Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

https://github.com/sdadas/polish-nlp-resources?fbclid=IwAR0tv_YbuBwfFirGFQvqaGdCSL6BV-9pNW8Wm3gkGiyXNajE_M9tpY0hIaM

Pre-trained models and language resources for Natural Language Processing in Polish
https://github.com/sdadas/polish-nlp-resources?fbclid=IwAR0tv_YbuBwfFirGFQvqaGdCSL6BV-9pNW8Wm3gkGiyXNajE_M9tpY0hIaM

language-models lexicons machine-learning natural-language-processing polish polish-language word-embedding

Last synced: 2 months ago
JSON representation

Pre-trained models and language resources for Natural Language Processing in Polish

Lists

README 

        
# Polish NLP resources



This repository contains pre-trained models and language resources for Natural Language Processing in Polish created during my research. Some of the models are also available on **[Huggingface Hub](https://huggingface.co/sdadas)**.



If you'd like to use any of those resources in your research please cite:



```bibtex

@Misc{polish-nlp-resources,

  author =       {S{\l}awomir Dadas},

  title =        {A repository of Polish {NLP} resources},

  howpublished = {Github},

  year =         {2019},

  url =          {https://github.com/sdadas/polish-nlp-resources/}

}

```



## Contents



- [Word embeddings](#word-embeddings)

  - [Word2Vec](#word2vec)

  - [FastText](#fasttext)

  - [GloVe](#glove)

  - [High dimensional word vectors](#high-dimensional-word-vectors)

  - [Compressed Word2Vec](#compressed-word2vec)

  - [Wikipedia2Vec](#wikipedia2vec)

- [Language models](#language-models)

  - [ELMo](#elmo)

  - [RoBERTa](#roberta)

  - [BART](#bart)

  - [GPT-2](#gpt-2)

  - [Longformer](#longformer)

- [Text encoders](#text-encoders)

- [Machine translation models](#machine-translation-models)

  - [Convolutional models for Fairseq](#convolutional-models-for-fairseq)

  - [T5-based models](#t5-based-models)

- [Fine-tuned models](#fine-tuned-models)

- [Dictionaries and lexicons](#dictionaries-and-lexicons)

- [Links to external resources](#links-to-external-resources)

  - [Repositories of linguistic tools and resources](#repositories-of-linguistic-tools-and-resources)

  - [Publicly available large Polish text corpora (> 1GB)](#publicly-available-large-polish-text-corpora--1gb)

  - [Models supporting Polish language](#models-supporting-polish-language)



## Word embeddings



The following section includes pre-trained word embeddings for Polish. Each model was trained on a corpus consisting of Polish Wikipedia dump, Polish books and articles, 1.5 billion tokens at total. 



### Word2Vec



Word2Vec trained with [Gensim](https://radimrehurek.com/gensim/). 100 dimensions, negative sampling, contains lemmatized words with 3 or more ocurrences in the corpus and additionally a set of pre-defined punctuation symbols, all numbers from 0 to 10'000, Polish forenames and lastnames. The archive contains embedding in gensim binary format. Example of usage:



```python

from gensim.models import KeyedVectors



if __name__ == '__main__':

    word2vec = KeyedVectors.load("word2vec_polish.bin")

    print(word2vec.similar_by_word("bierut"))

    

# [('cyrankiewicz', 0.818274736404419), ('gomułka', 0.7967918515205383), ('raczkiewicz', 0.7757788896560669), ('jaruzelski', 0.7737460732460022), ('pużak', 0.7667238712310791)]

```



[Download (GitHub)](https://github.com/sdadas/polish-nlp-resources/releases/download/v1.0/word2vec.zip)



### FastText



FastText trained with [Gensim](https://radimrehurek.com/gensim/). Vocabulary and dimensionality is identical to Word2Vec model. The archive contains embedding in gensim binary format. Example of usage:



```python

from gensim.models import KeyedVectors



if __name__ == '__main__':

    word2vec = KeyedVectors.load("fasttext_100_3_polish.bin")

    print(word2vec.similar_by_word("bierut"))

    

# [('bieruty', 0.9290274381637573), ('gierut', 0.8921363353729248), ('bieruta', 0.8906412124633789), ('bierutow', 0.8795544505119324), ('bierutowsko', 0.839280366897583)]

```



[Download (OneDrive)](https://witedupl-my.sharepoint.com/:u:/g/personal/dadass_wit_edu_pl/EeoDV_cq0KtAupMa0E9iIlEBMTVvw4OzABbPuAxUMFD8EA?e=5naF5z)



### GloVe



Global Vectors for Word Representation (GloVe) trained using the reference implementation from Stanford NLP. 100 dimensions, contains lemmatized words with 3 or more ocurrences in the corpus. Example of usage:



```python

from gensim.models import KeyedVectors



if __name__ == '__main__':

    word2vec = KeyedVectors.load_word2vec_format("glove_100_3_polish.txt")

    print(word2vec.similar_by_word("bierut"))

    

# [('cyrankiewicz', 0.8335597515106201), ('gomułka', 0.7793121337890625), ('bieruta', 0.7118682861328125), ('jaruzelski', 0.6743760108947754), ('minc', 0.6692837476730347)]

```



[Download (GitHub)](https://github.com/sdadas/polish-nlp-resources/releases/download/v1.0/glove.zip)



### High dimensional word vectors

Pre-trained vectors using the same vocabulary as above but with higher dimensionality. These vectors are more suitable for representing larger chunks of text such as sentences or documents using simple word aggregation methods (averaging, max pooling etc.) as more semantic information is preserved this way.



**GloVe** - **300d:** [Part 1 (GitHub)](https://github.com/sdadas/polish-nlp-resources/releases/download/glove-hd/glove_300_3_polish.zip.001), **500d:** [Part 1 (GitHub)](https://github.com/sdadas/polish-nlp-resources/releases/download/glove-hd/glove_500_3_polish.zip.001) [Part 2 (GitHub)](https://github.com/sdadas/polish-nlp-resources/releases/download/glove-hd/glove_500_3_polish.zip.002), **800d:** [Part 1 (GitHub)](https://github.com/sdadas/polish-nlp-resources/releases/download/glove-hd/glove_800_3_polish.zip.001) [Part 2 (GitHub)](https://github.com/sdadas/polish-nlp-resources/releases/download/glove-hd/glove_800_3_polish.zip.002) [Part 3 (GitHub)](https://github.com/sdadas/polish-nlp-resources/releases/download/glove-hd/glove_800_3_polish.zip.003) 



**Word2Vec** - [300d (OneDrive)](https://witedupl-my.sharepoint.com/:u:/g/personal/dadass_wit_edu_pl/EQ7QA6PkPupBtZYyP8kaafMB0z9FdHfME7kxm_tcRWh9hA?e=RGekMu), 

[500d (OneDrive)](https://witedupl-my.sharepoint.com/:u:/g/personal/dadass_wit_edu_pl/EfBT7WvY7eVHuQZIuUpnJzsBXTN2L896ldVvhRBiCmUH_A?e=F0LGVc), [800d (OneDrive)](https://witedupl-my.sharepoint.com/:u:/g/personal/dadass_wit_edu_pl/Eda0vUkicNpNk4oMf2eoLZkBTJbMTmymKqqZ_yoEXw98TA?e=rKu4pP)



**FastText** - [300d (OneDrive)](https://witedupl-my.sharepoint.com/:u:/g/personal/dadass_wit_edu_pl/ESj0xTXmTK5Jhiocp5Oxt7IBUUmaEjczFWvQn17c2QNgcg?e=9aory9), 

[500d (OneDrive)](https://witedupl-my.sharepoint.com/:u:/g/personal/dadass_wit_edu_pl/EViVRrF38fJMv1ihX2ARDNEBFFOE-MLSDHCcMG49IqEcCQ?e=g36NJ7), [800d (OneDrive)](https://witedupl-my.sharepoint.com/:u:/g/personal/dadass_wit_edu_pl/ESHkEJ7jLGlHoAIKiYdL0NkB_Z8VJyFcEHx3TpE7L1kNFg?e=FkoBgA)



### Compressed Word2Vec



This is a compressed version of the Word2Vec embedding model described above. For compression, we used the method described in [Compressing Word Embeddings via Deep Compositional Code Learning](https://arxiv.org/abs/1711.01068) by Shu and Nakayama. Compressed embeddings are suited for deployment on storage-poor devices such as mobile phones. The model weights 38MB, only 4.4% size of the original Word2Vec embeddings. Although the authors of the article claimed that compressing with their method doesn't hurt model performance, we noticed a slight but acceptable drop of accuracy when using compressed version of embeddings. Sample decoder class with usage:



```python

import gzip

from typing import Dict, Callable

import numpy as np



class CompressedEmbedding(object):



    def __init__(self, vocab_path: str, embedding_path: str, to_lowercase: bool=True):

        self.vocab_path: str = vocab_path

        self.embedding_path: str = embedding_path

        self.to_lower: bool = to_lowercase

        self.vocab: Dict[str, int] = self.__load_vocab(vocab_path)

        embedding = np.load(embedding_path)

        self.codes: np.ndarray = embedding[embedding.files[0]]

        self.codebook: np.ndarray = embedding[embedding.files[1]]

        self.m = self.codes.shape[1]

        self.k = int(self.codebook.shape[0] / self.m)

        self.dim: int = self.codebook.shape[1]



    def __load_vocab(self, vocab_path: str) -> Dict[str, int]:

        open_func: Callable = gzip.open if vocab_path.endswith(".gz") else open

        with open_func(vocab_path, "rt", encoding="utf-8") as input_file:

            return {line.strip():idx for idx, line in enumerate(input_file)}



    def vocab_vector(self, word: str):

        if word == "": return np.zeros(self.dim)

        val: str = word.lower() if self.to_lower else word

        index: int = self.vocab.get(val, self.vocab[""])

        codes = self.codes[index]

        code_indices = np.array([idx * self.k + offset for idx, offset in enumerate(np.nditer(codes))])

        return np.sum(self.codebook[code_indices], axis=0)



if __name__ == '__main__':

    word2vec = CompressedEmbedding("word2vec_100_3.vocab.gz", "word2vec_100_3.compressed.npz")

    print(word2vec.vocab_vector("bierut"))

```



[Download (GitHub)](https://github.com/sdadas/polish-nlp-resources/releases/download/v1.0/compressed.zip)



### Wikipedia2Vec

[Wikipedia2Vec](https://wikipedia2vec.github.io/) is a toolkit for learning joint representations of words and Wikipedia entities. We share Polish embeddings learned using a modified version of the library in which we added lemmatization and fixed some issues regarding parsing wiki dumps for languages other than English. Embedding models are available in sizes from 100 to 800 dimensions. A simple example: 



```python

from wikipedia2vec import Wikipedia2Vec



wiki2vec = Wikipedia2Vec.load("wiki2vec-plwiki-100.bin")

print(wiki2vec.most_similar(wiki2vec.get_entity("Bolesław Bierut")))

# (, 1.0), (, 0.75790733), (, 0.7276504),

# (, 0.7081445), (, 0.7043667) [...]

```



Download embeddings: [100d](https://witedupl-my.sharepoint.com/:u:/g/personal/dadass_wit_edu_pl/Ee_DFnilujxCiHmfUjRsqzUBBPst44eyCtmAnpB-Tq-ykw?e=ZzWIuf), [300d](https://witedupl-my.sharepoint.com/:u:/g/personal/dadass_wit_edu_pl/EWBzb1a89YJJku3vzFPObTUB5wTNaqIsznKT2AaKSP6xDQ?e=hhxSf0), [500d](https://witedupl-my.sharepoint.com/:u:/g/personal/dadass_wit_edu_pl/ERYsJUEo_DlKpUBBV_A86-0BrHDB88TJGr--WtzbKxhfJg?e=BPjH80), [800d](https://witedupl-my.sharepoint.com/:u:/g/personal/dadass_wit_edu_pl/EQJT8QyrMLFEqtC_1ZdOI54BzOQXIlvoQIbhra9EuIoV7w?e=SLfQrI). 



## Language models



### ELMo



Embeddings from Language Models (ELMo) is a contextual embedding presented in [Deep contextualized word representations](https://arxiv.org/abs/1802.05365) by Peters et al. Sample usage with PyTorch below, for a more detailed instructions for integrating ELMo with your model please refer to the official repositories [github.com/allenai/bilm-tf](https://github.com/allenai/bilm-tf) (Tensorflow) and [github.com/allenai/allennlp](https://github.com/allenai/allennlp) (PyTorch).



```python

from allennlp.commands.elmo import ElmoEmbedder



elmo = ElmoEmbedder("options.json", "weights.hdf5")

print(elmo.embed_sentence(["Zażółcić", "gęślą", "jaźń"]))

```



[Download (GitHub)](https://github.com/sdadas/polish-nlp-resources/releases/download/v1.0/elmo.zip)



### RoBERTa



Language model for Polish based on popular transformer architecture. We provide weights for improved BERT language model introduced in [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://arxiv.org/pdf/1907.11692.pdf). We provide two RoBERTa models for Polish - base and large model. A summary of pre-training parameters for each model is shown in the table below. We release two version of the each model: one in the [Fairseq](https://github.com/pytorch/fairseq) format and the other in the [HuggingFace Transformers](https://github.com/huggingface/transformers) format. More information about the models can be found in a [separate repository](https://github.com/sdadas/polish-roberta).



Model

L / H / A*

Batch size

Update steps

Corpus size

Fairseq

Transformers



  RoBERTa (base)

  12 / 768 / 12

  8k

  125k

  ~20GB

  

  v0.9.0

  

  

  v3.4

  



  RoBERTa‑v2 (base)

  12 / 768 / 12

  8k

  400k

  ~20GB

  

  v0.10.1

  

  

  v4.4

  



  RoBERTa (large)

  24 / 1024 / 16

  30k

  50k

  ~135GB

  

  v0.9.0

  

  

  v3.4

  



  RoBERTa‑v2 (large)

  24 / 1024 / 16

  2k

  400k

  ~200GB

  

  v0.10.2

  

  

  v4.14

  



  

  

  DistilRoBERTa

  6 / 768 / 12

  1k

  10ep.

  ~20GB

  

  n/a

  

  

  v4.13

  



\* L - the number of encoder blocks, H - hidden size, A - the number of attention heads 



Example in Fairseq:



```python

import os

from fairseq.models.roberta import RobertaModel, RobertaHubInterface

from fairseq import hub_utils



model_path = "roberta_large_fairseq"

loaded = hub_utils.from_pretrained(

    model_name_or_path=model_path,

    data_name_or_path=model_path,

    bpe="sentencepiece",

    sentencepiece_vocab=os.path.join(model_path, "sentencepiece.bpe.model"),

    load_checkpoint_heads=True,

    archive_map=RobertaModel.hub_models(),

    cpu=True

)

roberta = RobertaHubInterface(loaded['args'], loaded['task'], loaded['models'][0])

roberta.eval()

roberta.fill_mask('Druga wojna światowa zakończyła się w  roku.', topk=1)

roberta.fill_mask('Ludzie najbardziej boją się .', topk=1)

#[('Druga wojna światowa zakończyła się w 1945 roku.', 0.9345270991325378, ' 1945')]

#[('Ludzie najbardziej boją się śmierci.', 0.14140743017196655, ' śmierci')]

```



It is recommended to use the above models, but it is still possible to download [our old model](https://github.com/sdadas/polish-nlp-resources/releases/download/roberta/roberta.zip), trained on smaller batch size (2K) and smaller corpus (15GB).



### BART



BART is a transformer-based sequence to sequence model trained with a denoising objective. Can be used for fine-tuning on prediction tasks, just like regular BERT, as well as various text generation tasks such as machine translation, summarization, paraphrasing etc. We provide a Polish version of BART base model, trained on a large corpus of texts extracted from Common Crawl (200+ GB). More information on the BART architecture can be found in [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/abs/1910.13461). Example in HugginFace Transformers:



```python

import os

from transformers import BartForConditionalGeneration, PreTrainedTokenizerFast



model_dir = "bart_base_transformers"

tok = PreTrainedTokenizerFast(tokenizer_file=os.path.join(model_dir, "tokenizer.json"))

model = BartForConditionalGeneration.from_pretrained(model_dir)

sent = "Drugaświatowa zakończyła się wroku kapitulacją hitlerowskich"

batch = tok(sent, return_tensors='pt')

generated_ids = model.generate(batch['input_ids'])

print(tok.batch_decode(generated_ids, skip_special_tokens=True))

# ['Druga wojna światowa zakończyła się w 1945 roku kapitulacją hitlerowskich Niemiec.']

```



Download for [Fairseq v0.10](https://github.com/sdadas/polish-nlp-resources/releases/download/bart-base/bart_base_fairseq.zip) or [HuggingFace Transformers v4.0](https://github.com/sdadas/polish-nlp-resources/releases/download/bart-base/bart_base_transformers.zip).



### GPT-2



GPT-2 is a unidirectional transformer-based language model trained with an auto-regressive objective, originally introduced in the [Language Models are Unsupervised Multitask Learners](https://d4mucfpksywv.cloudfront.net/better-language-models/language_models_are_unsupervised_multitask_learners.pdf) paper. The original English GPT-2 was released in four sizes differing by the number of parameters: Small (112M), Medium (345M), Large (774M), XL (1.5B). 



#### Models for Huggingface Transformers



We provide Polish GPT-2 models for Huggingface Transformers. The models have been trained using [Megatron-LM](https://github.com/NVIDIA/Megatron-LM) library and then converted to the Huggingface format. The released checkpoints support longer contexts than the original GPT-2 by OpenAI. Small and medium models support up to 2048 tokens, twice as many as GPT-2 models and the same as GPT-3. Large and XL models support up to 1536 tokens. Example in Transformers:



```python

from transformers import pipeline



generator = pipeline("text-generation",  model="sdadas/polish-gpt2-medium")

results = generator("Policja skontrolowała trzeźwość kierowców",

  max_new_tokens=1024,  do_sample=True, repetition_penalty = 1.2,

  num_return_sequences=1, num_beams=1,  temperature=0.95,top_k=50, top_p=0.95

)

print(results[0].get("generated_text"))

# Policja skontrolowała trzeźwość kierowców. Teraz policjanci przypominają kierowcom o zachowaniu 

# bezpiecznej odległości i środkach ostrożności związanych z pandemią. - Kierujący po spożyciu 

# alkoholu są bardziej wyczuleni na innych uczestników ruchu drogowego oraz mają większą skłonność 

# do brawury i ryzykownego zachowania zwłaszcza wobec pieszych. Dodatkowo nie zawsze pamiętają oni 

# zasady obowiązujących u nas przepisów prawa regulujących kwestie dotyczące odpowiedzialności [...]

```

[Small](https://huggingface.co/sdadas/polish-gpt2-small), [Medium](https://huggingface.co/sdadas/polish-gpt2-medium), [Large](https://huggingface.co/sdadas/polish-gpt2-large), and [XL](https://huggingface.co/sdadas/polish-gpt2-xl) models are available on the Huggingface Hub



#### Models for Fairseq



We provide Polish versions of the medium and large GPT-2 models trained using Fairseq library. Example in Fairseq:



```python

import os

from fairseq import hub_utils

from fairseq.models.transformer_lm import TransformerLanguageModel



model_dir = "gpt2_medium_fairseq"

loaded = hub_utils.from_pretrained(

    model_name_or_path=model_dir,

    checkpoint_file="model.pt",

    data_name_or_path=model_dir,

    bpe="hf_byte_bpe",

    bpe_merges=os.path.join(model_dir, "merges.txt"),

    bpe_vocab=os.path.join(model_dir, "vocab.json"),

    load_checkpoint_heads=True,

    archive_map=TransformerLanguageModel.hub_models()

)

model = hub_utils.GeneratorHubInterface(loaded["args"], loaded["task"], loaded["models"])

model.eval()

result = model.sample(

    ["Policja skontrolowała trzeźwość kierowców"],

    beam=5, sampling=True, sampling_topk=50, sampling_topp=0.95,

    temperature=0.95, max_len_a=1, max_len_b=100, no_repeat_ngram_size=3

)

print(result[0])

# Policja skontrolowała trzeźwość kierowców pojazdów. Wszystko działo się na drodze gminnej, między Radwanowem 

# a Boguchowem. - Około godziny 12.30 do naszego komisariatu zgłosił się kierowca, którego zaniepokoiło 

# zachowanie kierującego w chwili wjazdu na tą drogę. Prawdopodobnie nie miał zapiętych pasów - informuje st. asp. 

# Anna Węgrzyniak z policji w Brzezinach. Okazało się, że kierujący był pod wpływem alkoholu. [...]

```



Download [medium](https://github.com/sdadas/polish-nlp-resources/releases/download/gpt-2/gpt2_medium_fairseq.7z) or [large](https://github.com/sdadas/polish-nlp-resources/releases/download/gpt-2/gpt2_large_fairseq.7z) model for Fairseq v0.10.



### Longformer



One of the main constraints of standard Transformer architectures is the limitation on the number of input tokens. There are several known models that allow processing of long documents, one of the popular ones being Longformer, introduced in the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150). We provide base and large versions of Polish Longformer model. The models were initialized with Polish RoBERTa (v2) weights and then fine-tuned on a corpus of long documents, ranging from 1024 to 4096 tokens. Example in Huggingface Transformers:



```python

from transformers import pipeline

fill_mask = pipeline('fill-mask', model='sdadas/polish-longformer-base-4096')

fill_mask('Stolica oraz największe miasto Francji to .')

```

[Base](https://huggingface.co/sdadas/polish-longformer-base-4096) and [large](https://huggingface.co/sdadas/polish-longformer-large-4096) models are available on the Huggingface Hub



## Text encoders

The purpose of text encoders is to produce a fixed-length vector representation for chunks of text, such as sentences or paragraphs. These models are used in semantic search, question answering, document clustering, dataset augmentation, plagiarism detection, and other tasks which involve measuring semantic similarity or relatedness between text passages.



### Paraphrase mining and semantic textual similarity

We share two models based on the [Sentence-Transformers](https://www.sbert.net/) library, trained using distillation method described in the paper [Making Monolingual Sentence Embeddings Multilingual using Knowledge Distillation](https://arxiv.org/abs/2004.09813). A corpus of 100 million parallel Polish-English sentence pairs from the [OPUS](https://opus.nlpl.eu/) project was used to train the models. You can download them from the Hugginface Hub using the links below.



Student model

Teacher model

Download



  polish-roberta-base-v2

  paraphrase-distilroberta-base-v2

  st-polish-paraphrase-from-distilroberta



  polish-roberta-base-v2

  paraphrase-mpnet-base-v2

  st-polish-paraphrase-from-mpnet



A simple example in Sentence-Transformers library:



```python

from sentence_transformers import SentenceTransformer

from sentence_transformers.util import cos_sim



sentences = ["Bardzo lubię jeść słodycze.", "Uwielbiam zajadać się słodkościami."]

model = SentenceTransformer("sdadas/st-polish-paraphrase-from-mpnet")

results = model.encode(sentences, convert_to_tensor=True, show_progress_bar=False)

print(cos_sim(results[0], results[1]))

# tensor([[0.9794]], device='cuda:0')

```

### MMLW 

MMLW (muszę mieć lepszą wiadomość) is a set of text encoders trained using [multilingual knowledge distillation method](https://arxiv.org/abs/2004.09813) on a diverse corpus of 60 million Polish-English text pairs, which included both sentence and paragraph aligned translations. The encoders are available in [Sentence-Transformers](https://www.sbert.net/) format. We used a two-step process to train the models. In the first step, the encoders were initialized with Polish RoBERTa and multilingual E5 checkpoints, and then distilled utilising English BGE as a teacher model. The resulting models from the distillation step can be used as general-purpose embeddings with applications in various tasks such as text similarity, document clustering, or fuzzy deduplication. The second step involved fine-tuning the obtained models on [Polish MS MARCO](https://huggingface.co/datasets/clarin-knext/msmarco-pl) dataset with contrastrive loss. The second stage models are adapted specifically for information retrieval tasks.



We provide a total of ten text encoders, five distilled and five fine-tuned for information retrieval. In the table below, we present the details of the released models.



  Base models

  Stage 1: Distilled models

  Stage 2: Retrieval models



  Student model

  Teacher model

  PL-MTEB
Score

  Download

  PIRB
NDCG@10

  Download



  Encoders based on Polish RoBERTa



  polish-roberta-base-v2

  bge-base-en

  61.05

  mmlw-roberta-base

  56.38

  mmlw-retrieval-roberta-base



  polish-roberta-large-v2

  bge-large-en

  63.23

  mmlw-roberta-large

  58.46

  mmlw-retrieval-roberta-large



  Encoders based on Multilingual E5



  multilingual-e5-small

  bge-small-en

  55.84

  mmlw-e5-small

  52.34

  mmlw-retrieval-e5-small



  multilingual-e5-base

  bge-base-en

  59.71

  mmlw-e5-base

  56.09

  mmlw-retrieval-e5-base



  multilingual-e5-large

  bge-large-en

  61.17

  mmlw-e5-large

  58.30

  mmlw-retrieval-e5-large



Please note that the developed models require the use of specific prefixes and suffixes when encoding texts. For RoBERTa-based encoders, each query should be preceded by the prefix "zapytanie: ", and no prefix is needed for passages. For E5-based models, queries should be prefixed with "query: " and passages with "passage: ". An example of how to use the models: 



```python

from sentence_transformers import SentenceTransformer

from sentence_transformers.util import cos_sim



query_prefix = "zapytanie: "      # "zapytanie: " for roberta, "query: " for e5

answer_prefix = ""                # empty for roberta, "passage: " for e5

queries = [query_prefix + "Jak dożyć 100 lat?"]

answers = [

    answer_prefix + "Trzeba zdrowo się odżywiać i uprawiać sport.",

    answer_prefix + "Trzeba pić alkohol, imprezować i jeździć szybkimi autami.",

    answer_prefix + "Gdy trwała kampania politycy zapewniali, że rozprawią się z zakazem niedzielnego handlu."

]

model = SentenceTransformer("sdadas/mmlw-retrieval-roberta-base")

queries_emb = model.encode(queries, convert_to_tensor=True, show_progress_bar=False)

answers_emb = model.encode(answers, convert_to_tensor=True, show_progress_bar=False)



best_answer = cos_sim(queries_emb, answers_emb).argmax().item()

print(answers[best_answer])

# Trzeba zdrowo się odżywiać i uprawiać sport.

```



## Machine translation models



This section includes pre-trained machine translation models.



### Convolutional models for Fairseq



We provide Polish-English and English-Polish convolutional neural machine translation models trained using [Fairseq](https://github.com/pytorch/fairseq) sequence modeling toolkit. Both models were trained on a parallel corpus of more than 40 million sentence pairs taken from [Opus](http://opus.nlpl.eu/) collection. Example of usage (`fairseq`, `sacremoses` and `subword-nmt` python packages are required to run this example):



```python

from fairseq.models import BaseFairseqModel



model_path = "/polish-english/"

model = BaseFairseqModel.from_pretrained(

    model_name_or_path=model_path,

    checkpoint_file="checkpoint_best.pt",

    data_name_or_path=model_path,

    tokenizer="moses",

    bpe="subword_nmt",

    bpe_codes="code",

    cpu=True

)

print(model.translate(sentence="Zespół astronomów odkrył w konstelacji Panny niezwykłą planetę.", beam=5))

# A team of astronomers discovered an extraordinary planet in the constellation of Virgo.

```



**Polish-English convolutional model:** [Download (GitHub)](https://github.com/sdadas/polish-nlp-resources/releases/download/nmt-models-conv/polish-english-conv.zip) \

**English-Polish convolutional model:** [Download (GitHub)](https://github.com/sdadas/polish-nlp-resources/releases/download/nmt-models-conv/english-polish-conv.zip)



### T5-based models



We share MT5 and Flan-T5 models fine-tuned for Polish-English and English-Polish translation. The models were trained on 70 million sentence pairs from [OPUS](http://opus.nlpl.eu/). You can download them from the Hugginface Hub using the links below. An example of how to use the models:



```python

from transformers import pipeline

generator = pipeline("translation", model="sdadas/flan-t5-base-translator-en-pl")

sentence = "A team of astronomers discovered an extraordinary planet in the constellation of Virgo."

print(generator(sentence, max_length=512))

# [{'translation_text': 'Zespół astronomów odkrył niezwykłą planetę w gwiazdozbiorze Panny.'}]

```



The following models are available on the Huggingface Hub: [mt5-base-translator-en-pl](https://huggingface.co/sdadas/mt5-base-translator-en-pl), [mt5-base-translator-pl-en](https://huggingface.co/sdadas/mt5-base-translator-pl-en), [flan-t5-base-translator-en-pl](https://huggingface.co/sdadas/flan-t5-base-translator-en-pl)



## Fine-tuned models



### ByT5-text-correction



A small multilingual utility model intended for simple text correction. It is designed to improve the quality of texts from the web, often lacking punctuation or proper word capitalization. The model was trained to perform three types of corrections: restoring punctuation in sentences, restoring word capitalization, and restoring diacritical marks for languages that include them.



The following languages are supported: Belarusian (be), Danish (da), German (de), Greek (el), English (en), Spanish (es), French (fr), Italian (it), Dutch (nl), Polish (pl), Portuguese (pt), Romanian (ro), Russian (ru), Slovak (sk), Swedish (sv), Ukrainian (uk). The model takes as input a sentence preceded by a language code prefix. For example:



```python

from transformers import pipeline

generator = pipeline("text2text-generation", model="sdadas/byt5-text-correction")

sentences = [

    " ciekaw jestem na co licza onuce stawiajace na sykulskiego w nadziei na zwrot ku rosji",

    " die frage die sich die europäer stellen müssen lautet ist es in unserem interesse die krise auf taiwan zu beschleunigen",

    " при своём рождении 26 августа 1910 года тереза получила имя агнес бояджиу"

]

generator(sentences, max_length=512)

# Ciekaw jestem na co liczą onuce stawiające na Sykulskiego w nadziei na zwrot ku Rosji.

# Die Frage, die sich die Europäer stellen müssen, lautet: Ist es in unserem Interesse, die Krise auf Taiwan zu beschleunigen?

# При своём рождении 26 августа 1910 года Тереза получила имя Агнес Бояджиу.

```



The model is available on the Huggingface Hub: [byt5-text-correction](https://huggingface.co/sdadas/byt5-text-correction)



## Dictionaries and lexicons



### Polish, English and foreign person names



This lexicon contains 346 thousand forenames and lastnames labeled as Polish, English or Foreign (other) crawled from multiple Internet sources.

Possible labels are: `P-N` (Polish forename), `P-L` (Polish lastname), `E-N` (English forename), `E-L` (English lastname), `F` (foreign / other). 

For each word, there is an additional flag indicating whether this name is also used as a common word in Polish (`C` for common, `U` for uncommon).



[Download (GitHub)](lexicons/names)



### Named entities extracted from SJP.PL



This dictionary consists mostly of the names of settlements, geographical regions, countries, continents and words derived from them (relational adjectives and inhabitant names). 

Besides that, it also contains names of popular brands, companies and common abbreviations of institutions' names.

This resource was created in a semi-automatic way, by extracting the words and their forms from SJP.PL using a set of heuristic rules and then manually filtering out words that weren't named entities.



[Download (GitHub)](lexicons/named-sjp)



## Links to external resources



### Repositories of linguistic tools and resources



- [Computational Linguistics in Poland - IPI PAN](http://clip.ipipan.waw.pl/LRT)

- [G4.19 Research Group, Wroclaw University of Technology](http://nlp.pwr.wroc.pl/narzedzia-i-zasoby)

- [CLARIN - repository of linguistic resources](https://clarin-pl.eu/dspace/)

- [Gonito.net - evaluation platform with some challenges for Polish](https://gonito.net)

- [Awesome NLP Polish (ksopyla)](https://github.com/ksopyla/awesome-nlp-polish)



### Publicly available large Polish text corpora (> 1GB)



- [OSCAR Corpus (Common Crawl extract)](https://oscar-corpus.com/)

- [CC-100 Web Crawl Data (Common Crawl extract)](http://data.statmt.org/cc-100/)

- [The Polish Parliamentary Corpus](http://clip.ipipan.waw.pl/PPC)

- [Redistributable subcorpora of the National Corpus of Polish](http://zil.ipipan.waw.pl/DistrNKJP)

- [Polish Wikipedia Dumps](https://dumps.wikimedia.org/plwiki/)

- [OPUS Parallel Corpora](https://opus.nlpl.eu/)

- [Corpus from PolEval 2018 Language Modeling Task](http://2018.poleval.pl/index.php/tasks/)

- [C4 and mC4 corpora (contains ~180GB of compressed Polish text)](https://huggingface.co/datasets/allenai/c4)

- [NLLB parallel corpus (1613 language pairs of which 43 include Polish)](https://huggingface.co/datasets/allenai/nllb)

- [CulturaX - a combination of mC4 and OSCAR corpora, cleaned and deduplicated](https://huggingface.co/datasets/uonlp/CulturaX)

- [MADLAD-400 (Multilingual Audited Dataset: Low-resource And Document-level)](https://huggingface.co/datasets/allenai/MADLAD-400)



### Models supporting Polish language



#### Sentence analysis (tokenization, lemmatization, POS tagging etc.)



- [SpaCy](https://spacy.io/models/pl) - A popular library for NLP in Python which includes Polish models for sentence analysis.

- [Stanza](https://stanfordnlp.github.io/stanza/) - A collection of neural NLP models for many languages from StndordNLP.

- [Trankit](https://github.com/nlp-uoregon/trankit) - A light-weight transformer-based python toolkit for multilingual natural language processing by the University of Oregon.

- [KRNNT](https://github.com/kwrobel-nlp/krnnt) and [KFTT](https://github.com/kwrobel-nlp/kftt) - Neural morphosyntactic taggers for Polish.

- [Morfeusz](http://morfeusz.sgjp.pl/) - A classic Polish morphosyntactic tagger.

- [Language Tool](https://github.com/languagetool-org/languagetool) - Java-based open source proofreading software for many languages with sentence analysis tools included.

- [Stempel](https://github.com/dzieciou/pystempel) - Algorythmic stemmer for Polish.

- [PoLemma](https://huggingface.co/amu-cai/polemma-large) - plT5-based lemmatizer of named entities and multi-word expressions for Polish, available in [small](https://huggingface.co/amu-cai/polemma-small), [base](https://huggingface.co/amu-cai/polemma-base) and [large](https://huggingface.co/amu-cai/polemma-large) sizes.



#### Machine translation

- [Marian-NMT](https://marian-nmt.github.io/) - An efficient C++ based implementation of neural translation models. Many pre-trained models are available, including those supporting Polish: [pl-de](https://huggingface.co/Helsinki-NLP/opus-mt-pl-de), [pl-en](https://huggingface.co/Helsinki-NLP/opus-mt-pl-en), [pl-es](https://huggingface.co/Helsinki-NLP/opus-mt-pl-es), [pl-fr](https://huggingface.co/Helsinki-NLP/opus-mt-pl-fr), [pl-sv](https://huggingface.co/Helsinki-NLP/opus-mt-pl-sv), [de-pl](https://huggingface.co/Helsinki-NLP/opus-mt-de-pl), [es-pl](https://huggingface.co/Helsinki-NLP/opus-mt-es-pl), [fr-pl](https://huggingface.co/Helsinki-NLP/opus-mt-fr-pl).

- [M2M](https://github.com/pytorch/fairseq/tree/master/examples/m2m_100) (2021) - A single massive machine translation architecture supporting direct translation for any pair from the list of 100 languages. Details in the paper [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/pdf/2010.11125.pdf).

- [mBART-50](https://huggingface.co/facebook/mbart-large-50-many-to-many-mmt) (2021) - A multilingual BART model fine-tuned for machine translation in 50 languages. Three machine translation models were published: [many-to-many](https://huggingface.co/facebook/mbart-large-50-many-to-many-mmt), [English-to-many](https://huggingface.co/facebook/mbart-large-50-one-to-many-mmt), and [many-to-English](https://huggingface.co/facebook/mbart-large-50-many-to-one-mmt). For more information see [Multilingual Translation with Extensible Multilingual Pretraining and Finetuning](https://arxiv.org/abs/2008.00401).

- [NLLB](https://github.com/facebookresearch/fairseq/tree/nllb) (2022) - NLLB (No Language Left Behind) is a project by Meta AI aiming to provide machine translation models for over 200 languages. A set of multilingual neural models ranging from 600M to 54.5B parameters is available for download. For more details see [No Language Left Behind: Scaling Human-Centered Machine Translation](https://research.facebook.com/publications/no-language-left-behind/).



#### Language models

- [Multilingual BERT](https://github.com/google-research/bert/blob/master/multilingual.md) (2018) - BERT (Bidirectional Encoder Representations from Transformers) is a model for generating contextual word representations. Multilingual cased model provided by Google supports 104 languages including Polish.

- [XLM-RoBERTa](https://github.com/pytorch/fairseq/tree/master/examples/xlmr) (2019) - Cross lingual language model trained on 2.5 terabytes of data from CommonCrawl and Wikipedia. Supports 100 languages including Polish. See [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/pdf/1911.02116.pdf) for details.

- [Slavic BERT](https://github.com/deepmipt/Slavic-BERT-NER#slavic-bert) (2019) - Multilingual BERT model supporting Bulgarian (bg), Czech (cs), Polish (pl) and Russian (ru) languages.

- [mT5](https://github.com/google-research/multilingual-t5) (2020) - Google's text-to-text transformer for 101 languages based on the T5 architecture. Details in the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934).

- [HerBERT](https://huggingface.co/allegro) (2020) - Polish BERT-based language model trained by Allegro for HuggingFace Transformers in [base](https://huggingface.co/allegro/herbert-base-cased) and [large](https://huggingface.co/allegro/herbert-large-cased) variant.

- [plT5](https://huggingface.co/allegro/plt5-large) (2021) - Polish version of the T5 model available in [small](https://huggingface.co/allegro/plt5-small), [base](https://huggingface.co/allegro/plt5-base) and [large](https://huggingface.co/allegro/plt5-large) sizes.

- [ByT5](https://huggingface.co/docs/transformers/model_doc/byt5) (2021) - A multilignual sequence to sequence model similar to T5, but using raw byte sequences as inputs instead of subword tokens. Introduced in the paper [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626).

- [XLM-RoBERTa-XL and XXL](https://github.com/pytorch/fairseq/blob/main/examples/xlmr/README.md) (2021) - Large-scale versions of XLM-RoBERTa models with 3.5 and 10.7 billion parameters respectively. For more information see [Larger-Scale Transformers for Multilingual Masked Language Modeling](https://arxiv.org/pdf/2105.00572.pdf).

- [mLUKE](https://huggingface.co/docs/transformers/model_doc/mluke) (2021) - A multilingual version of LUKE, Transformer-based language model enriched with entity metadata. The model supports 24 languages including Polish. For more information see [mLUKE: The Power of Entity Representations in Multilingual Pretrained Language Models](https://arxiv.org/pdf/2110.08151.pdf).

- [XGLM](https://huggingface.co/facebook/xglm-4.5B) (2021) - A GPT style autoregressive Transformer language model trained on a large-scale multilingual corpus. The model was published in several sizes, but only the 4.5B variant includes Polish language. For more information see [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668).

- [PapuGaPT2](https://huggingface.co/flax-community/papuGaPT2) (2021) - Polish GPT-like autoregressive models available in [base](https://huggingface.co/flax-community/papuGaPT2) and [large](https://huggingface.co/flax-community/papuGaPT2-large) sizes.

- [mGPT](https://huggingface.co/sberbank-ai/mGPT) (2022) - Another multilingual GPT style model with 1.3B parameters, covering 60 languages. The model has been trained by Sberbank AI. For more information see [mGPT: Few-Shot Learners Go Multilingual](https://arxiv.org/abs/2204.07580). 

- [Flan-T5](https://huggingface.co/docs/transformers/model_doc/flan-t5) (2022) - An improved version of T5 model, fine-tuned on a broad set of downstream tasks in multiple languages. Flan-T5 models can be used in zero-shot and few-shot scenarios, they can also be further fine-tuned for specific task. For more information see [Scaling Instruction-Finetuned Language Models](https://arxiv.org/pdf/2210.11416.pdf). 

- [XLM-V](https://huggingface.co/facebook/xlm-v-base) (2023) - A multilingual transformer-based language model utilising large vocabulary of 1 million tokens, which brings significant improvements on downstream tasks for some languages. Apart from a larger vocabulary, the model's architecture is similar to previously published XLM-R models. For more information see [XLM-V: Overcoming the Vocabulary Bottleneck in Multilingual Masked Language Models](https://arxiv.org/abs/2301.10472). 

- [umT5](https://console.cloud.google.com/storage/browser/t5-data/pretrained_models/t5x/umt5_xxl) (2023) - An improved mt5 model trained using a more uniform language distribution. For more information see [UniMax: Fairer and More Effective Language Sampling for Large-Scale Multilingual Pretraining](https://arxiv.org/pdf/2304.09151.pdf).

- [mLongT5](https://console.cloud.google.com/storage/browser/t5-data/pretrained_models/t5x/mlongt5) (2023) - A multilingual version of LongT5 which is an extension of the T5 model that handles long inputs of up to 16k tokens. Supports 101 languages including Polish. For more information see [mLongT5: A Multilingual and Efficient Text-To-Text Transformer for

Longer Sequences](https://arxiv.org/pdf/2305.11129.pdf).



#### Sentence encoders

- [Universal Sentence Encoder](https://tfhub.dev/google/universal-sentence-encoder-multilingual-large/1) (2019) - USE (Universal Sentence Encoder) generates sentence level langauge representations. Pre-trained multilingual model supports 16 langauges (Arabic, Chinese-simplified, Chinese-traditional, English, French, German, Italian, Japanese, Korean, Dutch, Polish, Portuguese, Spanish, Thai, Turkish, Russian).

- [LASER Language-Agnostic SEntence Representations](https://github.com/facebookresearch/LASER) (2019) - A multilingual sentence encoder by Facebook Research, supporting 93 languages.

- [LaBSE](https://tfhub.dev/google/LaBSE/1) (2020) - Language-agnostic BERT sentence embedding model supporting 109 languages. See [Language-agnostic BERT Sentence Embedding](https://arxiv.org/abs/2007.01852) for details.

- [Sentence Transformers](https://github.com/UKPLab/sentence-transformers) (2020) - Sentence-level models based on the transformer architecture. The library includes multilingual models supporting Polish. More information on multilingual knowledge distillation method used by the authors can be found in [Making Monolingual Sentence Embeddings Multilingual using Knowledge Distillation](https://arxiv.org/abs/2004.09813).

- [LASER2 and LASER3](https://github.com/facebookresearch/LASER/blob/main/nllb/README.md) (2022) - New versions of the LASER sentence encoder by Meta AI, developed as a part of the NLLB (No Language Left Behind) project. LASER2 supports the same set of languages as the first version of the encoder, which includes Polish. LASER3 adds support to less common languages, mostly low-resource African languages. See [Bitext Mining Using Distilled Sentence Representations for Low-Resource Languages](https://arxiv.org/pdf/2205.12654.pdf) for more details.

- [E5](https://huggingface.co/intfloat/multilingual-e5-large) (2022) - A general purpose text encoder which can be applied to a variety of tasks such as information retrieval, semantic textual similarity, text reranking, or clustering. The models were trained using a large dataset of text pairs extracted from CommonCrawl. Recently, multilingual E5 models supporting Polish have been published in [small](https://huggingface.co/intfloat/multilingual-e5-small), [base](https://huggingface.co/intfloat/multilingual-e5-base), and [large](https://huggingface.co/intfloat/multilingual-e5-large) versions. See [Text Embeddings by Weakly-Supervised Contrastive Pre-training](https://arxiv.org/abs/2212.03533) for more details.



#### Optical character recognition (OCR)

- [Easy OCR](https://github.com/JaidedAI/EasyOCR) - Optical character recognition toolkit with pre-trained models for over 40 languages, including Polish.

- [Tesseract](https://github.com/tesseract-ocr/tesseract) - Popular OCR software developed since 1980s, supporting over 100 languages. For integration with Python, wrappers such as [PyTesseract](https://github.com/madmaze/pytesseract) or [OCRMyPDF](https://github.com/ocrmypdf/OCRmyPDF) can be used. 



#### Speech processing (speech recognition, text-to-speech, voice cloning etc.)

- [Quartznet - Nvidia NeMo](https://catalog.ngc.nvidia.com/orgs/nvidia/teams/nemo/models/stt_pl_quartznet15x5) (2021) - Nvidia NeMo is a toolkit for building conversational AI models. Apart from the framework itself, Nvidia also published many models trained using their code, which includes a speech recognition model for Polish based on Quartznet architecture.

- [XLS-R](https://huggingface.co/facebook/wav2vec2-xls-r-300m) (2021) - XLS-R is a multilingual version of Wav2Vec 2.0 model by Meta AI, which is a large-scale pre-trained model for speech processing. The model is trained in a self-supervised way, so it needs to be fine-tuned for solving specific tasks such as ASR. Several fine-tuned checkpoints for Polish speech recognition exist on the HuggingFace Hub e.g. [wav2vec2-large-xlsr-53-polish](https://huggingface.co/jonatasgrosman/wav2vec2-large-xlsr-53-polish)

- [M-CTC-T](https://huggingface.co/speechbrain/m-ctc-t-large) (2021) - A speech recognition model from Meta AI, supporting 60 languages including Polish. For more information see [Pseudo-Labeling for Massively Multilingual Speech Recognition](https://arxiv.org/abs/2111.00161).

- [Whisper](https://github.com/openai/whisper/) (2022) - Whisper is a model released by OpenAI for ASR and other speech-related tasks, supporting 82 languages. The model is available in five sizes: tiny (39M params), base (74M), small (244M), medium (769M), and large (1.5B). More information can be found in the paper [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf).

- [MMS](https://github.com/facebookresearch/fairseq/blob/main/examples/mms/README.md) (2023) - Massively Multilingual Speech (MMS) is a large-scale multilingual speech foundation model published by Meta AI. Along with the pre-trained models, they also released checkpoints fine-tuned for specific tasks such as [speech recognition](https://github.com/facebookresearch/fairseq/blob/main/examples/mms/README.md#asr), [text to speech](https://github.com/facebookresearch/fairseq/blob/main/examples/mms/README.md#tts), and [language identification](https://github.com/facebookresearch/fairseq/blob/main/examples/mms/README.md#lid). For more information see [Scaling Speech Technology to 1,000+ Languages](https://research.facebook.com/publications/scaling-speech-technology-to-1000-languages/).

- [SeamlessM4T](https://github.com/facebookresearch/seamless_communication) (2023) - Multilingual and multitask model trained on text and speech data. It covers almost 100 languages including Polish and can perform automatic speech recognition (ASR) as well as multimodal translation tasks across languages: speech-to-text, text-to-speech, speech-to-speech, text-to-text. See [SeamlessM4T — Massively Multilingual & Multimodal Machine Translation](https://dl.fbaipublicfiles.com/seamless/seamless_m4t_paper.pdf) for more details.

- [SONAR](https://github.com/facebookresearch/SONAR#supported-languages-and-download-links) (2023) - Multilingual embeddings for speech and text with a set of additional models fine-tuned for specific tasks such as text translation, speech-to-text translation, or cross-lingual semantic similarity. See [SONAR: Sentence-Level Multimodal and Language-Agnostic Representations](https://ai.meta.com/research/publications/sonar-sentence-level-multimodal-and-language-agnostic-representations/) for more details.

- [XTTS](https://huggingface.co/coqui/XTTS-v1) (2023) - A text-to-speech model that allows voice cloning by using just a 3-second audio sample of the target voice. Supports 13 languages including Polish.



#### Multimodal models

- [Multilingual CLIP (SBert)](https://huggingface.co/sentence-transformers/clip-ViT-B-32-multilingual-v1) (2021) - CLIP (Contrastive Language-Image Pre-Training) is a neural network introducted by [OpenAI](https://github.com/openai/CLIP) which enables joint vector representations for images and text. It can be used for building image search engines. This is a multilingual version of CLIP trained by the authors of the [Sentence-Transformers](https://www.sbert.net/) library.

- [Multilingual CLIP (M-CLIP)](https://huggingface.co/M-CLIP/M-BERT-Base-ViT-B) (2021) - This is yet another multilingual version of CLIP supporting Polish language, trained by the Swedish Institute of Computer Science (SICS).

- [LayoutXLM](https://huggingface.co/microsoft/layoutxlm-base) (2021) - A multilingual version of [LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2) model, pre-trained on 30 million  documents in 53 languages. The model combines visual, spatial, and textual modalities to solve prediction problems on visually-rich documents, such as PDFs or DOCs. See [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) and [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) for details.