Add new pre-trained models BERTweet and PhoBERT

model_cards/vinai/bertweet-base/README.md

@@ -0,0 +1,71 @@
+# <a name="introduction"></a> BERTweet: A pre-trained language model for English Tweets 
+
+ - BERTweet is the first public large-scale language model pre-trained for English Tweets. BERTweet is trained based on the [RoBERTa](https://github.com/pytorch/fairseq/blob/master/examples/roberta/README.md)  pre-training procedure, using the same model configuration as [BERT-base](https://github.com/google-research/bert). 
+ - The corpus used to pre-train BERTweet consists of 850M English Tweets (16B word tokens ~ 80GB), containing 845M Tweets streamed from 01/2012 to 08/2019 and 5M Tweets related the **COVID-19** pandemic. 
+ - BERTweet does better than its competitors RoBERTa-base and [XLM-R-base](https://arxiv.org/abs/1911.02116) and outperforms previous state-of-the-art models on three downstream Tweet NLP tasks of Part-of-speech tagging, Named entity recognition and text classification.
+
+The general architecture and experimental results of BERTweet can be found in our EMNLP-2020 demo [paper](https://arxiv.org/abs/2005.10200):
+
+    @inproceedings{bertweet,
+    title     = {{BERTweet: A pre-trained language model for English Tweets}},
+    author    = {Dat Quoc Nguyen and Thanh Vu and Anh Tuan Nguyen},
+    booktitle = {Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations},
+    year      = {2020}
+    }
+
+**Please CITE** our paper when BERTweet is used to help produce published results or is incorporated into other software.
+
+For further information or requests, please go to [BERTweet's homepage](https://github.com/VinAIResearch/BERTweet)!
+
+## <a name="install2"></a> Installation 
+
+ - Python version >= 3.6
+ - [PyTorch](http://pytorch.org/) version >= 1.4.0
+ - `pip3 install transformers emoji`
+
+## <a name="models2"></a> Pre-trained model 
+
+Model | #params | Arch. | Pre-training data
+---|---|---|---
+`vinai/bertweet-base` | 135M | base | 845M English Tweets (80GB)
+
+
+## <a name="usage2"></a> Example usage 
+
+
+```python
+import torch
+from transformers import AutoModel, AutoTokenizer #, BertweetTokenizer
+
+bertweet = AutoModel.from_pretrained("vinai/bertweet-base")
+tokenizer = AutoTokenizer.from_pretrained("vinai/bertweet-base")
+#tokenizer = BertweetTokenizer.from_pretrained("vinai/bertweet-base")
+
+# INPUT TWEET IS ALREADY NORMALIZED!
+line = "SC has first two presumptive cases of coronavirus , DHEC confirms HTTPURL via @USER :cry:"
+
+input_ids = torch.tensor([tokenizer.encode(line)])
+
+with torch.no_grad():
+    features = bertweet(input_ids)  # Models outputs are now tuples
+```
+
+## <a name="preprocess"></a> Normalize raw input Tweets 
+
+Before applying `fastBPE` to the pre-training corpus of 850M English Tweets, we tokenized these  Tweets using `TweetTokenizer` from the NLTK toolkit and used the `emoji` package to translate emotion icons into text strings (here, each icon is referred to as a word token).   We also normalized the Tweets by converting user mentions and web/url links into special tokens `@USER` and `HTTPURL`, respectively. Thus it is recommended to also apply the same pre-processing step for BERTweet-based downstream applications w.r.t. the raw input Tweets.
+
+```python
+import torch
+from transformers import BertweetTokenizer
+
+# Load the BertweetTokenizer with a normalization mode if the input Tweet is raw
+tokenizer = BertweetTokenizer.from_pretrained("vinai/bertweet-base", normalization=True)
+
+# BERTweet's tokenizer can be also loaded in the "Auto" mode
+# from transformers import AutoTokenizer
+# tokenizer = AutoTokenizer.from_pretrained("vinai/bertweet-base", normalization=True)
+
+line = "SC has first two presumptive cases of coronavirus, DHEC confirms https://postandcourier.com/health/covid19/sc-has-first-two-presumptive-cases-of-coronavirus-dhec-confirms/article_bddfe4ae-5fd3-11ea-9ce4-5f495366cee6.html?utm_medium=social&utm_source=twitter&utm_campaign=user-share… via @postandcourier"
+
+input_ids = torch.tensor([tokenizer.encode(line)])
+```

model_cards/vinai/phobert-base/README.md

@@ -0,0 +1,51 @@
+# <a name="introduction"></a> PhoBERT: Pre-trained language models for Vietnamese 
+
+Pre-trained PhoBERT models are the state-of-the-art language models for Vietnamese ([Pho](https://en.wikipedia.org/wiki/Pho), i.e. "Phở", is a popular food in Vietnam): 
+
+ - Two PhoBERT versions of "base" and "large" are the first public large-scale monolingual language models pre-trained for Vietnamese. PhoBERT pre-training approach is based on [RoBERTa](https://github.com/pytorch/fairseq/blob/master/examples/roberta/README.md)  which optimizes the [BERT](https://github.com/google-research/bert) pre-training procedure for more robust performance.
+ - PhoBERT outperforms previous monolingual and multilingual approaches, obtaining new state-of-the-art performances on four downstream Vietnamese NLP tasks of Part-of-speech tagging, Dependency parsing, Named-entity recognition and Natural language inference.
+
+The general architecture and experimental results of PhoBERT can be found in our EMNLP-2020 Findings [paper](https://arxiv.org/abs/2003.00744):
+
+    @article{phobert,
+    title     = {{PhoBERT: Pre-trained language models for Vietnamese}},
+    author    = {Dat Quoc Nguyen and Anh Tuan Nguyen},
+    journal   = {Findings of EMNLP},
+    year      = {2020}
+    }
+
+**Please CITE** our paper when PhoBERT is used to help produce published results or is incorporated into other software.
+
+For further information or requests, please go to [PhoBERT's homepage](https://github.com/VinAIResearch/PhoBERT)!
+
+## Installation <a name="install2"></a>
+ - Python version >= 3.6
+ - [PyTorch](http://pytorch.org/) version >= 1.4.0
+ - `pip3 install transformers`
+
+## Pre-trained models <a name="models2"></a>
+
+
+Model | #params | Arch.	 | Pre-training data
+---|---|---|---
+`vinai/phobert-base` | 135M | base | 20GB  of texts
+`vinai/phobert-large` | 370M | large | 20GB  of texts
+
+## Example usage <a name="usage2"></a>
+
+```python
+import torch
+from transformers import AutoModel, AutoTokenizer #, PhobertTokenizer
+
+phobert = AutoModel.from_pretrained("vinai/phobert-base")
+tokenizer = AutoTokenizer.from_pretrained("vinai/phobert-base")
+#tokenizer = PhobertTokenizer.from_pretrained("vinai/phobert-base")
+
+# INPUT TEXT MUST BE ALREADY WORD-SEGMENTED!
+line = "Tôi là sinh_viên trường đại_học Công_nghệ ."
+
+input_ids = torch.tensor([tokenizer.encode(line)])
+
+with torch.no_grad():
+    features = phobert(input_ids)  # Models outputs are now tuples
+```

model_cards/vinai/phobert-large/README.md

@@ -0,0 +1,51 @@
+# <a name="introduction"></a> PhoBERT: Pre-trained language models for Vietnamese 
+
+Pre-trained PhoBERT models are the state-of-the-art language models for Vietnamese ([Pho](https://en.wikipedia.org/wiki/Pho), i.e. "Phở", is a popular food in Vietnam): 
+
+ - Two PhoBERT versions of "base" and "large" are the first public large-scale monolingual language models pre-trained for Vietnamese. PhoBERT pre-training approach is based on [RoBERTa](https://github.com/pytorch/fairseq/blob/master/examples/roberta/README.md)  which optimizes the [BERT](https://github.com/google-research/bert) pre-training procedure for more robust performance.
+ - PhoBERT outperforms previous monolingual and multilingual approaches, obtaining new state-of-the-art performances on four downstream Vietnamese NLP tasks of Part-of-speech tagging, Dependency parsing, Named-entity recognition and Natural language inference.
+
+The general architecture and experimental results of PhoBERT can be found in our EMNLP-2020 Findings [paper](https://arxiv.org/abs/2003.00744):
+
+    @article{phobert,
+    title     = {{PhoBERT: Pre-trained language models for Vietnamese}},
+    author    = {Dat Quoc Nguyen and Anh Tuan Nguyen},
+    journal   = {Findings of EMNLP},
+    year      = {2020}
+    }
+
+**Please CITE** our paper when PhoBERT is used to help produce published results or is incorporated into other software.
+
+For further information or requests, please go to [PhoBERT's homepage](https://github.com/VinAIResearch/PhoBERT)!
+
+## Installation <a name="install2"></a>
+ - Python version >= 3.6
+ - [PyTorch](http://pytorch.org/) version >= 1.4.0
+ - `pip3 install transformers`
+
+## Pre-trained models <a name="models2"></a>
+
+
+Model | #params | Arch.	 | Pre-training data
+---|---|---|---
+`vinai/phobert-base` | 135M | base | 20GB  of texts
+`vinai/phobert-large` | 370M | large | 20GB  of texts
+
+## Example usage <a name="usage2"></a>
+
+```python
+import torch
+from transformers import AutoModel, AutoTokenizer #, PhobertTokenizer
+
+phobert = AutoModel.from_pretrained("vinai/phobert-large")
+tokenizer = AutoTokenizer.from_pretrained("vinai/phobert-large")
+#tokenizer = PhobertTokenizer.from_pretrained("vinai/phobert-base")
+
+# INPUT TEXT MUST BE ALREADY WORD-SEGMENTED!
+line = "Tôi là sinh_viên trường đại_học Công_nghệ ."
+
+input_ids = torch.tensor([tokenizer.encode(line)])
+
+with torch.no_grad():
+    features = phobert(input_ids)  # Models outputs are now tuples
+```

src/transformers/__init__.py

@@ -145,6 +145,7 @@
 from .tokenization_bert import BasicTokenizer, BertTokenizer, BertTokenizerFast, WordpieceTokenizer
 from .tokenization_bert_generation import BertGenerationTokenizer
 from .tokenization_bert_japanese import BertJapaneseTokenizer, CharacterTokenizer, MecabTokenizer
+from .tokenization_bertweet import BertweetTokenizer
 from .tokenization_camembert import CamembertTokenizer
 from .tokenization_ctrl import CTRLTokenizer
 from .tokenization_distilbert import DistilBertTokenizer, DistilBertTokenizerFast
@@ -166,6 +167,7 @@
 from .tokenization_mobilebert import MobileBertTokenizer, MobileBertTokenizerFast
 from .tokenization_openai import OpenAIGPTTokenizer, OpenAIGPTTokenizerFast
 from .tokenization_pegasus import PegasusTokenizer
+from .tokenization_phobert import PhobertTokenizer
 from .tokenization_reformer import ReformerTokenizer
 from .tokenization_retribert import RetriBertTokenizer, RetriBertTokenizerFast
 from .tokenization_roberta import RobertaTokenizer, RobertaTokenizerFast

src/transformers/tokenization_auto.py

@@ -54,6 +54,7 @@
 from .tokenization_bert import BertTokenizer, BertTokenizerFast
 from .tokenization_bert_generation import BertGenerationTokenizer
 from .tokenization_bert_japanese import BertJapaneseTokenizer
+from .tokenization_bertweet import BertweetTokenizer
 from .tokenization_camembert import CamembertTokenizer
 from .tokenization_ctrl import CTRLTokenizer
 from .tokenization_distilbert import DistilBertTokenizer, DistilBertTokenizerFast
@@ -68,6 +69,7 @@
 from .tokenization_mobilebert import MobileBertTokenizer, MobileBertTokenizerFast
 from .tokenization_openai import OpenAIGPTTokenizer, OpenAIGPTTokenizerFast
 from .tokenization_pegasus import PegasusTokenizer
+from .tokenization_phobert import PhobertTokenizer
 from .tokenization_reformer import ReformerTokenizer
 from .tokenization_retribert import RetriBertTokenizer, RetriBertTokenizerFast
 from .tokenization_roberta import RobertaTokenizer, RobertaTokenizerFast
@@ -96,6 +98,8 @@
         (MarianConfig, (MarianTokenizer, None)),
         (BartConfig, (BartTokenizer, BartTokenizerFast)),
         (LongformerConfig, (LongformerTokenizer, LongformerTokenizerFast)),
+        (RobertaConfig, (BertweetTokenizer, None)),
+        (RobertaConfig, (PhobertTokenizer, None)),
         (RobertaConfig, (RobertaTokenizer, RobertaTokenizerFast)),
         (ReformerConfig, (ReformerTokenizer, None)),
         (ElectraConfig, (ElectraTokenizer, ElectraTokenizerFast)),