.

lm-spanish's Issues

Agila dataset availability

Hi! Thanks for all the work!

I would like to ask if the dataset used for the training of the agila-7b model is available for testing/downloading.

Potential issues with HF GPT2 Models

Hello,

I am using the GPT2 models available in HF, and running into a few issues. Firstly, there seems to be an issue with the tokenizer. Trying to calculate perplexity using the evaluate module, as follows:

from evaluate import load
perplexity = load("perplexity", module_type="metric")
results = perplexity.compute(predictions=["Hola, como estas?"], model_id="PlanTL-GOB-ES/gpt2-base-bne", device="cpu")

Gives the following error:

 ...
  File "/ikerlariak/aormazabal024/PhD/Poetry-Generation/demo/poetry-env-traganarru/lib/python3.8/site-packages/torch/nn/functional.py", line 2199, in embedding
    return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
IndexError: index out of range in self`

This seems to be related to the special tokens for <pad>, <s>, </s> and <unk> not being properly set (but are used by the evaluate module), as the only special token added in the tokenizer is <|endoftext|>. One can manually fix it for the local snapshot:

tokenizer.pad_token = '<pad>'
tokenizer.bos_token = '</s>'
tokenizer.eos_token = '</s>'
tokenizer.unk_token = '<unk>'
tokenizer.save_pretrained('[snapshot-path]')

However, even after fixing this, I am getting quite high perplexities compared to the 10-13 reported in the paper for all sentences I am trying (assuming per-word-perplexity is reported). Is it possible there was an issue when converting from fairseq to HF, and are the original fairseq models available somewhere to compare? Or maybe I am making a mistake when calculating the ppl, was there any tokenization done to the text apart from BPE (i.e. replacing newlines with , which is pretty standard in fairseq)?

GPT-2 state and GPT-j-6B

I would like to ask about the state of the GPT-2 model. Will it arrive soon at huggingface?

I would also like to ask if you have the intention of train GPT-j-6B. Training this model for some people would be impossible due to its hardware requirements, but you have Mare Nostrum, the dataset and the previous version GPT-2.

Dataset Availability

Hi!

Thanks for the work, I love it. I was wandering if the final clean corpus is available somewhere.

Tokenizer rare subwords

Hi, First at all many thanks for this amazing work in Spanish 😍.

from transformers import AutoModelForMaskedLM
from transformers import AutoTokenizer

model_checkpoint = "PlanTL-GOB-ES/roberta-large-bne"
model = AutoModelForMaskedLM.from_pretrained(model_checkpoint)
tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)

rare_tokens_vocab = [word for word in tokenizer.vocab if 'Ġ' in word]
print(len(rare_tokens_vocab))
# 37695 50262

Almost 75% or tokens contains "Ġ" char. It is really strange! Probably due to dirty text in the corpus ?? or why is the reason of so many tokens with "Ġ"??