FacebookAI/roberta-base

1---2language: en3tags:4- exbert5license: mit6datasets:7- bookcorpus8- wikipedia9---10 11# RoBERTa base model12 13Pretrained model on English language using a masked language modeling (MLM) objective. It was introduced in14[this paper](https://arxiv.org/abs/1907.11692) and first released in15[this repository](https://github.com/pytorch/fairseq/tree/master/examples/roberta). This model is case-sensitive: it16makes a difference between english and English.17 18Disclaimer: The team releasing RoBERTa did not write a model card for this model so this model card has been written by19the Hugging Face team.20 21## Model description22 23RoBERTa is a transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means24it was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of25publicly available data) with an automatic process to generate inputs and labels from those texts. 26 27More precisely, it was pretrained with the Masked language modeling (MLM) objective. Taking a sentence, the model28randomly masks 15% of the words in the input then run the entire masked sentence through the model and has to predict29the masked words. This is different from traditional recurrent neural networks (RNNs) that usually see the words one30after the other, or from autoregressive models like GPT which internally mask the future tokens. It allows the model to31learn a bidirectional representation of the sentence.32 33This way, the model learns an inner representation of the English language that can then be used to extract features34useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard35classifier using the features produced by the BERT model as inputs.36 37## Intended uses & limitations38 39You can use the raw model for masked language modeling, but it's mostly intended to be fine-tuned on a downstream task.40See the [model hub](https://huggingface.co/models?filter=roberta) to look for fine-tuned versions on a task that41interests you.42 43Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)44to make decisions, such as sequence classification, token classification or question answering. For tasks such as text45generation you should look at a model like GPT2.46 47### How to use48 49You can use this model directly with a pipeline for masked language modeling:50 51```python52>>> from transformers import pipeline53>>> unmasker = pipeline('fill-mask', model='roberta-base')54>>> unmasker("Hello I'm a <mask> model.")55 56[{'sequence': "<s>Hello I'm a male model.</s>",57  'score': 0.3306540250778198,58  'token': 2943,59  'token_str': 'Ġmale'},60 {'sequence': "<s>Hello I'm a female model.</s>",61  'score': 0.04655390977859497,62  'token': 2182,63  'token_str': 'Ġfemale'},64 {'sequence': "<s>Hello I'm a professional model.</s>",65  'score': 0.04232972860336304,66  'token': 2038,67  'token_str': 'Ġprofessional'},68 {'sequence': "<s>Hello I'm a fashion model.</s>",69  'score': 0.037216778844594955,70  'token': 2734,71  'token_str': 'Ġfashion'},72 {'sequence': "<s>Hello I'm a Russian model.</s>",73  'score': 0.03253649175167084,74  'token': 1083,75  'token_str': 'ĠRussian'}]76```77 78Here is how to use this model to get the features of a given text in PyTorch:79 80```python81from transformers import RobertaTokenizer, RobertaModel82tokenizer = RobertaTokenizer.from_pretrained('roberta-base')83model = RobertaModel.from_pretrained('roberta-base')84text = "Replace me by any text you'd like."85encoded_input = tokenizer(text, return_tensors='pt')86output = model(**encoded_input)87```88 89and in TensorFlow:90 91```python92from transformers import RobertaTokenizer, TFRobertaModel93tokenizer = RobertaTokenizer.from_pretrained('roberta-base')94model = TFRobertaModel.from_pretrained('roberta-base')95text = "Replace me by any text you'd like."96encoded_input = tokenizer(text, return_tensors='tf')97output = model(encoded_input)98```99 100### Limitations and bias101 102The training data used for this model contains a lot of unfiltered content from the internet, which is far from103neutral. Therefore, the model can have biased predictions:104 105```python106>>> from transformers import pipeline107>>> unmasker = pipeline('fill-mask', model='roberta-base')108>>> unmasker("The man worked as a <mask>.")109 110[{'sequence': '<s>The man worked as a mechanic.</s>',111  'score': 0.08702439814805984,112  'token': 25682,113  'token_str': 'Ġmechanic'},114 {'sequence': '<s>The man worked as a waiter.</s>',115  'score': 0.0819653645157814,116  'token': 38233,117  'token_str': 'Ġwaiter'},118 {'sequence': '<s>The man worked as a butcher.</s>',119  'score': 0.073323555290699,120  'token': 32364,121  'token_str': 'Ġbutcher'},122 {'sequence': '<s>The man worked as a miner.</s>',123  'score': 0.046322137117385864,124  'token': 18678,125  'token_str': 'Ġminer'},126 {'sequence': '<s>The man worked as a guard.</s>',127  'score': 0.040150221437215805,128  'token': 2510,129  'token_str': 'Ġguard'}]130 131>>> unmasker("The Black woman worked as a <mask>.")132 133[{'sequence': '<s>The Black woman worked as a waitress.</s>',134  'score': 0.22177888453006744,135  'token': 35698,136  'token_str': 'Ġwaitress'},137 {'sequence': '<s>The Black woman worked as a prostitute.</s>',138  'score': 0.19288744032382965,139  'token': 36289,140  'token_str': 'Ġprostitute'},141 {'sequence': '<s>The Black woman worked as a maid.</s>',142  'score': 0.06498628109693527,143  'token': 29754,144  'token_str': 'Ġmaid'},145 {'sequence': '<s>The Black woman worked as a secretary.</s>',146  'score': 0.05375480651855469,147  'token': 2971,148  'token_str': 'Ġsecretary'},149 {'sequence': '<s>The Black woman worked as a nurse.</s>',150  'score': 0.05245552211999893,151  'token': 9008,152  'token_str': 'Ġnurse'}]153```154 155This bias will also affect all fine-tuned versions of this model.156 157## Training data158 159The RoBERTa model was pretrained on the reunion of five datasets:160- [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038 unpublished books;161- [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and headers) ;162- [CC-News](https://commoncrawl.org/2016/10/news-dataset-available/), a dataset containing 63 millions English news163  articles crawled between September 2016 and February 2019.164- [OpenWebText](https://github.com/jcpeterson/openwebtext), an opensource recreation of the WebText dataset used to165  train GPT-2,166- [Stories](https://arxiv.org/abs/1806.02847) a dataset containing a subset of CommonCrawl data filtered to match the167  story-like style of Winograd schemas.168 169Together these datasets weigh 160GB of text.170 171## Training procedure172 173### Preprocessing174 175The texts are tokenized using a byte version of Byte-Pair Encoding (BPE) and a vocabulary size of 50,000. The inputs of176the model take pieces of 512 contiguous tokens that may span over documents. The beginning of a new document is marked177with `<s>` and the end of one by `</s>`178 179The details of the masking procedure for each sentence are the following:180- 15% of the tokens are masked.181- In 80% of the cases, the masked tokens are replaced by `<mask>`.182- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.183- In the 10% remaining cases, the masked tokens are left as is.184 185Contrary to BERT, the masking is done dynamically during pretraining (e.g., it changes at each epoch and is not fixed).186 187### Pretraining188 189The model was trained on 1024 V100 GPUs for 500K steps with a batch size of 8K and a sequence length of 512. The190optimizer used is Adam with a learning rate of 6e-4, \\(\beta_{1} = 0.9\\), \\(\beta_{2} = 0.98\\) and191\\(\epsilon = 1e-6\\), a weight decay of 0.01, learning rate warmup for 24,000 steps and linear decay of the learning192rate after.193 194## Evaluation results195 196When fine-tuned on downstream tasks, this model achieves the following results:197 198Glue test results:199 200| Task | MNLI | QQP  | QNLI | SST-2 | CoLA | STS-B | MRPC | RTE  |201|:----:|:----:|:----:|:----:|:-----:|:----:|:-----:|:----:|:----:|202|      | 87.6 | 91.9 | 92.8 | 94.8  | 63.6 | 91.2  | 90.2 | 78.7 |203 204 205### BibTeX entry and citation info206 207```bibtex208@article{DBLP:journals/corr/abs-1907-11692,209  author    = {Yinhan Liu and210               Myle Ott and211               Naman Goyal and212               Jingfei Du and213               Mandar Joshi and214               Danqi Chen and215               Omer Levy and216               Mike Lewis and217               Luke Zettlemoyer and218               Veselin Stoyanov},219  title     = {RoBERTa: {A} Robustly Optimized {BERT} Pretraining Approach},220  journal   = {CoRR},221  volume    = {abs/1907.11692},222  year      = {2019},223  url       = {http://arxiv.org/abs/1907.11692},224  archivePrefix = {arXiv},225  eprint    = {1907.11692},226  timestamp = {Thu, 01 Aug 2019 08:59:33 +0200},227  biburl    = {https://dblp.org/rec/journals/corr/abs-1907-11692.bib},228  bibsource = {dblp computer science bibliography, https://dblp.org}229}230```231 232<a href="https://huggingface.co/exbert/?model=roberta-base">233	<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">234</a>235