save intermediate

prophetnet modified

modify codes as suggested v1

add prophetnet test files

README.md

@@ -161,34 +161,36 @@ If you'd like to play with the examples, you must [install the library from sour
 1. **[ALBERT](https://huggingface.co/transformers/model_doc/albert.html)** (from Google Research and the Toyota Technological Institute at Chicago) released with the paper [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://arxiv.org/abs/1909.11942), by Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut.
 1. **[BART](https://huggingface.co/transformers/model_doc/bart.html)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/pdf/1910.13461.pdf) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
 1. **[BERT](https://huggingface.co/transformers/model_doc/bert.html)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
-1. **[BERT For Sequence Generation](https://huggingface.co/transformers/model_doc/bertgeneration.html)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
-1. **[Blenderbot](https://huggingface.co/transformers/master/model_doc/blenderbot.html)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
+1. **[BERT For Sequence Generation](https://tfhub.dev/s?module-type=text-generation&subtype=module,placeholder)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
 1. **[CamemBERT](https://huggingface.co/transformers/model_doc/camembert.html)** (from Inria/Facebook/Sorbonne) released with the paper [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
 1. **[CTRL](https://huggingface.co/transformers/model_doc/ctrl.html)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
-1. **[DeBERTa](https://huggingface.co/transformers/master/model_doc/deberta.html)** (from Microsoft Research) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
+1. **[DeBERTa](https://huggingface.co/transformers/model_doc/deberta.html)** (from Microsoft Research) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
 1. **[DialoGPT](https://huggingface.co/transformers/model_doc/dialogpt.html)** (from Microsoft Research) released with the paper [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.
 1. **[DistilBERT](https://huggingface.co/transformers/model_doc/distilbert.html)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/master/examples/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/master/examples/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/master/examples/distillation) and a German version of DistilBERT.
-1. **[DPR](https://huggingface.co/transformers/model_doc/dpr.html)** (from Facebook) released with the paper [Dense Passage Retrieval
+1. **[DPR](https://github.com/facebookresearch/DPR)** (from Facebook) released with the paper [Dense Passage Retrieval
 for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) by Vladimir Karpukhin, Barlas Oğuz, Sewon
 Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
 1. **[ELECTRA](https://huggingface.co/transformers/model_doc/electra.html)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
 1. **[FlauBERT](https://huggingface.co/transformers/model_doc/flaubert.html)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
-1. **[Funnel Transformer](https://huggingface.co/transformers/model_doc/funnel.html)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
+1. **[Funnel Transformer](https://github.com/laiguokun/Funnel-Transformer)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
 1. **[GPT](https://huggingface.co/transformers/model_doc/gpt.html)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
 1. **[GPT-2](https://huggingface.co/transformers/model_doc/gpt2.html)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
-1. **[LayoutLM](https://huggingface.co/transformers/model_doc/layoutlm.html)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
+1. **[LayoutLM](https://github.com/microsoft/unilm/tree/master/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
 1. **[Longformer](https://huggingface.co/transformers/model_doc/longformer.html)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-1. **[LXMERT](https://huggingface.co/transformers/model_doc/lxmert.html)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
+1. **[LXMERT](https://github.com/airsplay/lxmert)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
 1. **[MarianMT](https://huggingface.co/transformers/model_doc/marian.html)** Machine translation models trained using [OPUS](http://opus.nlpl.eu/) data by Jörg Tiedemann. The [Marian Framework](https://marian-nmt.github.io/) is being developed by the Microsoft Translator Team.
-1. **[MBart](https://huggingface.co/transformers/model_doc/mbart.html)** (from Facebook) released with the paper  [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
-1. **[Pegasus](https://huggingface.co/transformers/model_doc/pegasus.html)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777)> by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
+1. **[MBart](https://github.com/pytorch/fairseq/tree/master/examples/mbart)** (from Facebook) released with the paper  [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
+1. **[MMBT](https://github.com/facebookresearch/mmbt/)** (from Facebook), released together with the paper a [Supervised Multimodal Bitransformers for Classifying Images and Text](https://arxiv.org/pdf/1909.02950.pdf) by Douwe Kiela, Suvrat Bhooshan, Hamed Firooz, Davide Testuggine.
+1. **[Pegasus](https://github.com/google-research/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777)> by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
+1. **[ProphetNet](https://github.com/microsoft/ProphetNet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
 1. **[Reformer](https://huggingface.co/transformers/model_doc/reformer.html)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
 1. **[RoBERTa](https://huggingface.co/transformers/model_doc/roberta.html)** (from Facebook), released together with the paper a [Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
 ultilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/master/examples/distillation) and a German version of DistilBERT.
-1. **[SqueezeBert](https://huggingface.co/transformers/master/model_doc/squeezebert.html)** released with the paper [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) by Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer.
+1. **[SqueezeBert](https://huggingface.co/transformers/model_doc/squeezebert.html)** released with the paper [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) by Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer.
 1. **[T5](https://huggingface.co/transformers/model_doc/t5.html)** (from Google AI) released with the paper [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://arxiv.org/abs/1910.10683) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
 1. **[Transformer-XL](https://huggingface.co/transformers/model_doc/transformerxl.html)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
 1. **[XLM](https://huggingface.co/transformers/model_doc/xlm.html)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
+1. **[XLMProphetNet](https://github.com/microsoft/ProphetNet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
 1. **[XLM-RoBERTa](https://huggingface.co/transformers/model_doc/xlmroberta.html)** (from Facebook AI), released together with the paper [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.
 1. **[XLNet](https://huggingface.co/transformers/model_doc/xlnet.html)** (from Google/CMU) released with the paper [​XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) by Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
 1. **[Other community models](https://huggingface.co/models)**, contributed by the [community](https://huggingface.co/users).

docs/source/index.rst

@@ -57,104 +57,119 @@ conversion utilities for the following models:
 ..
     This list is updated automatically from the README with `make fix-copies`. Do not update manually!
 
-1. :doc:`ALBERT <model_doc/albert>` (from Google Research and the Toyota Technological Institute at Chicago) released
-   with the paper `ALBERT: A Lite BERT for Self-supervised Learning of Language Representations
-   <https://arxiv.org/abs/1909.11942>`__, by Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush
-   Sharma, Radu Soricut.
-2. :doc:`BART <model_doc/bart>` (from Facebook) released with the paper `BART: Denoising Sequence-to-Sequence
-   Pre-training for Natural Language Generation, Translation, and Comprehension
+1. `ALBERT <https://huggingface.co/transformers/model_doc/albert.html>`__ (from Google Research and the Toyota
+   Technological Institute at Chicago) released with the paper `ALBERT: A Lite BERT for Self-supervised Learning of
+   Language Representations <https://arxiv.org/abs/1909.11942>`__, by Zhenzhong Lan, Mingda Chen, Sebastian Goodman,
+   Kevin Gimpel, Piyush Sharma, Radu Soricut.
+2. `BART <https://huggingface.co/transformers/model_doc/bart.html>`__ (from Facebook) released with the paper `BART:
+   Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension
    <https://arxiv.org/pdf/1910.13461.pdf>`__ by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman
    Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
-3. :doc:`BERT <model_doc/bert>` (from Google) released with the paper `BERT: Pre-training of Deep Bidirectional
-   Transformers for Language Understanding <https://arxiv.org/abs/1810.04805>`__ by Jacob Devlin, Ming-Wei Chang,
-   Kenton Lee and Kristina Toutanova.
-4. :doc:`BERT For Sequence Generation <model_doc/bertgeneration>` (from Google) released with the paper `Leveraging
-   Pre-trained Checkpoints for Sequence Generation Tasks <https://arxiv.org/abs/1907.12461>`__ by Sascha Rothe, Shashi
-   Narayan, Aliaksei Severyn.
-5. `Blenderbot <https://huggingface.co/transformers/master/model_doc/blenderbot.html>`__ (from Facebook) released with
-   the paper `Recipes for building an open-domain chatbot <https://arxiv.org/abs/2004.13637>`__ by Stephen Roller,
-   Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan
-   Boureau, Jason Weston.
-6. :doc:`CamemBERT <model_doc/camembert>` (from Inria/Facebook/Sorbonne) released with the paper `CamemBERT: a Tasty
-   French Language Model <https://arxiv.org/abs/1911.03894>`__ by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz
-   Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
-7. :doc:`CTRL <model_doc/ctrl>` (from Salesforce) released with the paper `CTRL: A Conditional Transformer Language
-   Model for Controllable Generation <https://arxiv.org/abs/1909.05858>`__ by Nitish Shirish Keskar*, Bryan McCann*,
-   Lav R. Varshney, Caiming Xiong and Richard Socher.
-8. `DeBERTa <https://huggingface.co/transformers/master/model_doc/deberta.html>`__ (from Microsoft Research) released
-   with the paper `DeBERTa: Decoding-enhanced BERT with Disentangled Attention <https://arxiv.org/abs/2006.03654>`__ by
+3. `BERT <https://huggingface.co/transformers/model_doc/bert.html>`__ (from Google) released with the paper `BERT:
+   Pre-training of Deep Bidirectional Transformers for Language Understanding <https://arxiv.org/abs/1810.04805>`__ by
+   Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
+4. `BERT For Sequence Generation <https://tfhub.dev/s?module-type=text-generation&subtype=module,placeholder>`__ (from
+   Google) released with the paper `Leveraging Pre-trained Checkpoints for Sequence Generation Tasks
+   <https://arxiv.org/abs/1907.12461>`__ by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+5. `CamemBERT <https://huggingface.co/transformers/model_doc/camembert.html>`__ (from Inria/Facebook/Sorbonne) released
+   with the paper `CamemBERT: a Tasty French Language Model <https://arxiv.org/abs/1911.03894>`__ by Louis Martin*,
+   Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé
+   Seddah and Benoît Sagot.
+6. `CTRL <https://huggingface.co/transformers/model_doc/ctrl.html>`__ (from Salesforce) released with the paper `CTRL:
+   A Conditional Transformer Language Model for Controllable Generation <https://arxiv.org/abs/1909.05858>`__ by Nitish
+   Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
+7. `DeBERTa <https://huggingface.co/transformers/model_doc/deberta.html>`__ (from Microsoft Research) released with the
+   paper `DeBERTa: Decoding-enhanced BERT with Disentangled Attention <https://arxiv.org/abs/2006.03654>`__ by
    Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
-9. :doc:`DialoGPT <model_doc/dialogpt>` (from Microsoft Research) released with the paper `DialoGPT: Large-Scale
-   Generative Pre-training for Conversational Response Generation <https://arxiv.org/abs/1911.00536>`__ by Yizhe Zhang,
-   Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.
-10. :doc:`DistilBERT <model_doc/distilbert>` (from HuggingFace), released together with the paper `DistilBERT, a
-    distilled version of BERT: smaller, faster, cheaper and lighter <https://arxiv.org/abs/1910.01108>`__ by Victor
-    Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into `DistilGPT2
-    <https://github.com/huggingface/transformers/tree/master/examples/distillation>`__, RoBERTa into `DistilRoBERTa
-    <https://github.com/huggingface/transformers/tree/master/examples/distillation>`__, Multilingual BERT into
-    `DistilmBERT <https://github.com/huggingface/transformers/tree/master/examples/distillation>`__ and a German
-    version of DistilBERT.
-11. :doc:`DPR <model_doc/dpr>` (from Facebook) released with the paper `Dense Passage Retrieval for Open-Domain
-    Question Answering <https://arxiv.org/abs/2004.04906>`__ by Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick
-    Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
-12. :doc:`ELECTRA <model_doc/electra>` (from Google Research/Stanford University) released with the paper `ELECTRA:
-    Pre-training text encoders as discriminators rather than generators <https://arxiv.org/abs/2003.10555>`__ by Kevin
-    Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
-13. :doc:`FlauBERT <model_doc/flaubert>` (from CNRS) released with the paper `FlauBERT: Unsupervised Language Model
-    Pre-training for French <https://arxiv.org/abs/1912.05372>`__ by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne,
-    Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
-14. :doc:`Funnel Transformer <model_doc/funnel>` (from CMU/Google Brain) released with the paper `Funnel-Transformer:
-    Filtering out Sequential Redundancy for Efficient Language Processing <https://arxiv.org/abs/2006.03236>`__ by
-    Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
-15. :doc:`GPT <model_doc/gpt>` (from OpenAI) released with the paper `Improving Language Understanding by Generative
-    Pre-Training <https://blog.openai.com/language-unsupervised/>`__ by Alec Radford, Karthik Narasimhan, Tim Salimans
-    and Ilya Sutskever.
-16. :doc:`GPT-2 <model_doc/gpt2>` (from OpenAI) released with the paper `Language Models are Unsupervised Multitask
-    Learners <https://blog.openai.com/better-language-models/>`__ by Alec Radford*, Jeffrey Wu*, Rewon Child, David
-    Luan, Dario Amodei** and Ilya Sutskever**.
-17. :doc:`LayoutLM <model_doc/layoutlm>` (from Microsoft Research Asia) released with the paper `LayoutLM: Pre-training
-    of Text and Layout for Document Image Understanding <https://arxiv.org/abs/1912.13318>`__ by Yiheng Xu, Minghao Li,
-    Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
-18. :doc:`Longformer <model_doc/longformer>` (from AllenAI) released with the paper `Longformer: The Long-Document
-    Transformer <https://arxiv.org/abs/2004.05150>`__ by Iz Beltagy, Matthew E. Peters, Arman Cohan.
-19. :doc:`LXMERT <model_doc/lxmert>` (from UNC Chapel Hill) released with the paper `LXMERT: Learning Cross-Modality
-    Encoder Representations from Transformers for Open-Domain Question Answering <https://arxiv.org/abs/1908.07490>`__
-    by Hao Tan and Mohit Bansal.
-20. :doc:`MarianMT <model_doc/marian>` Machine translation models trained using `OPUS <http://opus.nlpl.eu/>`__ data by
-    Jörg Tiedemann. The `Marian Framework <https://marian-nmt.github.io/>`__ is being developed by the Microsoft
-    Translator Team.
-21. :doc:`MBart <model_doc/mbart>` (from Facebook) released with the paper  `Multilingual Denoising Pre-training for
-    Neural Machine Translation <https://arxiv.org/abs/2001.08210>`__ by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li,
-    Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
-22. :doc:`Pegasus <model_doc/pegasus>` (from Google) released with the paper `PEGASUS: Pre-training with Extracted
-    Gap-sentences for Abstractive Summarization <https://arxiv.org/abs/1912.08777>`__> by Jingqing Zhang, Yao Zhao,
-    Mohammad Saleh and Peter J. Liu.
-23. :doc:`Reformer <model_doc/reformer>` (from Google Research) released with the paper `Reformer: The Efficient
-    Transformer <https://arxiv.org/abs/2001.04451>`__ by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
-24. :doc:`RoBERTa <model_doc/roberta>` (from Facebook), released together with the paper a `Robustly Optimized BERT
-    Pretraining Approach <https://arxiv.org/abs/1907.11692>`__ by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar
-    Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov. ultilingual BERT into `DistilmBERT
+8. `DialoGPT <https://huggingface.co/transformers/model_doc/dialogpt.html>`__ (from Microsoft Research) released with
+   the paper `DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation
+   <https://arxiv.org/abs/1911.00536>`__ by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang
+   Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.
+9. `DistilBERT <https://huggingface.co/transformers/model_doc/distilbert.html>`__ (from HuggingFace), released together
+   with the paper `DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter
+   <https://arxiv.org/abs/1910.01108>`__ by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been
+   applied to compress GPT2 into `DistilGPT2
+   <https://github.com/huggingface/transformers/tree/master/examples/distillation>`__, RoBERTa into `DistilRoBERTa
+   <https://github.com/huggingface/transformers/tree/master/examples/distillation>`__, Multilingual BERT into
+   `DistilmBERT <https://github.com/huggingface/transformers/tree/master/examples/distillation>`__ and a German version
+   of DistilBERT.
+10. `DPR <https://github.com/facebookresearch/DPR>`__ (from Facebook) released with the paper `Dense Passage Retrieval
+    for Open-Domain Question Answering <https://arxiv.org/abs/2004.04906>`__ by Vladimir Karpukhin, Barlas Oğuz, Sewon
+    Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
+11. `ELECTRA <https://huggingface.co/transformers/model_doc/electra.html>`__ (from Google Research/Stanford University)
+    released with the paper `ELECTRA: Pre-training text encoders as discriminators rather than generators
+    <https://arxiv.org/abs/2003.10555>`__ by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
+12. `FlauBERT <https://huggingface.co/transformers/model_doc/flaubert.html>`__ (from CNRS) released with the paper
+    `FlauBERT: Unsupervised Language Model Pre-training for French <https://arxiv.org/abs/1912.05372>`__ by Hang Le,
+    Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé,
+    Laurent Besacier, Didier Schwab.
+13. `Funnel Transformer <https://github.com/laiguokun/Funnel-Transformer>`__ (from CMU/Google Brain) released with the
+    paper `Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing
+    <https://arxiv.org/abs/2006.03236>`__ by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
+14. `GPT <https://huggingface.co/transformers/model_doc/gpt.html>`__ (from OpenAI) released with the paper `Improving
+    Language Understanding by Generative Pre-Training <https://blog.openai.com/language-unsupervised/>`__ by Alec
+    Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
+15. `GPT-2 <https://huggingface.co/transformers/model_doc/gpt2.html>`__ (from OpenAI) released with the paper `Language
+    Models are Unsupervised Multitask Learners <https://blog.openai.com/better-language-models/>`__ by Alec Radford*,
+    Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
+16. `LayoutLM <https://github.com/microsoft/unilm/tree/master/layoutlm>`__ (from Microsoft Research Asia) released with
+    the paper `LayoutLM: Pre-training of Text and Layout for Document Image Understanding
+    <https://arxiv.org/abs/1912.13318>`__ by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
+17. `Longformer <https://huggingface.co/transformers/model_doc/longformer.html>`__ (from AllenAI) released with the
+    paper `Longformer: The Long-Document Transformer <https://arxiv.org/abs/2004.05150>`__ by Iz Beltagy, Matthew E.
+    Peters, Arman Cohan.
+18. `LXMERT <https://github.com/airsplay/lxmert>`__ (from UNC Chapel Hill) released with the paper `LXMERT: Learning
+    Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering
+    <https://arxiv.org/abs/1908.07490>`__ by Hao Tan and Mohit Bansal.
+19. `MarianMT <https://huggingface.co/transformers/model_doc/marian.html>`__ Machine translation models trained using
+    `OPUS <http://opus.nlpl.eu/>`__ data by Jörg Tiedemann. The `Marian Framework <https://marian-nmt.github.io/>`__ is
+    being developed by the Microsoft Translator Team.
+20. `MBart <https://github.com/pytorch/fairseq/tree/master/examples/mbart>`__ (from Facebook) released with the paper 
+    `Multilingual Denoising Pre-training for Neural Machine Translation <https://arxiv.org/abs/2001.08210>`__ by Yinhan
+    Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
+21. `MMBT <https://github.com/facebookresearch/mmbt/>`__ (from Facebook), released together with the paper a
+    `Supervised Multimodal Bitransformers for Classifying Images and Text <https://arxiv.org/pdf/1909.02950.pdf>`__ by
+    Douwe Kiela, Suvrat Bhooshan, Hamed Firooz, Davide Testuggine.
+22. `Pegasus <https://github.com/google-research/pegasus>`__ (from Google) released with the paper `PEGASUS:
+    Pre-training with Extracted Gap-sentences for Abstractive Summarization <https://arxiv.org/abs/1912.08777>`__> by
+    Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
+23. `ProphetNet <https://github.com/microsoft/ProphetNet>`__ (from Microsoft Research) released with the paper
+    `ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training <https://arxiv.org/abs/2001.04063>`__
+    by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
+24. `Reformer <https://huggingface.co/transformers/model_doc/reformer.html>`__ (from Google Research) released with the
+    paper `Reformer: The Efficient Transformer <https://arxiv.org/abs/2001.04451>`__ by Nikita Kitaev, Łukasz Kaiser,
+    Anselm Levskaya.
+25. `RoBERTa <https://huggingface.co/transformers/model_doc/roberta.html>`__ (from Facebook), released together with
+    the paper a `Robustly Optimized BERT Pretraining Approach <https://arxiv.org/abs/1907.11692>`__ by Yinhan Liu, Myle
+    Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
+    ultilingual BERT into `DistilmBERT
     <https://github.com/huggingface/transformers/tree/master/examples/distillation>`__ and a German version of
     DistilBERT.
-25. `SqueezeBert <https://huggingface.co/transformers/master/model_doc/squeezebert.html>`__ released with the paper
+26. `SqueezeBert <https://huggingface.co/transformers/model_doc/squeezebert.html>`__ released with the paper
     `SqueezeBERT: What can computer vision teach NLP about efficient neural networks?
     <https://arxiv.org/abs/2006.11316>`__ by Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer.
-26. :doc:`T5 <model_doc/t5>` (from Google AI) released with the paper `Exploring the Limits of Transfer Learning with a
-    Unified Text-to-Text Transformer <https://arxiv.org/abs/1910.10683>`__ by Colin Raffel and Noam Shazeer and Adam
-    Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
-27. :doc:`Transformer-XL <model_doc/transformerxl>` (from Google/CMU) released with the paper `Transformer-XL:
-    Attentive Language Models Beyond a Fixed-Length Context <https://arxiv.org/abs/1901.02860>`__ by Zihang Dai*,
-    Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
-28. :doc:`XLM <model_doc/xlm>` (from Facebook) released together with the paper `Cross-lingual Language Model
-    Pretraining <https://arxiv.org/abs/1901.07291>`__ by Guillaume Lample and Alexis Conneau.
-29. :doc:`XLM-RoBERTa <model_doc/xlmroberta>` (from Facebook AI), released together with the paper `Unsupervised
-    Cross-lingual Representation Learning at Scale <https://arxiv.org/abs/1911.02116>`__ by Alexis Conneau*, Kartikay
-    Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke
-    Zettlemoyer and Veselin Stoyanov.
-30. :doc:`XLNet <model_doc/xlnet>` (from Google/CMU) released with the paper `​XLNet: Generalized Autoregressive
-    Pretraining for Language Understanding <https://arxiv.org/abs/1906.08237>`__ by Zhilin Yang*, Zihang Dai*, Yiming
-    Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
-31. `Other community models <https://huggingface.co/models>`__, contributed by the `community
+27. `T5 <https://huggingface.co/transformers/model_doc/t5.html>`__ (from Google AI) released with the paper `Exploring
+    the Limits of Transfer Learning with a Unified Text-to-Text Transformer <https://arxiv.org/abs/1910.10683>`__ by
+    Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi
+    Zhou and Wei Li and Peter J. Liu.
+28. `Transformer-XL <https://huggingface.co/transformers/model_doc/transformerxl.html>`__ (from Google/CMU) released
+    with the paper `Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context
+    <https://arxiv.org/abs/1901.02860>`__ by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le,
+    Ruslan Salakhutdinov.
+29. `XLM <https://huggingface.co/transformers/model_doc/xlm.html>`__ (from Facebook) released together with the paper
+    `Cross-lingual Language Model Pretraining <https://arxiv.org/abs/1901.07291>`__ by Guillaume Lample and Alexis
+    Conneau.
+30. `XLMProphetNet <https://github.com/microsoft/ProphetNet>`__ (from Microsoft Research) released with the paper
+    `ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training <https://arxiv.org/abs/2001.04063>`__
+    by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
+31. `XLM-RoBERTa <https://huggingface.co/transformers/model_doc/xlmroberta.html>`__ (from Facebook AI), released
+    together with the paper `Unsupervised Cross-lingual Representation Learning at Scale
+    <https://arxiv.org/abs/1911.02116>`__ by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary,
+    Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.
+32. `XLNet <https://huggingface.co/transformers/model_doc/xlnet.html>`__ (from Google/CMU) released with the paper
+    `​XLNet: Generalized Autoregressive Pretraining for Language Understanding <https://arxiv.org/abs/1906.08237>`__ by
+    Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
+33. `Other community models <https://huggingface.co/models>`__, contributed by the `community
     <https://huggingface.co/users>`__.
 
 .. toctree::

docs/source/pretrained_models.rst

@@ -294,10 +294,10 @@ For a list that includes community-uploaded models, refer to `https://huggingfac
 |                    | ``t5-11B``                                                 | | ~11B parameters with 24-layers, 1024-hidden-state, 65536 feed-forward hidden-state, 128-heads,                                      |
 |                    |                                                            | | Trained on English text: the Colossal Clean Crawled Corpus (C4)                                                                     |
 +--------------------+------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
-| XLM-RoBERTa        | ``xlm-roberta-base``                                       | | ~270M parameters with 12-layers, 768-hidden-state, 3072 feed-forward hidden-state, 8-heads,                                         |
+| XLM-RoBERTa        | ``xlm-roberta-base``                                       | | ~125M parameters with 12-layers, 768-hidden-state, 3072 feed-forward hidden-state, 8-heads,                                         |
 |                    |                                                            | | Trained on on 2.5 TB of newly created clean CommonCrawl data in 100 languages                                                       |
 |                    +------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
-|                    | ``xlm-roberta-large``                                      | | ~550M parameters with 24-layers, 1024-hidden-state, 4096 feed-forward hidden-state, 16-heads,                                       |
+|                    | ``xlm-roberta-large``                                      | | ~355M parameters with 24-layers, 1027-hidden-state, 4096 feed-forward hidden-state, 16-heads,                                       |
 |                    |                                                            | | Trained on 2.5 TB of newly created clean CommonCrawl data in 100 languages                                                          |
 +--------------------+------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------+
 | FlauBERT           | ``flaubert/flaubert_small_cased``                          | | 6-layer, 512-hidden, 8-heads, 54M parameters                                                                                        |

examples/seq2seq/README.md

@@ -19,7 +19,7 @@ For `bertabs` instructions, see [`bertabs/README.md`](bertabs/README.md).
 #### XSUM:
 ```bash
 cd examples/seq2seq
-wget https://cdn-datasets.huggingface.co/summarization/xsum.tar.gz
+wget https://s3.amazonaws.com/datasets.huggingface.co/summarization/xsum.tar.gz
 tar -xzvf xsum.tar.gz
 export XSUM_DIR=${PWD}/xsum
 ```
@@ -29,7 +29,7 @@ To use your own data, copy that files format. Each article to be summarized is o
 #### CNN/DailyMail
 ```bash
 cd examples/seq2seq
-wget https://cdn-datasets.huggingface.co/summarization/cnn_dm_v2.tgz
+wget https://s3.amazonaws.com/datasets.huggingface.co/summarization/cnn_dm_v2.tgz
 tar -xzvf cnn_dm_v2.tgz  # empty lines removed
 mv cnn_cln cnn_dm
 export CNN_DIR=${PWD}/cnn_dm
@@ -39,7 +39,7 @@ this should make a directory called `cnn_dm/` with 6 files.
 #### WMT16 English-Romanian Translation Data:
 download with this command:
 ```bash
-wget https://cdn-datasets.huggingface.co/translation/wmt_en_ro.tar.gz
+wget https://s3.amazonaws.com/datasets.huggingface.co/translation/wmt_en_ro.tar.gz
 tar -xzvf wmt_en_ro.tar.gz
 export ENRO_DIR=${PWD}/wmt_en_ro
 ```
@@ -47,7 +47,7 @@ this should make a directory called `wmt_en_ro/` with 6 files.
 
 #### WMT English-German:
 ```bash
-wget https://cdn-datasets.huggingface.co/translation/wmt_en_de.tgz
+wget https://s3.amazonaws.com/datasets.huggingface.co/translation/wmt_en_de.tgz
 tar -xzvf wmt_en_de.tgz
 export DATA_DIR=${PWD}/wmt_en_de
 ```

examples/seq2seq/finetune_bart_tiny.sh

@@ -1,7 +1,7 @@
 # Script for verifying that run_bart_sum can be invoked from its directory
 
 # Get tiny dataset with cnn_dm format (4 examples for train, val, test)
-wget https://cdn-datasets.huggingface.co/summarization/cnn_tiny.tgz
+wget https://s3.amazonaws.com/datasets.huggingface.co/summarization/cnn_tiny.tgz
 tar -xzvf cnn_tiny.tgz
 rm cnn_tiny.tgz
 

examples/seq2seq/finetune_trainer.py

@@ -4,7 +4,7 @@ import sys
 from dataclasses import dataclass, field
 from typing import Optional
 
-from seq2seq_trainer import Seq2SeqTrainer, arg_to_scheduler_choices
+from seq2seq_trainer import Seq2SeqTrainer
 from transformers import (
     AutoConfig,
     AutoModelForSeq2SeqLM,
@@ -63,9 +63,6 @@ class Seq2SeqTrainingArguments(TrainingArguments):
     attention_dropout: Optional[float] = field(
         default=None, metadata={"help": "Attention dropout probability. Goes into model.config."}
     )
-    lr_scheduler: Optional[str] = field(
-        default="linear", metadata={"help": f"Which lr scheduler to use. Selected in {arg_to_scheduler_choices}"}
-    )
 
 
 @dataclass

examples/seq2seq/precomputed_pseudo_labels.md

@@ -4,24 +4,24 @@ These are the generations of various large models on various large **training**
 ### Available Pseudo-labels
 | Dataset | Model                       | Link                                                                                   | Rouge Scores       | Notes                                                                                                       
 |---------|-----------------------------|----------------------------------------------------------------------------------------|--------------------|-------------------------------------------------------------------------------------------------------------
-| XSUM    | `facebook/bart-large-xsum`    | [download](https://cdn-datasets.huggingface.co/pseudo/xsum/bart_xsum_pl.tgz)          | 49.8/28.0/42.5     |                                                                                                             
-| XSUM    | `google/pegasus-xsum`         | [download](https://cdn-datasets.huggingface.co/pseudo/xsum/pegasus_xsum.tgz)          | 53.3/32.7/46.5     |                                                                                                             
-| XSUM    | `facebook/bart-large-xsum`    | [download](https://cdn-datasets.huggingface.co/pseudo/xsum/xsum_pl2_bart.tgz)         |                   | Bart pseudolabels filtered to those with Rouge2 > 10.0 w GT.                                                 
-| CNN/DM  | `sshleifer/pegasus-cnn-ft-v2` | [download](https://cdn-datasets.huggingface.co/pseudo/cnn_dm/pegasus_cnn_cnn_pls.tgz) | 47.316/26.65/44.56 | do not worry about the fact that train.source is one line shorter.                                          
-| CNN/DM  | `facebook/bart-large-cnn`     | [download](https://cdn-datasets.huggingface.co/pseudo/cnn_dm/cnn_bart_pl.tgz)         |                    | 5K (2%) are missing, there should be 282173                                                                 
-| CNN/DM  | `google/pegasus-xsum`         | [download](https://cdn-datasets.huggingface.co/pseudo/cnn_dm/pegasus_xsum_on_cnn.tgz) | 21.5/6.76/25       | extra labels for xsum distillation  Used max_source_length=512, (and all other pegasus-xsum configuration). 
-| EN-RO   | `Helsinki-NLP/opus-mt-en-ro`  | [download](https://cdn-datasets.huggingface.co/pseudo/wmt_en_ro/opus_mt_en_ro.tgz) |       |  
-| EN-RO   | `facebook/mbart-large-en-ro`  | [download](https://cdn-datasets.huggingface.co/pseudo/wmt_en_ro/mbart_large_en_ro.tgz) |       |  
+| XSUM    | `facebook/bart-large-xsum`    | [download](https://s3.amazonaws.com/datasets.huggingface.co/pseudo/xsum/bart_xsum_pl.tgz)          | 49.8/28.0/42.5     |                                                                                                             
+| XSUM    | `google/pegasus-xsum`         | [download](https://s3.amazonaws.com/datasets.huggingface.co/pseudo/xsum/pegasus_xsum.tgz)          | 53.3/32.7/46.5     |                                                                                                             
+| XSUM    | `facebook/bart-large-xsum`    | [download](https://s3.amazonaws.com/datasets.huggingface.co/pseudo/xsum/xsum_pl2_bart.tgz)         |                   | Bart pseudolabels filtered to those with Rouge2 > 10.0 w GT.                                                 
+| CNN/DM  | `sshleifer/pegasus-cnn-ft-v2` | [download](https://s3.amazonaws.com/datasets.huggingface.co/pseudo/cnn_dm/pegasus_cnn_cnn_pls.tgz) | 47.316/26.65/44.56 | do not worry about the fact that train.source is one line shorter.                                          
+| CNN/DM  | `facebook/bart-large-cnn`     | [download](https://s3.amazonaws.com/datasets.huggingface.co/pseudo/cnn_dm/cnn_bart_pl.tgz)         |                    | 5K (2%) are missing, there should be 282173                                                                 
+| CNN/DM  | `google/pegasus-xsum`         | [download](https://s3.amazonaws.com/datasets.huggingface.co/pseudo/cnn_dm/pegasus_xsum_on_cnn.tgz) | 21.5/6.76/25       | extra labels for xsum distillation  Used max_source_length=512, (and all other pegasus-xsum configuration). 
+| EN-RO   | `Helsinki-NLP/opus-mt-en-ro`  | [download](https://s3.amazonaws.com/datasets.huggingface.co/pseudo/wmt_en_ro/opus_mt_en_ro.tgz) |       |  
+| EN-RO   | `facebook/mbart-large-en-ro`  | [download](https://s3.amazonaws.com/datasets.huggingface.co/pseudo/wmt_en_ro/mbart_large_en_ro.tgz) |       |  
 
 
 (EN_RO = WMT 2016 English-Romanian).
 
 Example Download Command:
 ```bash
-curl -S https://cdn-datasets.huggingface.co/pseudo/xsum/bart_xsum_pl.tgz | tar -xvz -C .
+curl -S https://s3.amazonaws.com/datasets.huggingface.co/pseudo/xsum/bart_xsum_pl.tgz | tar -xvz -C .
 ```
 ### Generating New Pseudolabels
-Here is the command I used to generate the pseudolabels in the second row of the table, after downloading XSUM from [here](https://cdn-datasets.huggingface.co/summarization/xsum.tar.gz). 
+Here is the command I used to generate the pseudolabels in the second row of the table, after downloading XSUM from [here](https://s3.amazonaws.com/datasets.huggingface.co/summarization/xsum.tar.gz). 
 
 ```bash                                                                         
 python -m torch.distributed.launch --nproc_per_node=8 run_distributed_eval.py \

examples/seq2seq/seq2seq_trainer.py

@@ -8,16 +8,7 @@ from torch.utils.data import DistributedSampler, RandomSampler
 from transformers import Trainer
 from transformers.configuration_fsmt import FSMTConfig
 from transformers.file_utils import is_torch_tpu_available
-from transformers.optimization import (
-    Adafactor,
-    AdamW,
-    get_constant_schedule,
-    get_constant_schedule_with_warmup,
-    get_cosine_schedule_with_warmup,
-    get_cosine_with_hard_restarts_schedule_with_warmup,
-    get_linear_schedule_with_warmup,
-    get_polynomial_decay_schedule_with_warmup,
-)
+from transformers.optimization import Adafactor, AdamW, get_linear_schedule_with_warmup
 from transformers.trainer_pt_utils import get_tpu_sampler
 
 
@@ -29,16 +20,6 @@ except ImportError:
 
 logger = logging.getLogger(__name__)
 
-arg_to_scheduler = {
-    "linear": get_linear_schedule_with_warmup,
-    "cosine": get_cosine_schedule_with_warmup,
-    "cosine_w_restarts": get_cosine_with_hard_restarts_schedule_with_warmup,
-    "polynomial": get_polynomial_decay_schedule_with_warmup,
-    "constant": get_constant_schedule,
-    "constant_w_warmup": get_constant_schedule_with_warmup,
-}
-arg_to_scheduler_choices = sorted(arg_to_scheduler.keys())
-
 
 class Seq2SeqTrainer(Trainer):
     def __init__(self, config, data_args, *args, **kwargs):
@@ -81,21 +62,9 @@ class Seq2SeqTrainer(Trainer):
                 )
 
         if self.lr_scheduler is None:
-            self.lr_scheduler = self._get_lr_scheduler(num_training_steps)
-        else:  # ignoring --lr_scheduler
-            logger.warn("scheduler is passed to `Seq2SeqTrainer`, `--lr_scheduler` arg is ignored.")
-
-    def _get_lr_scheduler(self, num_training_steps):
-        schedule_func = arg_to_scheduler[self.args.lr_scheduler]
-        if self.args.lr_scheduler == "constant":
-            scheduler = schedule_func(self.optimizer)
-        elif self.args.lr_scheduler == "constant_w_warmup":
-            scheduler = schedule_func(self.optimizer, num_warmup_steps=self.args.warmup_steps)
-        else:
-            scheduler = schedule_func(
+            self.lr_scheduler = get_linear_schedule_with_warmup(
                 self.optimizer, num_warmup_steps=self.args.warmup_steps, num_training_steps=num_training_steps
             )
-        return scheduler
 
     def _get_train_sampler(self) -> Optional[torch.utils.data.sampler.Sampler]:
         if isinstance(self.train_dataset, torch.utils.data.IterableDataset):

examples/text-classification/run_tf_text_classification.py

@@ -96,9 +96,6 @@ def get_tfds(
         else None
     )
 
-    if train_ds is not None:
-        train_ds = train_ds.apply(tf.data.experimental.assert_cardinality(len(ds[datasets.Split.TRAIN])))
-
     val_ds = (
         tf.data.Dataset.from_generator(
             gen_val,
@@ -109,9 +106,6 @@ def get_tfds(
         else None
     )
 
-    if val_ds is not None:
-        val_ds = val_ds.apply(tf.data.experimental.assert_cardinality(len(ds[datasets.Split.VALIDATION])))
-
     test_ds = (
         tf.data.Dataset.from_generator(
             gen_test,
@@ -122,9 +116,6 @@ def get_tfds(
         else None
     )
 
-    if test_ds is not None:
-        test_ds = test_ds.apply(tf.data.experimental.assert_cardinality(len(ds[datasets.Split.TEST])))
-
     return train_ds, val_ds, test_ds, label2id
 
 

model_cards/joeddav/xlm-roberta-large-xnli/README.md

@@ -9,7 +9,6 @@ datasets:
 - xnli
 widget:
 - text: "За кого вы голосуете в 2020 году? <sep> This text is about politique."
-license: mit
 ---
 
 # xlm-roberta-large-xnli

model_cards/microsoft/prophetnet-large-uncased-cnndm/README.md

@@ -0,0 +1,30 @@
+## prophetnet-large-uncased-cnndm
+Fine-tuned weights(converted from [original fairseq version repo](https://github.com/microsoft/ProphetNet)) for [ProphetNet](https://arxiv.org/abs/2001.04063) on summarization task CNN/DailyMail.  
+ProphetNet is a new pre-trained language model for sequence-to-sequence learning with a novel self-supervised objective called future n-gram prediction.  
+ProphetNet is able to predict more future tokens with a n-stream decoder. The original implementation is Fairseq version at [github repo](https://github.com/microsoft/ProphetNet).   
+
+### Usage
+A quick usage is like: 
+```
+from transformers import ProphetNetTokenizer, ProphetNetForConditionalGeneration, ProphetNetConfig
+
+model = ProphetNetForConditionalGeneration.from_pretrained('microsoft/prophetnet-large-uncased-cnndm')
+tokenizer = ProphetNetTokenizer.from_pretrained('microsoft/prophetnet-large-uncased-cnndm')
+
+ARTICLE_TO_SUMMARIZE = "USTC was founded in Beijing by the Chinese Academy of Sciences (CAS) in September 1958. The Director of CAS, Mr. Guo Moruo was appointed the first president of USTC. USTC's founding mission was to develop a high-level science and technology workforce, as deemed critical for development of China's economy, defense, and science and technology education. The establishment was hailed as \"A Major Event in the History of Chinese Education and Science.\" CAS has supported USTC by combining most of its institutes with the departments of the university. USTC is listed in the top 16 national key universities, becoming the youngest national key university.".lower()
+inputs = tokenizer([ARTICLE_TO_SUMMARIZE], max_length=100, return_tensors='pt')
+
+# Generate Summary
+summary_ids = model.generate(inputs['input_ids'], num_beams=4, max_length=512, early_stopping=True)
+print([tokenizer.decode(g) for g in summary_ids])
+```
+Here, [X_SEP] is used as a special token to seperate sentences.
+### Citation
+```bibtex
+@article{yan2020prophetnet,
+  title={Prophetnet: Predicting future n-gram for sequence-to-sequence pre-training},
+  author={Yan, Yu and Qi, Weizhen and Gong, Yeyun and Liu, Dayiheng and Duan, Nan and Chen, Jiusheng and Zhang, Ruofei and Zhou, Ming},
+  journal={arXiv preprint arXiv:2001.04063},
+  year={2020}
+}
+```

model_cards/microsoft/prophetnet-large-uncased/README.md

@@ -0,0 +1,17 @@
+## prophetnet-large-uncased
+Pretrained weights for [ProphetNet](https://arxiv.org/abs/2001.04063).  
+ProphetNet is a new pre-trained language model for sequence-to-sequence learning with a novel self-supervised objective called future n-gram prediction.  
+ProphetNet is able to predict more future tokens with a n-stream decoder. The original implementation is Fairseq version at [github repo](https://github.com/microsoft/ProphetNet).   
+
+### Usage
+Please see [the official repository](https://github.com/microsoft/ProphetNet) for details.
+
+### Citation
+```bibtex
+@article{yan2020prophetnet,
+  title={Prophetnet: Predicting future n-gram for sequence-to-sequence pre-training},
+  author={Yan, Yu and Qi, Weizhen and Gong, Yeyun and Liu, Dayiheng and Duan, Nan and Chen, Jiusheng and Zhang, Ruofei and Zhou, Ming},
+  journal={arXiv preprint arXiv:2001.04063},
+  year={2020}
+}
+```

model_cards/microsoft/xprophetnet-large-wiki100-cased-xglue-ntg/README.md

@@ -0,0 +1,32 @@
+## xprophetnet-large-wiki100-cased-xglue-ntg
+Cross-lingual version [ProphetNet](https://arxiv.org/abs/2001.04063), pretrained on [wiki100 xGLUE dataset](https://arxiv.org/abs/2004.01401) and finetuned on xGLUE cross-lingual News Titles Generation task.  
+ProphetNet is a new pre-trained language model for sequence-to-sequence learning with a novel self-supervised objective called future n-gram prediction.  
+ProphetNet is able to predict more future tokens with a n-stream decoder. The original implementation is Fairseq version at [github repo](https://github.com/microsoft/ProphetNet).   
+
+xProphetNet is also served as the baseline model for xGLUE cross-lingual natural language generation tasks.  
+For xGLUE corss-lingual NLG tasks, xProphetNet is finetuned with English data, but inference with both English and other zero-shot language data.  
+### Usage
+A quick usage is like: 
+```
+from transformers import ProphetNetTokenizer, ProphetNetForConditionalGeneration, ProphetNetConfig
+
+model = ProphetNetForConditionalGeneration.from_pretrained('microsoft/xprophetnet-large-wiki100-cased-xglue-ntg')
+tokenizer = ProphetNetTokenizer.from_pretrained('microsoft/xprophetnet-large-wiki100-cased-xglue-ntg')
+
+EN_SENTENCE = "Microsoft Corporation intends to officially end free support for the Windows 7 operating system after January 14, 2020, according to the official portal of the organization. From that day, users of this system will not be able to receive security updates, which could make their computers vulnerable to cyber attacks."
+RU_SENTENCE = "орпорация Microsoft намерена официально прекратить бесплатную поддержку операционной системы Windows 7 после 14 января 2020 года, сообщается на официальном портале организации . С указанного дня пользователи этой системы не смогут получать обновления безопасности, из-за чего их компьютеры могут стать уязвимыми к кибератакам."
+ZH_SENTENCE = "根据该组织的官方门户网站，微软公司打算在2020年1月14日之后正式终止对Windows 7操作系统的免费支持。从那时起，该系统的用户将无法接收安全更新，这可能会使他们的计算机容易受到网络攻击。"
+inputs = tokenizer([EN_SENTENCE, RU_SENTENCE, ZH_SENTENCE], padding=True, max_length=256, return_tensors='pt')
+
+summary_ids = model.generate(inputs['input_ids'], num_beams=4, max_length=100, early_stopping=True)
+print([tokenizer.decode(g) for g in summary_ids])  
+```
+### Citation
+```bibtex
+@article{yan2020prophetnet,
+  title={Prophetnet: Predicting future n-gram for sequence-to-sequence pre-training},
+  author={Yan, Yu and Qi, Weizhen and Gong, Yeyun and Liu, Dayiheng and Duan, Nan and Chen, Jiusheng and Zhang, Ruofei and Zhou, Ming},
+  journal={arXiv preprint arXiv:2001.04063},
+  year={2020}
+}
+```

model_cards/microsoft/xprophetnet-large-wiki100-cased-xglue-qg/README.md

@@ -0,0 +1,31 @@
+## xprophetnet-large-wiki100-cased-xglue-ntg
+Cross-lingual version [ProphetNet](https://arxiv.org/abs/2001.04063), pretrained on [wiki100 xGLUE dataset](https://arxiv.org/abs/2004.01401) and finetuned on xGLUE cross-lingual Question Generation task.  
+ProphetNet is a new pre-trained language model for sequence-to-sequence learning with a novel self-supervised objective called future n-gram prediction.  
+ProphetNet is able to predict more future tokens with a n-stream decoder. The original implementation is Fairseq version at [github repo](https://github.com/microsoft/ProphetNet).   
+
+xProphetNet is also served as the baseline model for xGLUE cross-lingual natural language generation tasks.  
+For xGLUE corss-lingual NLG tasks, xProphetNet is finetuned with English data, but inference with both English and other zero-shot language data.  
+### Usage
+A quick usage is like: 
+```
+from transformers import ProphetNetTokenizer, ProphetNetForConditionalGeneration, ProphetNetConfig
+
+model = ProphetNetForConditionalGeneration.from_pretrained('microsoft/xprophetnet-large-wiki100-cased-xglue-qg')
+tokenizer = ProphetNetTokenizer.from_pretrained('microsoft/xprophetnet-large-wiki100-cased-xglue-qg')
+
+EN_SENTENCE = "Google left China in 2010"
+ZH_SENTENCE = "Google在2010年离开中国"
+inputs = tokenizer([EN_SENTENCE, ZH_SENTENCE], padding=True, max_length=256, return_tensors='pt')
+
+summary_ids = model.generate(inputs['input_ids'], num_beams=4, max_length=100, early_stopping=True)
+print([tokenizer.decode(g) for g in summary_ids])  
+```
+### Citation
+```bibtex
+@article{yan2020prophetnet,
+  title={Prophetnet: Predicting future n-gram for sequence-to-sequence pre-training},
+  author={Yan, Yu and Qi, Weizhen and Gong, Yeyun and Liu, Dayiheng and Duan, Nan and Chen, Jiusheng and Zhang, Ruofei and Zhou, Ming},
+  journal={arXiv preprint arXiv:2001.04063},
+  year={2020}
+}
+```

model_cards/microsoft/xprophetnet-large-wiki100-cased/README.md

@@ -0,0 +1,19 @@
+## xprophetnet-large-wiki100-cased
+Cross-lingual version [ProphetNet](https://arxiv.org/abs/2001.04063), pretrained on [wiki100 xGLUE dataset](https://arxiv.org/abs/2004.01401).  
+ProphetNet is a new pre-trained language model for sequence-to-sequence learning with a novel self-supervised objective called future n-gram prediction.  
+ProphetNet is able to predict more future tokens with a n-stream decoder. The original implementation is Fairseq version at [github repo](https://github.com/microsoft/ProphetNet).   
+
+xProphetNet is also served as the baseline model for xGLUE cross-lingual natural language generation tasks.  
+For xGLUE corss-lingual NLG tasks, xProphetNet is finetuned with English data, but inference with both English and other zero-shot language data.  
+### Usage
+Please see [the official repository](https://github.com/microsoft/ProphetNet/tree/master/xProphetNet) for details.
+
+### Citation
+```bibtex
+@article{yan2020prophetnet,
+  title={Prophetnet: Predicting future n-gram for sequence-to-sequence pre-training},
+  author={Yan, Yu and Qi, Weizhen and Gong, Yeyun and Liu, Dayiheng and Duan, Nan and Chen, Jiusheng and Zhang, Ruofei and Zhou, Ming},
+  journal={arXiv preprint arXiv:2001.04063},
+  year={2020}
+}
+```

setup.py

@@ -3,10 +3,7 @@ Simple check list from AllenNLP repo: https://github.com/allenai/allennlp/blob/m
 
 To create the package for pypi.
 
-1. Change the version in __init__.py, setup.py as well as docs/source/conf.py. Remove the master from the links in
-   the new models of the README:
-   (https://huggingface.co/transformers/master/model_doc/ -> https://huggingface.co/transformers/model_doc/)
-   then run `make fix-copies` to fix the index of the documentation.
+1. Change the version in __init__.py, setup.py as well as docs/source/conf.py.
 
 2. Unpin specific versions from setup.py that use a git install.
 

src/transformers/__init__.py

@@ -54,6 +54,7 @@ from .configuration_mmbt import MMBTConfig
 from .configuration_mobilebert import MOBILEBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, MobileBertConfig
 from .configuration_openai import OPENAI_GPT_PRETRAINED_CONFIG_ARCHIVE_MAP, OpenAIGPTConfig
 from .configuration_pegasus import PegasusConfig
+from .configuration_prophetnet import PROPHETNET_PRETRAINED_CONFIG_ARCHIVE_MAP, ProphetNetConfig
 from .configuration_rag import RagConfig
 from .configuration_reformer import REFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, ReformerConfig
 from .configuration_retribert import RETRIBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, RetriBertConfig
@@ -63,6 +64,7 @@ from .configuration_t5 import T5_PRETRAINED_CONFIG_ARCHIVE_MAP, T5Config
 from .configuration_transfo_xl import TRANSFO_XL_PRETRAINED_CONFIG_ARCHIVE_MAP, TransfoXLConfig
 from .configuration_utils import PretrainedConfig
 from .configuration_xlm import XLM_PRETRAINED_CONFIG_ARCHIVE_MAP, XLMConfig
+from .configuration_xlm_prophetnet import XLM_PROPHETNET_PRETRAINED_CONFIG_ARCHIVE_MAP, XLMProphetNetConfig
 from .configuration_xlm_roberta import XLM_ROBERTA_PRETRAINED_CONFIG_ARCHIVE_MAP, XLMRobertaConfig
 from .configuration_xlnet import XLNET_PRETRAINED_CONFIG_ARCHIVE_MAP, XLNetConfig
 from .data import (
@@ -185,6 +187,7 @@ from .tokenization_mobilebert import MobileBertTokenizer, MobileBertTokenizerFas
 from .tokenization_openai import OpenAIGPTTokenizer, OpenAIGPTTokenizerFast
 from .tokenization_pegasus import PegasusTokenizer, PegasusTokenizerFast
 from .tokenization_phobert import PhobertTokenizer
+from .tokenization_prophetnet import ProphetNetTokenizer
 from .tokenization_rag import RagTokenizer
 from .tokenization_reformer import ReformerTokenizer, ReformerTokenizerFast
 from .tokenization_retribert import RetriBertTokenizer, RetriBertTokenizerFast
@@ -203,6 +206,7 @@ from .tokenization_utils_base import (
 )
 from .tokenization_utils_fast import PreTrainedTokenizerFast
 from .tokenization_xlm import XLMTokenizer
+from .tokenization_xlm_prophetnet import XLMProphetNetTokenizer
 from .tokenization_xlm_roberta import XLMRobertaTokenizer, XLMRobertaTokenizerFast
 from .tokenization_xlnet import SPIECE_UNDERLINE, XLNetTokenizer, XLNetTokenizerFast
 
@@ -443,6 +447,12 @@ if is_torch_available():
         load_tf_weights_in_openai_gpt,
     )
     from .modeling_pegasus import PegasusForConditionalGeneration
+    from .modeling_prophetnet import (
+        PROPHETNET_PRETRAINED_MODEL_ARCHIVE_LIST,
+        ProphetNetForConditionalGeneration,
+        ProphetNetModel,
+        ProphetNetPreTrainedModel,
+    )
     from .modeling_rag import RagModel, RagSequenceForGeneration, RagTokenForGeneration
     from .modeling_reformer import (
         REFORMER_PRETRAINED_MODEL_ARCHIVE_LIST,
@@ -503,6 +513,11 @@ if is_torch_available():
         XLMPreTrainedModel,
         XLMWithLMHeadModel,
     )
+    from .modeling_xlm_prophetnet import (
+        XLM_PROPHETNET_PRETRAINED_MODEL_ARCHIVE_LIST,
+        XLMProphetNetForConditionalGeneration,
+        XLMProphetNetModel,
+    )
     from .modeling_xlm_roberta import (
         XLM_ROBERTA_PRETRAINED_MODEL_ARCHIVE_LIST,
         XLMRobertaForCausalLM,

src/transformers/configuration_auto.py

@@ -41,6 +41,7 @@ from .configuration_mbart import MBART_PRETRAINED_CONFIG_ARCHIVE_MAP, MBartConfi
 from .configuration_mobilebert import MobileBertConfig
 from .configuration_openai import OPENAI_GPT_PRETRAINED_CONFIG_ARCHIVE_MAP, OpenAIGPTConfig
 from .configuration_pegasus import PegasusConfig
+from .configuration_prophetnet import PROPHETNET_PRETRAINED_CONFIG_ARCHIVE_MAP, ProphetNetConfig
 from .configuration_rag import RagConfig
 from .configuration_reformer import ReformerConfig
 from .configuration_retribert import RETRIBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, RetriBertConfig
@@ -50,6 +51,7 @@ from .configuration_t5 import T5_PRETRAINED_CONFIG_ARCHIVE_MAP, T5Config
 from .configuration_transfo_xl import TRANSFO_XL_PRETRAINED_CONFIG_ARCHIVE_MAP, TransfoXLConfig
 from .configuration_utils import PretrainedConfig
 from .configuration_xlm import XLM_PRETRAINED_CONFIG_ARCHIVE_MAP, XLMConfig
+from .configuration_xlm_prophetnet import XLM_PROPHETNET_PRETRAINED_CONFIG_ARCHIVE_MAP, XLMProphetNetConfig
 from .configuration_xlm_roberta import XLM_ROBERTA_PRETRAINED_CONFIG_ARCHIVE_MAP, XLMRobertaConfig
 from .configuration_xlnet import XLNET_PRETRAINED_CONFIG_ARCHIVE_MAP, XLNetConfig
 
@@ -84,6 +86,8 @@ ALL_PRETRAINED_CONFIG_ARCHIVE_MAP = dict(
         DPR_PRETRAINED_CONFIG_ARCHIVE_MAP,
         DEBERTA_PRETRAINED_CONFIG_ARCHIVE_MAP,
         SQUEEZEBERT_PRETRAINED_CONFIG_ARCHIVE_MAP,
+        XLM_PROPHETNET_PRETRAINED_CONFIG_ARCHIVE_MAP,
+        PROPHETNET_PRETRAINED_CONFIG_ARCHIVE_MAP,
     ]
     for key, value, in pretrained_map.items()
 )
@@ -116,6 +120,8 @@ CONFIG_MAPPING = OrderedDict(
         ("gpt2", GPT2Config),
         ("transfo-xl", TransfoXLConfig),
         ("xlnet", XLNetConfig),
+        ("xlm-prophetnet", XLMProphetNetConfig),
+        ("prophetnet", ProphetNetConfig),
         ("xlm", XLMConfig),
         ("ctrl", CTRLConfig),
         ("electra", ElectraConfig),
@@ -164,6 +170,8 @@ MODEL_NAMES_MAPPING = OrderedDict(
         ("layoutlm", "LayoutLM"),
         ("dpr", "DPR"),
         ("rag", "RAG"),
+        ("xlm-prophetnet", "XLMProphetNet"),
+        ("prophetnet", "ProphetNet"),
     ]
 )
 

src/transformers/configuration_prophetnet.py

@@ -0,0 +1,106 @@
+# coding=utf-8
+# Copyright 2020 The Microsoft Authors and The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" ProphetNet model configuration """
+
+
+import logging
+
+from .configuration_utils import PretrainedConfig
+
+
+logger = logging.getLogger(__name__)
+
+PROPHETNET_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "microsoft/prophetnet-large-uncased": "https://s3.amazonaws.com/models.huggingface.co/bert/microsoft/prophetnet-large-uncased/config.json",
+}
+
+
+class ProphetNetConfig(PretrainedConfig):
+    r"""
+    Configuration class for ProphetNet. Parameters are renamed from the fairseq implementation
+    """
+    model_type = "prophetnet"
+
+    def __init__(
+        self,
+        activation_dropout=0.1,
+        activation_function="gelu",
+        vocab_size=30522,
+        hidden_size=1024,
+        encoder_ffn_dim=4096,
+        num_encoder_layers=12,
+        num_encoder_attention_heads=16,
+        decoder_ffn_dim=4096,
+        num_decoder_layers=12,
+        num_decoder_attention_heads=16,
+        encoder_layerdrop=0.1,
+        decoder_layerdrop=0.1,
+        attention_dropout=0.1,
+        dropout=0.1,
+        max_position_embeddings=512,
+        init_std=0.02,
+        is_encoder_decoder=True,
+        pad_token_id=0,
+        bos_token_id=1,
+        eos_token_id=2,
+        decoder_start_token_id=0,
+        ngram=2,
+        num_buckets=32,
+        relative_max_distance=128,
+        disable_ngram_loss=False,
+        eps=0.0,
+        **common_kwargs
+    ):
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            is_encoder_decoder=is_encoder_decoder,
+            decoder_start_token_id=decoder_start_token_id,
+            **common_kwargs,
+        )
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.encoder_ffn_dim = encoder_ffn_dim
+        self.num_encoder_layers = num_encoder_layers
+        self.num_encoder_attention_heads = num_encoder_attention_heads
+        self.encoder_layerdrop = encoder_layerdrop
+        self.decoder_layerdrop = decoder_layerdrop
+        self.decoder_ffn_dim = decoder_ffn_dim
+        self.num_decoder_layers = num_decoder_layers
+        self.num_decoder_attention_heads = num_decoder_attention_heads
+        self.max_position_embeddings = max_position_embeddings
+        self.init_std = init_std  # Normal(0, this parameter)
+        self.activation_function = activation_function
+
+        # parameters for prophetnet
+        self.ngram = ngram
+        self.num_buckets = num_buckets
+        self.relative_max_distance = relative_max_distance
+        self.disable_ngram_loss = disable_ngram_loss
+        self.eps = eps
+
+        # 3 Types of Dropout
+        self.attention_dropout = attention_dropout
+        self.activation_dropout = activation_dropout
+        self.dropout = dropout
+
+    @property
+    def num_attention_heads(self) -> int:
+        return self.num_encoder_attention_heads
+
+    @property
+    def num_hidden_layers(self) -> int:
+        return self.num_encoder_layers + self.num_decoder_layers

src/transformers/configuration_xlm_prophetnet.py

@@ -0,0 +1,22 @@
+""" XLM-ProphetNet model configuration """
+
+
+import logging
+
+from .configuration_prophetnet import ProphetNetConfig
+
+
+logger = logging.getLogger(__name__)
+
+XLM_PROPHETNET_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "microsoft/xprophetnet-large-wiki100-cased": "https://s3.amazonaws.com/models.huggingface.co/bert/microsoft/xprophetnet-large-wiki100-cased/config.json",
+}
+
+
+class XLMProphetNetConfig(ProphetNetConfig):
+    """
+    This class overrides :class:`~transformers.RobertaConfig`. Please check the
+    superclass for the appropriate documentation alongside usage examples.
+    """
+
+    model_type = "xlm-prophetnet"

src/transformers/convert_prophetnet_original_pytorch_checkpoint_to_pytorch.py

@@ -0,0 +1,144 @@
+# coding=utf-8
+# Copyright 2018 The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Convert RoBERTa checkpoint."""
+
+
+import argparse
+
+from transformers import logging
+from transformers.modeling_prophetnet import ProphetNetForConditionalGeneration
+from transformers.modeling_xlm_prophetnet import XLMProphetNetForConditionalGeneration
+
+# transformers_old should correspond to branch `save_old_prophetnet_model_structure` here
+from transformers_old.modeling_prophetnet import (
+    ProphetNetForConditionalGeneration as ProphetNetForConditionalGenerationOld,
+)
+from transformers_old.modeling_xlm_prophetnet import (
+    XLMProphetNetForConditionalGeneration as XLMProphetNetForConditionalGenerationOld,
+)
+
+
+logger = logging.get_logger(__name__)
+logging.set_verbosity_info()
+
+
+def convert_prophetnet_checkpoint_to_pytorch(prophetnet_checkpoint_path: str, pytorch_dump_folder_path: str):
+    """
+    Copy/paste/tweak prohpetnet's weights to our prophetnet structure.
+    """
+    if "xprophetnet" in prophetnet_checkpoint_path:
+        prophet_old = XLMProphetNetForConditionalGenerationOld.from_pretrained(prophetnet_checkpoint_path)
+        prophet, loading_info = XLMProphetNetForConditionalGeneration.from_pretrained(
+            prophetnet_checkpoint_path, output_loading_info=True
+        )
+    else:
+        prophet_old = ProphetNetForConditionalGenerationOld.from_pretrained(prophetnet_checkpoint_path)
+        prophet, loading_info = ProphetNetForConditionalGeneration.from_pretrained(
+            prophetnet_checkpoint_path, output_loading_info=True
+        )
+
+    mapping = {
+        "ngram_self_attn_layer_norm": "self_attn_layer_norm",
+        "cross_attn": "encoder_attn",
+        "cross_attn_layer_norm": "encoder_attn_layer_norm",
+        "feed_forward_layer_norm": "final_layer_norm",
+        "feed_forward": "",
+        "intermediate": "fc1",
+        "output": "fc2",
+        "key_proj_bias": "bias_k",
+        "value_proj_bias": "bias_v",
+        "key_proj_weight": "k_proj_weight",
+        "value_proj_weight": "v_proj_weight",
+        "query_proj_weight": "q_proj_weight",
+        "key_proj": "k_proj",
+        "value_proj": "v_proj",
+        "query_proj": "q_proj",
+        "word_embeddings": "embed_tokens",
+        "embeddings_layer_norm": "emb_layer_norm",
+    }
+
+    for key in loading_info["missing_keys"]:
+        attributes = key.split(".")
+
+        if attributes[0] == "lm_head":
+            model = prophet
+            old_model = prophet_old
+        else:
+            model = prophet.prophetnet
+            old_model = prophet_old.model
+
+        is_key_init = False
+        for attribute in attributes:
+            if attribute in mapping:
+                old_attribute = mapping[attribute]
+            else:
+                old_attribute = attribute
+
+            if attribute == "weight":
+                assert old_model.weight.shape == model.weight.shape, "Shapes have to match!"
+                model.weight = old_model.weight
+                logger.info(f"{attribute} is initialized.")
+                is_key_init = True
+                break
+            elif attribute == "bias":
+                assert old_model.bias.shape == model.bias.shape, "Shapes have to match!"
+                model.bias = old_model.bias
+                logger.info(f"{attribute} is initialized")
+                is_key_init = True
+                break
+            elif attribute in [
+                "in_proj_weight",
+                "key_proj_weight",
+                "value_proj_weight",
+                "query_proj_weight",
+                "in_proj_bias",
+                "key_proj_bias",
+                "value_proj_bias",
+            ]:
+                old_model_weight = getattr(old_model, old_attribute)
+                assert getattr(model, attribute).shape == old_model_weight.shape, "Shapes have to match!"
+                setattr(model, attribute, old_model_weight)
+                is_key_init = True
+                break
+
+            if attribute.isdigit():
+                model = model[int(attribute)]
+                old_model = old_model[int(old_attribute)]
+            else:
+                model = getattr(model, attribute)
+
+                if old_attribute == "":
+                    old_model = old_model
+                else:
+                    old_model = getattr(old_model, old_attribute)
+
+        if not is_key_init:
+            raise ValueError(f"{key} was not correctly initialized!")
+
+    print(f"Saving model to {pytorch_dump_folder_path}")
+    prophet.save_pretrained(pytorch_dump_folder_path)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    # Required parameters
+    parser.add_argument(
+        "--prophetnet_checkpoint_path", default=None, type=str, required=True, help="Path the official PyTorch dump."
+    )
+    parser.add_argument(
+        "--pytorch_dump_folder_path", default=None, type=str, required=True, help="Path to the output PyTorch model."
+    )
+    args = parser.parse_args()
+    convert_prophetnet_checkpoint_to_pytorch(args.prophetnet_checkpoint_path, args.pytorch_dump_folder_path)

src/transformers/modeling_auto.py

@@ -43,6 +43,7 @@ from .configuration_auto import (
     MobileBertConfig,
     OpenAIGPTConfig,
     PegasusConfig,
+    ProphetNetConfig,
     ReformerConfig,
     RetriBertConfig,
     RobertaConfig,
@@ -50,6 +51,7 @@ from .configuration_auto import (
     T5Config,
     TransfoXLConfig,
     XLMConfig,
+    XLMProphetNetConfig,
     XLMRobertaConfig,
     XLNetConfig,
     replace_list_option_in_docstrings,
@@ -155,6 +157,7 @@ from .modeling_mobilebert import (
 )
 from .modeling_openai import OpenAIGPTLMHeadModel, OpenAIGPTModel
 from .modeling_pegasus import PegasusForConditionalGeneration
+from .modeling_prophetnet import ProphetNetForConditionalGeneration, ProphetNetModel
 from .modeling_rag import (  # noqa: F401 - need to import all RagModels to be in globals() function
     RagModel,
     RagSequenceForGeneration,
@@ -194,6 +197,7 @@ from .modeling_xlm import (
     XLMModel,
     XLMWithLMHeadModel,
 )
+from .modeling_xlm_prophetnet import XLMProphetNetForConditionalGeneration, XLMProphetNetModel
 from .modeling_xlm_roberta import (
     XLMRobertaForCausalLM,
     XLMRobertaForMaskedLM,
@@ -247,6 +251,8 @@ MODEL_MAPPING = OrderedDict(
         (BertGenerationConfig, BertGenerationEncoder),
         (DebertaConfig, DebertaModel),
         (DPRConfig, DPRQuestionEncoder),
+        (XLMProphetNetConfig, XLMProphetNetModel),
+        (ProphetNetConfig, ProphetNetModel),
     ]
 )
 
@@ -359,6 +365,8 @@ MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING = OrderedDict(
         (BartConfig, BartForConditionalGeneration),
         (FSMTConfig, FSMTForConditionalGeneration),
         (EncoderDecoderConfig, EncoderDecoderModel),
+        (XLMProphetNetConfig, XLMProphetNetForConditionalGeneration),
+        (ProphetNetConfig, ProphetNetForConditionalGeneration),
     ]
 )
 

src/transformers/modeling_lxmert.py

@@ -964,7 +964,7 @@ class LxmertModel(LxmertPreTrainedModel):
         # Process the visual attention mask
         if visual_attention_mask is not None:
             extended_visual_attention_mask = visual_attention_mask.unsqueeze(1).unsqueeze(2)
-            extended_visual_attention_mask = extended_visual_attention_mask.to(dtype=self.dtype)
+            extended_visual_attention_mask = extended_visual_attention_mask.to(dtype=next(self.parameters()).dtype)
             extended_visual_attention_mask = (1.0 - extended_visual_attention_mask) * -10000.0
         else:
             extended_visual_attention_mask = None

src/transformers/modeling_utils.py

@@ -489,12 +489,17 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin):
             self._tie_or_clone_weights(output_embeddings, self.get_input_embeddings())
 
         if self.config.is_encoder_decoder and self.config.tie_encoder_decoder:
+            if hasattr(self, self.base_model_prefix):
+                self = getattr(self, self.base_model_prefix)
             self._tie_encoder_decoder_weights(self.encoder, self.decoder, self.base_model_prefix)
 
     @staticmethod
     def _tie_encoder_decoder_weights(encoder: nn.Module, decoder: nn.Module, base_model_prefix: str):
         uninitialized_encoder_weights: List[str] = []
-        assert decoder.__class__ == encoder.__class__, f"{decoder.__class__} and {encoder.__class__} have to be equal."
+        if decoder.__class__ != encoder.__class__:
+            logger.info(
+                f"{decoder.__class__} and {encoder.__class__} are not equal. In this case make sure that all encoder weights are correctly initialized."
+            )
 
         def tie_encoder_to_decoder_recursively(
             decoder_pointer: nn.Module,
@@ -527,7 +532,9 @@ class PreTrainedModel(nn.Module, ModuleUtilsMixin, GenerationMixin):
                     if name.isdigit():
                         encoder_name = str(int(name) + encoder_layer_pos)
                         decoder_name = name
-                        if not isinstance(decoder_modules[decoder_name], type(encoder_modules[encoder_name])):
+                        if not isinstance(decoder_modules[decoder_name], type(encoder_modules[encoder_name])) and len(
+                            encoder_modules
+                        ) != len(decoder_modules):
                             # this can happen if the name corresponds to the position in a list module list of layers
                             # in this case the decoder has added a cross-attention that the encoder does not have
                             # thus skip this step and substract one layer pos from encoder

src/transformers/modeling_xlm_prophetnet.py

@@ -0,0 +1,31 @@
+from .configuration_xlm_prophetnet import XLMProphetNetConfig
+from .modeling_prophetnet import ProphetNetForConditionalGeneration, ProphetNetModel
+from .utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+_TOKENIZER_FOR_DOC = "XLMProphetNetTokenizer"
+
+XLM_PROPHETNET_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "microsoft/xprophetnet-large-wiki100-cased",
+    # See all ProphetNet models at https://huggingface.co/models?filter=xprophetnet
+]
+
+
+class XLMProphetNetModel(ProphetNetModel):
+    """
+    This class overrides :class:`~transformers.ProphetNetModel`. Please check the
+    superclass for the appropriate documentation alongside usage examples.
+    """
+
+    config_class = XLMProphetNetConfig
+
+
+class XLMProphetNetForConditionalGeneration(ProphetNetForConditionalGeneration):
+    """
+    This class overrides :class:`~transformers.ProphetNetForConditionalGeneration`. Please check the
+    superclass for the appropriate documentation alongside usage examples.
+    """
+
+    config_class = XLMProphetNetConfig

src/transformers/tokenization_auto.py

@@ -43,6 +43,7 @@ from .configuration_auto import (
     MobileBertConfig,
     OpenAIGPTConfig,
     PegasusConfig,
+    ProphetNetConfig,
     RagConfig,
     ReformerConfig,
     RetriBertConfig,
@@ -51,6 +52,7 @@ from .configuration_auto import (
     T5Config,
     TransfoXLConfig,
     XLMConfig,
+    XLMProphetNetConfig,
     XLMRobertaConfig,
     XLNetConfig,
     replace_list_option_in_docstrings,
@@ -82,6 +84,7 @@ from .tokenization_mobilebert import MobileBertTokenizer, MobileBertTokenizerFas
 from .tokenization_openai import OpenAIGPTTokenizer, OpenAIGPTTokenizerFast
 from .tokenization_pegasus import PegasusTokenizer, PegasusTokenizerFast
 from .tokenization_phobert import PhobertTokenizer
+from .tokenization_prophetnet import ProphetNetTokenizer
 from .tokenization_rag import RagTokenizer
 from .tokenization_reformer import ReformerTokenizer, ReformerTokenizerFast
 from .tokenization_retribert import RetriBertTokenizer, RetriBertTokenizerFast
@@ -90,6 +93,7 @@ from .tokenization_squeezebert import SqueezeBertTokenizer, SqueezeBertTokenizer
 from .tokenization_t5 import T5Tokenizer, T5TokenizerFast
 from .tokenization_transfo_xl import TransfoXLTokenizer
 from .tokenization_xlm import XLMTokenizer
+from .tokenization_xlm_prophetnet import XLMProphetNetTokenizer
 from .tokenization_xlm_roberta import XLMRobertaTokenizer, XLMRobertaTokenizerFast
 from .tokenization_xlnet import XLNetTokenizer, XLNetTokenizerFast
 from .utils import logging
@@ -136,6 +140,8 @@ TOKENIZER_MAPPING = OrderedDict(
         (BertGenerationConfig, (BertGenerationTokenizer, None)),
         (DebertaConfig, (DebertaTokenizer, None)),
         (RagConfig, (RagTokenizer, None)),
+        (XLMProphetNetConfig, (XLMProphetNetTokenizer, None)),
+        (ProphetNetConfig, (ProphetNetTokenizer, None)),
     ]
 )
 

src/transformers/tokenization_bert.py

@@ -398,7 +398,6 @@ class BasicTokenizer(object):
         """
         # union() returns a new set by concatenating the two sets.
         never_split = self.never_split.union(set(never_split)) if never_split else self.never_split
-        text = self._clean_text(text)
 
         # This was added on November 1st, 2018 for the multilingual and Chinese
         # models. This is also applied to the English models now, but it doesn't

src/transformers/tokenization_prophetnet.py

@@ -0,0 +1,289 @@
+# coding=utf-8
+# Copyright 2020 The Microsoft Authors and The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import collections
+import logging
+import os
+from typing import List, Optional
+
+from .tokenization_bert import BasicTokenizer, WordpieceTokenizer
+from .tokenization_utils import PreTrainedTokenizer
+
+
+logger = logging.getLogger(__name__)
+
+VOCAB_FILES_NAMES = {"vocab_file": "prophetnet.tokenizer"}
+
+PRETRAINED_VOCAB_FILES_MAP = {
+    "vocab_file": {
+        "microsoft/prophetnet-large-uncased": "https://s3.amazonaws.com/models.huggingface.co/bert/microsoft/prophetnet-large-uncased/prophetnet.tokenizer",
+    }
+}
+
+PRETRAINED_INIT_CONFIGURATION = {
+    "microsoft/prophetnet-large-uncased": {"do_lower_case": True, "xprophetnet_tokenizer": False},
+}
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    "microsoft/prophetnet-large-uncased": 512,
+}
+
+
+def load_vocab(vocab_file):
+    """Loads a vocabulary file into a dictionary."""
+    vocab = collections.OrderedDict()
+    with open(vocab_file, "r", encoding="utf-8") as reader:
+        tokens = reader.readlines()
+    for index, token in enumerate(tokens):
+        token = token.rstrip("\n")
+        vocab[token] = index
+    return vocab
+
+
+class ProphetNetTokenizer(PreTrainedTokenizer):
+    r"""
+    Construct a ProphetNetTokenizer. Based on WordPiece.
+
+    This tokenizer inherits from :class:`~transformers.PreTrainedTokenizer` which contains most of the main methods.
+    Users should refer to this superclass for more information regarding those methods.
+
+    Args:
+        vocab_file (:obj:`str`):
+            File containing the vocabulary.
+        do_lower_case (:obj:`bool`, `optional`, defaults to :obj:`True`):
+            Whether or not to lowercase the input when tokenizing.
+        do_basic_tokenize (:obj:`bool`, `optional`, defaults to :obj:`True`):
+            Whether or not to do basic tokenization before WordPiece.
+        never_split (:obj:`Iterable`, `optional`):
+            Collection of tokens which will never be split during tokenization. Only has an effect when
+            :obj:`do_basic_tokenize=True`
+        unk_token (:obj:`str`, `optional`, defaults to :obj:`"[UNK]"`):
+            The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this
+            token instead.
+        sep_token (:obj:`str`, `optional`, defaults to :obj:`"[SEP]"`):
+            The separator token, which is used when building a sequence from multiple sequences, e.g. two sequences
+            for sequence classification or for a text and a question for question answering.
+            It is also used as the last token of a sequence built with special tokens.
+        pad_token (:obj:`str`, `optional`, defaults to :obj:`"[PAD]"`):
+            The token used for padding, for example when batching sequences of different lengths.
+        cls_token (:obj:`str`, `optional`, defaults to :obj:`"[CLS]"`):
+            The classifier token which is used when doing sequence classification (classification of the whole
+            sequence instead of per-token classification). It is the first token of the sequence when built with
+            special tokens.
+        mask_token (:obj:`str`, `optional`, defaults to :obj:`"[MASK]"`):
+            The token used for masking values. This is the token used when training this model with masked language
+            modeling. This is the token which the model will try to predict.
+        tokenize_chinese_chars (:obj:`bool`, `optional`, defaults to :obj:`True`):
+            Whether or not to tokenize Chinese characters.
+
+            This should likely be deactivated for Japanese (see this `issue
+            <https://github.com/huggingface/transformers/issues/328>`__).
+        strip_accents: (:obj:`bool`, `optional`):
+            Whether or not to strip all accents. If this option is not specified, then it will be determined by the
+            value for :obj:`lowercase` (as in the original BERT).
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    pretrained_init_configuration = PRETRAINED_INIT_CONFIGURATION
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+
+    def __init__(
+        self,
+        vocab_file,
+        do_lower_case=True,
+        do_basic_tokenize=True,
+        never_split=None,
+        unk_token="[UNK]",
+        sep_token="[SEP]",
+        x_sep_token="[X_SEP]",
+        pad_token="[PAD]",
+        mask_token="[MASK]",
+        tokenize_chinese_chars=True,
+        strip_accents=None,
+        **kwargs
+    ):
+        super().__init__(
+            unk_token=unk_token,
+            sep_token=sep_token,
+            pad_token=pad_token,
+            mask_token=mask_token,
+            x_sep_token=x_sep_token,
+            **kwargs,
+        )
+        self.unique_no_split_tokens.append(x_sep_token)
+
+        if not os.path.isfile(vocab_file):
+            raise ValueError(
+                "Can't find a vocabulary file at path '{}'. To load the vocabulary from a Google pretrained "
+                "model use `tokenizer = ProphetNetTokenizer.from_pretrained(PRETRAINED_MODEL_NAME)`".format(vocab_file)
+            )
+        self.vocab = load_vocab(vocab_file)
+        self.ids_to_tokens = collections.OrderedDict([(ids, tok) for tok, ids in self.vocab.items()])
+        self.do_basic_tokenize = do_basic_tokenize
+        if do_basic_tokenize:
+            self.basic_tokenizer = BasicTokenizer(
+                do_lower_case=do_lower_case,
+                never_split=never_split,
+                tokenize_chinese_chars=tokenize_chinese_chars,
+                strip_accents=strip_accents,
+            )
+        self.wordpiece_tokenizer = WordpieceTokenizer(vocab=self.vocab, unk_token=self.unk_token)
+
+    @property
+    def vocab_size(self):
+        return len(self.vocab)
+
+    def get_vocab(self):
+        return dict(self.vocab, **self.added_tokens_encoder)
+
+    def _tokenize(self, text):
+        split_tokens = []
+        if self.do_basic_tokenize:
+            for token in self.basic_tokenizer.tokenize(text, never_split=self.all_special_tokens):
+
+                # If the token is part of the never_split set
+                if token in self.basic_tokenizer.never_split:
+                    split_tokens.append(token)
+                else:
+                    split_tokens += self.wordpiece_tokenizer.tokenize(token)
+        else:
+            split_tokens = self.wordpiece_tokenizer.tokenize(text)
+        return split_tokens
+
+    def _convert_token_to_id(self, token):
+        """ Converts a token (str) in an id using the vocab. """
+        return self.vocab.get(token, self.vocab.get(self.unk_token))
+
+    def _convert_id_to_token(self, index):
+        """Converts an index (integer) in a token (str) using the vocab."""
+        return self.ids_to_tokens.get(index, self.unk_token)
+
+    def convert_tokens_to_string(self, tokens):
+        """ Converts a sequence of tokens (string) in a single string. """
+        out_string = " ".join(tokens).replace(" ##", "").strip()
+        return out_string
+
+    def get_special_tokens_mask(
+        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None, already_has_special_tokens: bool = False
+    ) -> List[int]:
+        """
+        Retrieve sequence ids from a token list that has no special tokens added. This method is called when adding
+        special tokens using the tokenizer ``prepare_for_model`` method.
+
+        Args:
+            token_ids_0 (:obj:`List[int]`):
+                List of IDs.
+            token_ids_1 (:obj:`List[int]`, `optional`):
+                Optional second list of IDs for sequence pairs.
+            already_has_special_tokens (:obj:`bool`, `optional`, defaults to :obj:`False`):
+                Whether or not the token list is already formatted with special tokens for the model.
+
+        Returns:
+            :obj:`List[int]`: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
+        """
+        if already_has_special_tokens:
+            if token_ids_1 is not None:
+                raise ValueError(
+                    "You should not supply a second sequence if the provided sequence of "
+                    "ids is already formated with special tokens for the model."
+                )
+            return list(map(lambda x: 1 if x in [self.sep_token_id, self.cls_token_id] else 0, token_ids_0))
+
+        if token_ids_1 is None:
+            return ([0] * len(token_ids_0)) + [1]
+        return ([0] * len(token_ids_0)) + [1] + ([0] * len(token_ids_1)) + [1]
+
+    def create_token_type_ids_from_sequences(
+        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
+    ) -> List[int]:
+        """
+        Create a mask from the two sequences passed to be used in a sequence-pair classification task.
+        A ProphetNet sequence pair mask has the following format:
+
+        ::
+
+            0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1
+            | first sequence    | second sequence |
+
+        If :obj:`token_ids_1` is :obj:`None`, this method only returns the first portion of the mask (0s).
+
+        Args:
+            token_ids_0 (:obj:`List[int]`):
+                List of IDs.
+            token_ids_1 (:obj:`List[int]`, `optional`):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            :obj:`List[int]`: List of `token type IDs <../glossary.html#token-type-ids>`_ according to the given
+            sequence(s).
+        """
+        sep = [self.sep_token_id]
+        if token_ids_1 is None:
+            return len(token_ids_0 + sep) * [0]
+        return len(token_ids_0 + sep) * [0] + len(token_ids_1 + sep) * [1]
+
+    def save_vocabulary(self, vocab_path):
+        """
+        Save the vocabulary (copy original file) and special tokens file to a directory.
+
+        Args:
+            vocab_path (:obj:`str`):
+                The directory in which to save the vocabulary.
+
+        Returns:
+            :obj:`Tuple(str)`: Paths to the files saved.
+        """
+        index = 0
+        if os.path.isdir(vocab_path):
+            vocab_file = os.path.join(vocab_path, VOCAB_FILES_NAMES["vocab_file"])
+        else:
+            vocab_file = vocab_path
+        with open(vocab_file, "w", encoding="utf-8") as writer:
+            for token, token_index in sorted(self.vocab.items(), key=lambda kv: kv[1]):
+                if index != token_index:
+                    logger.warning(
+                        "Saving vocabulary to {}: vocabulary indices are not consecutive."
+                        " Please check that the vocabulary is not corrupted!".format(vocab_file)
+                    )
+                    index = token_index
+                writer.write(token + "\n")
+                index += 1
+        return (vocab_file,)
+
+    def build_inputs_with_special_tokens(
+        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
+    ) -> List[int]:
+        """
+        Build model inputs from a sequence or a pair of sequence for sequence classification tasks
+        by concatenating and adding special tokens.
+        A BERT sequence has the following format:
+
+        - single sequence: ``[CLS] X [SEP]``
+        - pair of sequences: ``[CLS] A [SEP] B [SEP]``
+
+        Args:
+            token_ids_0 (:obj:`List[int]`):
+                List of IDs to which the special tokens will be added.
+            token_ids_1 (:obj:`List[int]`, `optional`):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            :obj:`List[int]`: List of `input IDs <../glossary.html#input-ids>`__ with the appropriate special tokens.
+        """
+        if token_ids_1 is None:
+            return token_ids_0 + [self.sep_token_id]
+        sep = [self.sep_token_id]
+        return token_ids_0 + sep + token_ids_1 + sep

src/transformers/tokenization_xlm_prophetnet.py

@@ -0,0 +1,314 @@
+# coding=utf-8
+# Copyright 2020 The Microsoft Authors and The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import collections
+import logging
+import os
+from shutil import copyfile
+from typing import List, Optional
+
+from .tokenization_utils import PreTrainedTokenizer
+
+
+logger = logging.getLogger(__name__)
+
+SPIECE_UNDERLINE = "▁"
+
+VOCAB_FILES_NAMES = {"vocab_file": "prophetnet.tokenizer"}
+
+PRETRAINED_VOCAB_FILES_MAP = {
+    "vocab_file": {
+        "microsoft/xprophetnet-large-wiki100-cased": "https://cdn.huggingface.co/microsoft/xprophetnet-large-wiki100-cased/prophetnet.tokenizer",
+    }
+}
+
+PRETRAINED_INIT_CONFIGURATION = {
+    "microsoft/xprophetnet-large-wiki100-cased": {"do_lower_case": False, "xprophetnet_tokenizer": True},
+}
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    "microsoft/xprophetnet-large-wiki100-cased": 512,
+}
+
+
+def load_vocab(vocab_file):
+    """Loads a vocabulary file into a dictionary."""
+    vocab = collections.OrderedDict()
+    with open(vocab_file, "r", encoding="utf-8") as reader:
+        tokens = reader.readlines()
+    for index, token in enumerate(tokens):
+        token = token.rstrip("\n")
+        vocab[token] = index
+    return vocab
+
+
+class XLMProphetNetTokenizer(PreTrainedTokenizer):
+    """
+    Adapted from :class:`~transfomers.RobertaTokenizer` and class:`~transfomers.XLNetTokenizer`. Based on
+    `SentencePiece <https://github.com/google/sentencepiece>`__.
+
+    This tokenizer inherits from :class:`~transformers.PreTrainedTokenizer` which contains most of the main methods.
+    Users should refer to this superclass for more information regarding those methods.
+
+    Args:
+        vocab_file (:obj:`str`):
+            Path to the vocabulary file.
+        bos_token (:obj:`str`, `optional`, defaults to :obj:`"<s>"`):
+            The beginning of sequence token that was used during pretraining. Can be used a sequence classifier token.
+
+            .. note::
+
+                When building a sequence using special tokens, this is not the token that is used for the beginning
+                of sequence. The token used is the :obj:`cls_token`.
+        eos_token (:obj:`str`, `optional`, defaults to :obj:`"</s>"`):
+            The end of sequence token.
+
+            .. note::
+
+                When building a sequence using special tokens, this is not the token that is used for the end
+                of sequence. The token used is the :obj:`sep_token`.
+        sep_token (:obj:`str`, `optional`, defaults to :obj:`"</s>"`):
+            The separator token, which is used when building a sequence from multiple sequences, e.g. two sequences
+            for sequence classification or for a text and a question for question answering.
+            It is also used as the last token of a sequence built with special tokens.
+        cls_token (:obj:`str`, `optional`, defaults to :obj:`"<s>"`):
+            The classifier token which is used when doing sequence classification (classification of the whole
+            sequence instead of per-token classification). It is the first token of the sequence when built with
+            special tokens.
+        unk_token (:obj:`str`, `optional`, defaults to :obj:`"<unk>"`):
+            The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this
+            token instead.
+        pad_token (:obj:`str`, `optional`, defaults to :obj:`"<pad>"`):
+            The token used for padding, for example when batching sequences of different lengths.
+        mask_token (:obj:`str`, `optional`, defaults to :obj:`"<mask>"`):
+            The token used for masking values. This is the token used when training this model with masked language
+            modeling. This is the token which the model will try to predict.
+        additional_special_tokens (:obj:`List[str]`, `optional`, defaults to :obj:`["<s>NOTUSED", "</s>NOTUSED"]`):
+            Additional special tokens used by the tokenizer.
+
+    Attributes:
+        sp_model (:obj:`SentencePieceProcessor`):
+            The `SentencePiece` processor that is used for every conversion (string, tokens and IDs).
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+    model_input_names = ["attention_mask"]
+
+    def __init__(
+        self,
+        vocab_file,
+        bos_token="[SEP]",
+        eos_token="[SEP]",
+        sep_token="[SEP]",
+        unk_token="[UNK]",
+        pad_token="[PAD]",
+        mask_token="[MASK]",
+        **kwargs
+    ):
+        super().__init__(
+            bos_token=bos_token,
+            eos_token=eos_token,
+            unk_token=unk_token,
+            sep_token=sep_token,
+            pad_token=pad_token,
+            mask_token=mask_token,
+            **kwargs,
+        )
+
+        try:
+            import sentencepiece as spm
+        except ImportError:
+            logger.warning(
+                "You need to install SentencePiece to use XLMRobertaTokenizer: https://github.com/google/sentencepiece"
+                "pip install sentencepiece"
+            )
+            raise
+
+        self.sp_model = spm.SentencePieceProcessor()
+        self.sp_model.Load(str(vocab_file))
+        self.vocab_file = vocab_file
+
+        # Original fairseq vocab and spm vocab must be "aligned":
+        # Vocab    |    0    |    1    |   2    |    3    |  4  |  5  |  6  |   7   |   8   |  9
+        # -------- | ------- | ------- | ------ | ------- | --- | --- | --- | ----- | ----- | ----
+        # fairseq  | '<s>'   | '<pad>' | '</s>' | '<unk>' | ',' | '.' | '▁' | 's'   | '▁de' | '-'
+        # spm      | '<unk>' | '<s>'   | '</s>' | ','     | '.' | '▁' | 's' | '▁de' | '-'   | '▁a'
+
+        # put special tokens and [unused] tokens into the vocab
+        self.fairseq_tokens_to_ids = {"[PAD]": 0, "[CLS]": 1, "[SEP]": 2, "[UNK]": 3, "[MASK]": 4}
+
+        for i in range(10):
+            tok = "[unused{}]".format(i)
+            self.fairseq_tokens_to_ids[tok] = 5 + i
+
+        # The first "real" token "," has position 15 in the embedding vocab and position 3 in the spm vocab
+        self.fairseq_offset = 12
+        self.fairseq_ids_to_tokens = {v: k for k, v in self.fairseq_tokens_to_ids.items()}
+        for k in self.fairseq_tokens_to_ids.keys():
+            self.unique_no_split_tokens.append(k)
+
+    def __getstate__(self):
+        state = self.__dict__.copy()
+        state["sp_model"] = None
+        return state
+
+    def __setstate__(self, d):
+        self.__dict__ = d
+        try:
+            import sentencepiece as spm
+        except ImportError:
+            logger.warning(
+                "You need to install SentencePiece to use XLMRobertaTokenizer: https://github.com/google/sentencepiece"
+                "pip install sentencepiece"
+            )
+            raise
+        self.sp_model = spm.SentencePieceProcessor()
+        self.sp_model.Load(self.vocab_file)
+
+    def get_special_tokens_mask(
+        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None, already_has_special_tokens: bool = False
+    ) -> List[int]:
+        """
+        Retrieve sequence ids from a token list that has no special tokens added. This method is called when adding
+        special tokens using the tokenizer ``prepare_for_model`` method.
+
+        Args:
+            token_ids_0 (:obj:`List[int]`):
+                List of IDs.
+            token_ids_1 (:obj:`List[int]`, `optional`):
+                Optional second list of IDs for sequence pairs.
+            already_has_special_tokens (:obj:`bool`, `optional`, defaults to :obj:`False`):
+                Whether or not the token list is already formatted with special tokens for the model.
+
+        Returns:
+            :obj:`List[int]`: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
+        """
+
+        if already_has_special_tokens:
+            if token_ids_1 is not None:
+                raise ValueError(
+                    "You should not supply a second sequence if the provided sequence of "
+                    "ids is already formated with special tokens for the model."
+                )
+            return list(map(lambda x: 1 if x in [self.sep_token_id, self.cls_token_id] else 0, token_ids_0))
+
+        if token_ids_1 is None:
+            return ([0] * len(token_ids_0)) + [1]
+        return ([0] * len(token_ids_0)) + [1] + ([0] * len(token_ids_1)) + [1]
+
+    def create_token_type_ids_from_sequences(
+        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
+    ) -> List[int]:
+        """
+        Create a mask from the two sequences passed to be used in a sequence-pair classification task.
+        XLM-RoBERTa does not make use of token type ids, therefore a list of zeros is returned.
+
+        Args:
+            token_ids_0 (:obj:`List[int]`):
+                List of IDs.
+            token_ids_1 (:obj:`List[int]`, `optional`):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            :obj:`List[int]`: List of zeros.
+
+        """
+
+        sep = [self.sep_token_id]
+
+        if token_ids_1 is None:
+            return len(token_ids_0 + sep) * [0]
+        return len(token_ids_0 + sep + sep + token_ids_1 + sep) * [0]
+
+    @property
+    def vocab_size(self):
+        return len(self.sp_model) + self.fairseq_offset
+
+    def get_vocab(self):
+        vocab = {self.convert_ids_to_tokens(i): i for i in range(self.vocab_size)}
+        vocab.update(self.added_tokens_encoder)
+        return vocab
+
+    def _tokenize(self, text):
+        return self.sp_model.EncodeAsPieces(text)
+
+    def _convert_token_to_id(self, token):
+        """ Converts a token (str) in an id using the vocab. """
+        if token in self.fairseq_tokens_to_ids:
+            return self.fairseq_tokens_to_ids[token]
+        spm_id = self.sp_model.PieceToId(token)
+
+        # Need to return unknown token if the SP model returned 0
+        return spm_id + self.fairseq_offset if spm_id else self.unk_token_id
+
+    def _convert_id_to_token(self, index):
+        """Converts an index (integer) in a token (str) using the vocab."""
+        if index in self.fairseq_ids_to_tokens:
+            return self.fairseq_ids_to_tokens[index]
+        return self.sp_model.IdToPiece(index - self.fairseq_offset)
+
+    def convert_tokens_to_string(self, tokens):
+        """Converts a sequence of tokens (strings for sub-words) in a single string."""
+        out_string = "".join(tokens).replace(SPIECE_UNDERLINE, " ").strip()
+        return out_string
+
+    def save_vocabulary(self, save_directory):
+        """
+        Save the sentencepiece vocabulary (copy original file) and special tokens file to a directory.
+
+        Args:
+            save_directory (:obj:`str`):
+                The directory in which to save the vocabulary.
+
+        Returns:
+            :obj:`Tuple(str)`: Paths to the files saved.
+        """
+        if not os.path.isdir(save_directory):
+            logger.error("Vocabulary path ({}) should be a directory".format(save_directory))
+            return
+        out_vocab_file = os.path.join(save_directory, VOCAB_FILES_NAMES["vocab_file"])
+
+        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file):
+            copyfile(self.vocab_file, out_vocab_file)
+
+        return (out_vocab_file,)
+
+    def build_inputs_with_special_tokens(
+        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
+    ) -> List[int]:
+        """
+        Build model inputs from a sequence or a pair of sequence for sequence classification tasks
+        by concatenating and adding special tokens.
+        A ProphetNet sequence has the following format:
+
+        - single sequence: ``X [SEP]``
+        - pair of sequences: ``A [SEP] B [SEP]``
+
+        Args:
+            token_ids_0 (:obj:`List[int]`):
+                List of IDs to which the special tokens will be added
+            token_ids_1 (:obj:`List[int]`, `optional`, defaults to :obj:`None`):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            :obj:`List[int]`: list of `input IDs <../glossary.html#input-ids>`__ with the appropriate special tokens.
+        """
+
+        if token_ids_1 is None:
+            return token_ids_0 + [self.sep_token_id]
+        sep = [self.sep_token_id]
+        return token_ids_0 + sep + token_ids_1 + sep

src/transformers/tokenization_xlm_roberta.py

@@ -21,12 +21,13 @@ from typing import List, Optional
 
 from .tokenization_utils import PreTrainedTokenizer
 from .tokenization_utils_fast import PreTrainedTokenizerFast
-from .tokenization_xlnet import SPIECE_UNDERLINE
 from .utils import logging
 
 
 logger = logging.get_logger(__name__)
 
+SPIECE_UNDERLINE = "▁"
+
 VOCAB_FILES_NAMES = {"vocab_file": "sentencepiece.bpe.model"}
 
 PRETRAINED_VOCAB_FILES_MAP = {

src/transformers/utils/dummy_pt_objects.py

@@ -1296,6 +1296,36 @@ class PegasusForConditionalGeneration:
         requires_pytorch(self)
 
 
+PROPHETNET_PRETRAINED_MODEL_ARCHIVE_LIST = None
+
+
+class ProphetNetForConditionalGeneration:
+    def __init__(self, *args, **kwargs):
+        requires_pytorch(self)
+
+    @classmethod
+    def from_pretrained(self, *args, **kwargs):
+        requires_pytorch(self)
+
+
+class ProphetNetModel:
+    def __init__(self, *args, **kwargs):
+        requires_pytorch(self)
+
+    @classmethod
+    def from_pretrained(self, *args, **kwargs):
+        requires_pytorch(self)
+
+
+class ProphetNetPreTrainedModel:
+    def __init__(self, *args, **kwargs):
+        requires_pytorch(self)
+
+    @classmethod
+    def from_pretrained(self, *args, **kwargs):
+        requires_pytorch(self)
+
+
 class RagModel:
     def __init__(self, *args, **kwargs):
         requires_pytorch(self)
@@ -1697,6 +1727,27 @@ class XLMWithLMHeadModel:
         requires_pytorch(self)
 
 
+XLM_PROPHETNET_PRETRAINED_MODEL_ARCHIVE_LIST = None
+
+
+class XLMProphetNetForConditionalGeneration:
+    def __init__(self, *args, **kwargs):
+        requires_pytorch(self)
+
+    @classmethod
+    def from_pretrained(self, *args, **kwargs):
+        requires_pytorch(self)
+
+
+class XLMProphetNetModel:
+    def __init__(self, *args, **kwargs):
+        requires_pytorch(self)
+
+    @classmethod
+    def from_pretrained(self, *args, **kwargs):
+        requires_pytorch(self)
+
+
 XLM_ROBERTA_PRETRAINED_MODEL_ARCHIVE_LIST = None
 
 

tests/test_modeling_common.py

@@ -112,6 +112,7 @@ class ModelTesterMixin:
             model.eval()
             with torch.no_grad():
                 outputs = model(**self._prepare_for_class(inputs_dict, model_class))
+
             out_2 = outputs[0].cpu().numpy()
             out_2[np.isnan(out_2)] = 0
 
@@ -152,6 +153,7 @@ class ModelTesterMixin:
             with torch.no_grad():
                 first = model(**self._prepare_for_class(inputs_dict, model_class))[0]
                 second = model(**self._prepare_for_class(inputs_dict, model_class))[0]
+
             out_1 = first.cpu().numpy()
             out_2 = second.cpu().numpy()
             out_1 = out_1[~np.isnan(out_1)]
@@ -186,7 +188,7 @@ class ModelTesterMixin:
         seq_len = getattr(self.model_tester, "seq_length", None)
         decoder_seq_length = getattr(self.model_tester, "decoder_seq_length", seq_len)
         encoder_seq_length = getattr(self.model_tester, "encoder_seq_length", seq_len)
-        decoder_key_length = getattr(self.model_tester, "key_length", decoder_seq_length)
+        decoder_key_length = getattr(self.model_tester, "decoder_key_length", decoder_seq_length)
         encoder_key_length = getattr(self.model_tester, "key_length", encoder_seq_length)
         chunk_length = getattr(self.model_tester, "chunk_length", None)
         if chunk_length is not None and hasattr(self.model_tester, "num_hashes"):
@@ -238,6 +240,7 @@ class ModelTesterMixin:
                 if model_class in MODEL_FOR_QUESTION_ANSWERING_MAPPING.values():
                     correct_outlen += 1  # start_logits and end_logits instead of only 1 output
                     decoder_attention_idx += 1
+
                 self.assertEqual(out_len, correct_outlen)
 
                 decoder_attentions = outputs[decoder_attention_idx]
@@ -822,6 +825,7 @@ class ModelTesterMixin:
             model.eval()
 
             inputs = copy.deepcopy(self._prepare_for_class(inputs_dict, model_class))
+
             if not self.is_encoder_decoder:
                 input_ids = inputs["input_ids"]
                 del inputs["input_ids"]
@@ -839,7 +843,7 @@ class ModelTesterMixin:
                 inputs["decoder_inputs_embeds"] = wte(decoder_input_ids)
 
             with torch.no_grad():
-                model(**inputs)
+                model(**inputs)[0]
 
     def test_lm_head_model_random_no_beam_search_generate(self):
         config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()

tests/test_modeling_encoder_decoder.py

@@ -24,6 +24,7 @@ from .test_modeling_bert import BertModelTester
 from .test_modeling_bert_generation import BertGenerationEncoderTester
 from .test_modeling_common import ids_tensor
 from .test_modeling_gpt2 import GPT2ModelTester
+from .test_modeling_prophetnet import ProphetNetModelTester
 from .test_modeling_roberta import RobertaModelTester
 
 
@@ -41,6 +42,7 @@ if is_torch_available():
         EncoderDecoderConfig,
         EncoderDecoderModel,
         GPT2LMHeadModel,
+        ProphetNetForConditionalGeneration,
         RobertaForCausalLM,
         RobertaModel,
     )
@@ -663,3 +665,55 @@ class GPT2EncoderDecoderModelTest(EncoderDecoderMixin, unittest.TestCase):
 
     def test_encoder_decoder_model_shared_weights(self):
         pass
+
+
+class ProphetNetModelEncoderDecoderModelTest(EncoderDecoderMixin, unittest.TestCase):
+    def get_encoder_decoder_model(self, config, decoder_config):
+        encoder_model = BertModel(config)
+        decoder_model = ProphetNetForConditionalGeneration(decoder_config).get_decoder()
+        return encoder_model, decoder_model
+
+    def prepare_config_and_inputs(self):
+        model_tester_encoder = BertModelTester(self, batch_size=13)
+        model_tester_decoder = ProphetNetModelTester(self, batch_size=13)
+        encoder_config_and_inputs = model_tester_encoder.prepare_config_and_inputs()
+        decoder_config_and_inputs = model_tester_decoder.prepare_config_and_inputs()
+        (
+            config,
+            input_ids,
+            token_type_ids,
+            input_mask,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+        ) = encoder_config_and_inputs
+        (
+            decoder_config,
+            input_ids,
+            decoder_input_ids,
+            attention_mask,
+            decoder_attention_mask,
+            lm_labels,
+        ) = decoder_config_and_inputs
+
+        # make sure that cross attention layers are added
+        decoder_config.add_cross_attention = True
+        #  disable cache for now
+        decoder_config.use_cache = False
+        return {
+            "config": config,
+            "input_ids": input_ids,
+            "attention_mask": input_mask,
+            "decoder_config": decoder_config,
+            "decoder_input_ids": decoder_input_ids,
+            "decoder_attention_mask": decoder_attention_mask,
+            "labels": lm_labels,
+        }
+
+    def get_pretrained_model(self):
+        return EncoderDecoderModel.from_encoder_decoder_pretrained(
+            "bert-base-cased", "patrickvonplaten/prophetnet-large-uncased-decoder"
+        )
+
+    def test_encoder_decoder_model_shared_weights(self):
+        pass

tests/test_modeling_prophetnet.py

@@ -0,0 +1,648 @@
+# coding=utf-8
+# Copyright 2018 The Google AI Language Team Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import copy
+import tempfile
+import unittest
+
+from transformers import is_torch_available
+from transformers.testing_utils import require_torch, slow, torch_device
+
+from .test_configuration_common import ConfigTester
+from .test_modeling_common import ModelTesterMixin, ids_tensor
+
+
+if is_torch_available():
+    import torch
+
+    from transformers import ProphetNetConfig, ProphetNetForConditionalGeneration, ProphetNetModel, ProphetNetTokenizer
+
+
+class ProphetNetModelTester:
+    def __init__(
+        self,
+        parent,
+        vocab_size=99,
+        batch_size=13,
+        hidden_size=16,
+        encoder_seq_length=7,
+        decoder_seq_length=9,
+        # For common tests
+        is_training=True,
+        use_attention_mask=True,
+        use_labels=True,
+        decoder_start_token_id=0,
+        encoder_ffn_dim=32,
+        num_encoder_layers=4,
+        num_encoder_attention_heads=4,
+        decoder_ffn_dim=32,
+        num_decoder_layers=4,
+        num_decoder_attention_heads=4,
+        max_position_embeddings=30,
+        is_encoder_decoder=True,
+        pad_token_id=0,
+        bos_token_id=1,
+        eos_token_id=2,
+        ngram=1,
+        num_buckets=32,
+        relative_max_distance=128,
+        disable_ngram_loss=False,
+        scope=None,
+    ):
+
+        self.parent = parent
+        self.batch_size = batch_size
+        self.encoder_seq_length = encoder_seq_length
+        self.decoder_seq_length = decoder_seq_length
+        # For common tests
+        self.seq_length = self.decoder_seq_length
+        self.is_training = is_training
+        self.use_attention_mask = use_attention_mask
+        self.use_labels = use_labels
+
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_decoder_layers
+        self.num_encoder_layers = num_encoder_layers
+        self.num_decoder_layers = num_decoder_layers
+        self.decoder_ffn_dim = decoder_ffn_dim
+        self.encoder_ffn_dim = encoder_ffn_dim
+        self.num_attention_heads = num_decoder_attention_heads
+        self.num_encoder_attention_heads = num_encoder_attention_heads
+        self.num_decoder_attention_heads = num_decoder_attention_heads
+        self.eos_token_id = eos_token_id
+        self.bos_token_id = bos_token_id
+        self.pad_token_id = pad_token_id
+        self.decoder_start_token_id = decoder_start_token_id
+        self.ngram = ngram
+        self.num_buckets = num_buckets
+        self.relative_max_distance = relative_max_distance
+        self.disable_ngram_loss = disable_ngram_loss
+        self.max_position_embeddings = max_position_embeddings
+        self.is_encoder_decoder = is_encoder_decoder
+
+        self.scope = None
+        self.decoder_key_length = 2 * decoder_seq_length
+
+    def prepare_config_and_inputs(self):
+        input_ids = ids_tensor([self.batch_size, self.encoder_seq_length], self.vocab_size)
+        decoder_input_ids = ids_tensor([self.batch_size, self.decoder_seq_length], self.vocab_size)
+
+        attention_mask = None
+        decoder_attention_mask = None
+        if self.use_attention_mask:
+            attention_mask = ids_tensor([self.batch_size, self.encoder_seq_length], vocab_size=2)
+            decoder_attention_mask = ids_tensor([self.batch_size, self.decoder_seq_length], vocab_size=2)
+
+        lm_labels = None
+        if self.use_labels:
+            lm_labels = ids_tensor([self.batch_size, self.decoder_seq_length], self.vocab_size)
+
+        config = ProphetNetConfig(
+            vocab_size=self.vocab_size,
+            hidden_size=self.hidden_size,
+            num_encoder_layers=self.num_encoder_layers,
+            num_decoder_layers=self.num_decoder_layers,
+            decoder_ffn_dim=self.decoder_ffn_dim,
+            encoder_ffn_dim=self.encoder_ffn_dim,
+            num_encoder_attention_heads=self.num_encoder_attention_heads,
+            num_decoder_attention_heads=self.num_decoder_attention_heads,
+            eos_token_id=self.eos_token_id,
+            bos_token_id=self.bos_token_id,
+            pad_token_id=self.pad_token_id,
+            decoder_start_token_id=self.decoder_start_token_id,
+            ngram=self.ngram,
+            num_buckets=self.num_buckets,
+            relative_max_distance=self.relative_max_distance,
+            disable_ngram_loss=self.disable_ngram_loss,
+            max_position_embeddings=self.max_position_embeddings,
+            is_encoder_decoder=self.is_encoder_decoder,
+            return_dict=True,
+        )
+
+        return (
+            config,
+            input_ids,
+            decoder_input_ids,
+            attention_mask,
+            decoder_attention_mask,
+            lm_labels,
+        )
+
+    def check_prepare_lm_labels_via_shift_left(
+        self,
+        config,
+        input_ids,
+        decoder_input_ids,
+        attention_mask,
+        decoder_attention_mask,
+        lm_labels,
+    ):
+        model = ProphetNetModel(config=config)
+        model.to(torch_device)
+        model.eval()
+
+        # make sure that lm_labels are correctly padded from the right
+        lm_labels.masked_fill_((lm_labels == self.decoder_start_token_id), self.eos_token_id)
+
+        # add casaul pad token mask
+        triangular_mask = torch.tril(lm_labels.new_ones(lm_labels.shape)).logical_not()
+        lm_labels.masked_fill_(triangular_mask, self.pad_token_id)
+        decoder_input_ids = model._shift_right(lm_labels)
+
+        for i, (decoder_input_ids_slice, lm_labels_slice) in enumerate(zip(decoder_input_ids, lm_labels)):
+            # first item
+            self.parent.assertEqual(decoder_input_ids_slice[0].item(), self.decoder_start_token_id)
+            if i < decoder_input_ids_slice.shape[-1]:
+                if i < decoder_input_ids.shape[-1] - 1:
+                    # items before diagonal
+                    self.parent.assertListEqual(
+                        decoder_input_ids_slice[1 : i + 1].tolist(), lm_labels_slice[:i].tolist()
+                    )
+                # pad items after diagonal
+                if i < decoder_input_ids.shape[-1] - 2:
+                    self.parent.assertListEqual(
+                        decoder_input_ids_slice[i + 2 :].tolist(), lm_labels_slice[i + 1 : -1].tolist()
+                    )
+            else:
+                # all items after square
+                self.parent.assertListEqual(decoder_input_ids_slice[1:].tolist(), lm_labels_slice[:-1].tolist())
+
+    def create_and_check_model(
+        self,
+        config,
+        input_ids,
+        decoder_input_ids,
+        attention_mask,
+        decoder_attention_mask,
+        lm_labels,
+    ):
+        model = ProphetNetModel(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_ids=input_ids,
+            decoder_input_ids=decoder_input_ids,
+            attention_mask=attention_mask,
+            decoder_attention_mask=decoder_attention_mask,
+        )
+        result = model(input_ids=input_ids, decoder_input_ids=decoder_input_ids)
+        decoder_output = result.last_hidden_state
+        decoder_past = result.past_key_values
+        encoder_output = result.encoder_last_hidden_state
+
+        self.parent.assertEqual(encoder_output.size(), (self.batch_size, self.encoder_seq_length, self.hidden_size))
+        self.parent.assertEqual(decoder_output.size(), (self.batch_size, self.decoder_seq_length, self.hidden_size))
+        # There should be `num_layers` key value embeddings stored in decoder_past
+        self.parent.assertEqual(len(decoder_past), config.num_layers)
+        # There should be a self attn key, a self attn value, a cross attn key and a cross attn value stored in each decoder_past tuple
+        self.parent.assertEqual(len(decoder_past[0]), 4)
+
+    def create_and_check_with_lm_head(
+        self,
+        config,
+        input_ids,
+        decoder_input_ids,
+        attention_mask,
+        decoder_attention_mask,
+        lm_labels,
+    ):
+        model = ProphetNetForConditionalGeneration(config=config).to(torch_device).eval()
+        outputs = model(
+            input_ids=input_ids,
+            decoder_input_ids=decoder_input_ids,
+            decoder_attention_mask=decoder_attention_mask,
+            labels=lm_labels,
+        )
+        self.parent.assertEqual(len(outputs), 4)
+        self.parent.assertEqual(outputs["logits"].size(), (self.batch_size, self.decoder_seq_length, self.vocab_size))
+        self.parent.assertEqual(outputs["loss"].size(), ())
+
+    def create_and_check_decoder_model_past(
+        self,
+        config,
+        input_ids,
+        decoder_input_ids,
+        attention_mask,
+        decoder_attention_mask,
+        lm_labels,
+    ):
+        model = ProphetNetModel(config=config).get_decoder().to(torch_device).eval()
+        # first forward pass
+        outputs = model(input_ids, use_cache=True)
+        outputs_use_cache_conf = model(input_ids)
+        outputs_no_past = model(input_ids, use_cache=False)
+
+        self.parent.assertTrue(len(outputs) == len(outputs_use_cache_conf))
+        self.parent.assertTrue(len(outputs) == len(outputs_no_past) + 1)
+
+        output, past_key_values = outputs.to_tuple()
+
+        # create hypothetical next token and extent to next_input_ids
+        next_tokens = ids_tensor((self.batch_size, 1), config.vocab_size)
+
+        # append to next input_ids and
+        next_input_ids = torch.cat([input_ids, next_tokens], dim=-1)
+
+        output_from_no_past = model(next_input_ids)["last_hidden_state"]
+        output_from_past = model(next_tokens, past_key_values=past_key_values)["last_hidden_state"]
+
+        # select random slice
+        random_slice_idx = ids_tensor((1,), output_from_past.shape[-1]).item()
+        output_from_no_past_slice = output_from_no_past[:, next_input_ids.shape[-1] - 1, random_slice_idx].detach()
+        output_from_past_slice = output_from_past[:, 0, random_slice_idx].detach()
+
+        # test that outputs are equal for slice
+        assert torch.allclose(output_from_past_slice, output_from_no_past_slice, atol=1e-3)
+
+    def create_and_check_decoder_model_attention_mask_past(
+        self,
+        config,
+        input_ids,
+        decoder_input_ids,
+        attention_mask,
+        decoder_attention_mask,
+        lm_labels,
+    ):
+        model = ProphetNetModel(config=config).get_decoder()
+        model.to(torch_device)
+        model.eval()
+
+        # create attention mask
+        attn_mask = torch.ones(input_ids.shape, dtype=torch.long, device=torch_device)
+
+        half_seq_length = input_ids.shape[-1] // 2
+        attn_mask[:, half_seq_length:] = 0
+
+        # first forward pass
+        output, past_key_values = model(input_ids, attention_mask=attn_mask, use_cache=True).to_tuple()
+
+        # create hypothetical next token and extent to next_input_ids
+        next_tokens = ids_tensor((self.batch_size, 1), config.vocab_size)
+
+        # change a random masked slice from input_ids
+        random_seq_idx_to_change = ids_tensor((1,), half_seq_length).item() + 1
+        random_other_next_tokens = ids_tensor((self.batch_size, 1), config.vocab_size).squeeze(-1)
+        input_ids[:, -random_seq_idx_to_change] = random_other_next_tokens
+
+        # append to next input_ids and attn_mask
+        next_input_ids = torch.cat([input_ids, next_tokens], dim=-1)
+        attn_mask = torch.cat(
+            [attn_mask, torch.ones((attn_mask.shape[0], 1), dtype=torch.long, device=torch_device)],
+            dim=1,
+        )
+
+        # get two different outputs
+        output_from_no_past = model(next_input_ids)["last_hidden_state"]
+        output_from_past = model(next_tokens, past_key_values=past_key_values)["last_hidden_state"]
+
+        # select random slice
+        random_slice_idx = ids_tensor((1,), output_from_past.shape[-1]).item()
+        output_from_no_past_slice = output_from_no_past[:, next_input_ids.shape[-1] - 1, random_slice_idx].detach()
+        output_from_past_slice = output_from_past[:, 0, random_slice_idx].detach()
+
+        # test that outputs are equal for slice
+        assert torch.allclose(output_from_past_slice, output_from_no_past_slice, atol=1e-2)
+
+    def create_and_check_generate_with_past_key_value_states(
+        self,
+        config,
+        input_ids,
+        decoder_input_ids,
+        attention_mask,
+        decoder_attention_mask,
+        lm_labels,
+    ):
+        model = ProphetNetForConditionalGeneration(config=config).to(torch_device).eval()
+        torch.manual_seed(0)
+        output_without_past_cache = model.generate(
+            input_ids[:1], num_beams=2, max_length=5, do_sample=True, use_cache=False
+        )
+        torch.manual_seed(0)
+        output_with_past_cache = model.generate(input_ids[:1], num_beams=2, max_length=5, do_sample=True)
+        self.parent.assertTrue(torch.all(output_with_past_cache == output_without_past_cache))
+
+    def create_and_check_model_fp16_forward(
+        self,
+        config,
+        input_ids,
+        decoder_input_ids,
+        attention_mask,
+        decoder_attention_mask,
+        lm_labels,
+    ):
+        model = ProphetNetModel(config=config).to(torch_device).half().eval()
+        output = model(input_ids, decoder_input_ids=input_ids, attention_mask=attention_mask)["last_hidden_state"]
+        self.parent.assertFalse(torch.isnan(output).any().item())
+
+    def create_and_check_encoder_decoder_shared_weights(
+        self,
+        config,
+        input_ids,
+        decoder_input_ids,
+        attention_mask,
+        decoder_attention_mask,
+        lm_labels,
+    ):
+        for model_class in [ProphetNetModel, ProphetNetForConditionalGeneration]:
+            torch.manual_seed(0)
+            model = model_class(config=config).to(torch_device).eval()
+            # load state dict copies weights but does not tie them
+
+            if model_class == ProphetNetForConditionalGeneration:
+                model.prophetnet.encoder.load_state_dict(model.prophetnet.decoder.state_dict(), strict=False)
+            else:
+                model.encoder.load_state_dict(model.decoder.state_dict(), strict=False)
+
+            torch.manual_seed(0)
+            tied_config = copy.deepcopy(config)
+            tied_config.tie_encoder_decoder = True
+            tied_model = model_class(config=tied_config).to(torch_device).eval()
+
+            model_result = model(
+                input_ids=input_ids,
+                decoder_input_ids=decoder_input_ids,
+                attention_mask=attention_mask,
+                decoder_attention_mask=decoder_attention_mask,
+                return_dict=True,
+            )
+
+            tied_model_result = tied_model(
+                input_ids=input_ids,
+                decoder_input_ids=decoder_input_ids,
+                attention_mask=attention_mask,
+                decoder_attention_mask=decoder_attention_mask,
+                return_dict=True,
+            )
+
+            # check that models has less parameters
+            self.parent.assertLess(
+                sum(p.numel() for p in tied_model.parameters()), sum(p.numel() for p in model.parameters())
+            )
+            random_slice_idx = ids_tensor((1,), model_result[0].shape[-1]).item()
+
+            # check that outputs are equal
+            self.parent.assertTrue(
+                torch.allclose(
+                    model_result[0][0, :, random_slice_idx], tied_model_result[0][0, :, random_slice_idx], atol=1e-4
+                )
+            )
+
+            # check that outputs after saving and loading are equal
+            with tempfile.TemporaryDirectory() as tmpdirname:
+                tied_model.save_pretrained(tmpdirname)
+                tied_model = model_class.from_pretrained(tmpdirname)
+                tied_model.to(torch_device)
+                tied_model.eval()
+
+                # check that models has less parameters
+                self.parent.assertLess(
+                    sum(p.numel() for p in tied_model.parameters()), sum(p.numel() for p in model.parameters())
+                )
+                random_slice_idx = ids_tensor((1,), model_result[0].shape[-1]).item()
+
+                tied_model_result = tied_model(
+                    input_ids=input_ids,
+                    decoder_input_ids=decoder_input_ids,
+                    attention_mask=attention_mask,
+                    decoder_attention_mask=decoder_attention_mask,
+                )
+
+                # check that outputs are equal
+                self.parent.assertTrue(
+                    torch.allclose(
+                        model_result[0][0, :, random_slice_idx],
+                        tied_model_result[0][0, :, random_slice_idx],
+                        atol=1e-4,
+                    )
+                )
+
+    def check_fast_integration(
+        self,
+        config,
+        *args,
+    ):
+        input_ids = torch.tensor([[7, 4, 78, 0, 24, 52, 43]], device=torch_device, dtype=torch.long)
+        decoder_input_ids = torch.tensor([[12, 62, 25, 11, 47, 15, 14]], device=torch_device, dtype=torch.long)
+        attention_mask = torch.tensor([[1, 1, 1, 0, 1, 0, 0]], device=torch_device, dtype=torch.long)
+        decoder_attention_mask = torch.tensor([[1, 1, 1, 0, 0, 1, 0]], device=torch_device, dtype=torch.long)
+        lm_labels = torch.tensor([[62, 25, 11, 47, 15, 14, 24]], device=torch_device, dtype=torch.long)
+        torch.manual_seed(0)
+        config.ngram = 4
+        model = ProphetNetForConditionalGeneration(config=config)
+        model.to(torch_device)
+        model.eval()
+        with torch.no_grad():
+            result = model(
+                input_ids=input_ids,
+                decoder_input_ids=decoder_input_ids,
+                attention_mask=attention_mask,
+                decoder_attention_mask=decoder_attention_mask,
+                labels=lm_labels,
+                return_dict=True,
+            )
+
+        self.parent.assertTrue(torch.allclose(result.loss, torch.tensor(128.234, device=torch_device), atol=1e-3))
+
+        expected_logit_slice = torch.tensor(
+            [-0.0544, 0.0091, -0.0378, -0.1237, -0.0582, -0.0591, 0.0049], device=torch_device
+        )
+        self.parent.assertTrue(torch.allclose(result.logits[0, :, 1], expected_logit_slice, atol=1e-3))
+
+    def prepare_config_and_inputs_for_common(self):
+        config_and_inputs = self.prepare_config_and_inputs()
+        (
+            config,
+            input_ids,
+            decoder_input_ids,
+            attention_mask,
+            decoder_attention_mask,
+            lm_labels,
+        ) = config_and_inputs
+
+        inputs_dict = {
+            "input_ids": input_ids,
+            "attention_mask": attention_mask,
+            "decoder_input_ids": decoder_input_ids,
+            "decoder_attention_mask": decoder_attention_mask,
+            "use_cache": False,
+        }
+        return config, inputs_dict
+
+
+@require_torch
+class ProphetNetModelTest(ModelTesterMixin, unittest.TestCase):
+
+    all_model_classes = (ProphetNetModel, ProphetNetForConditionalGeneration) if is_torch_available() else ()
+    all_generative_model_classes = (ProphetNetForConditionalGeneration,) if is_torch_available() else ()
+    test_pruning = False
+    test_torchscript = False
+    test_resize_embeddings = False
+    test_headmasking = False
+    is_encoder_decoder = True
+
+    def setUp(self):
+        self.model_tester = ProphetNetModelTester(self)
+        self.config_tester = ConfigTester(self, config_class=ProphetNetConfig)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    def test_fast_integration(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.check_fast_integration(*config_and_inputs)
+
+    def test_shared_weights(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_encoder_decoder_shared_weights(*config_and_inputs)
+
+    def test_shift_labels_via_shift_left(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.check_prepare_lm_labels_via_shift_left(*config_and_inputs)
+
+    def test_decoder_model_past(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_decoder_model_past(*config_and_inputs)
+
+    def test_decoder_model_attn_mask_past(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_decoder_model_attention_mask_past(*config_and_inputs)
+
+    def test_decoder_model_generate(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_generate_with_past_key_value_states(*config_and_inputs)
+
+    def test_fp16_forward(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_model_fp16_forward(*config_and_inputs)
+
+
+class ProphetNetModelIntegrationTest(unittest.TestCase):
+    @slow
+    def test_pretrained_checkpoint_hidden_states(self):
+        model = ProphetNetForConditionalGeneration.from_pretrained("microsoft/prophetnet-large-uncased")
+        model.to(torch_device)
+
+        # encoder-decoder outputs
+        encoder_ids = torch.tensor(
+            [
+                [
+                    2871,
+                    102,
+                    2048,
+                    3176,
+                    2780,
+                    1997,
+                    2871,
+                    26727,
+                    2169,
+                    2097,
+                    12673,
+                    1996,
+                    8457,
+                    2006,
+                    2049,
+                    8240,
+                    2859,
+                    2799,
+                    1012,
+                    2023,
+                    6512,
+                    2038,
+                    2174,
+                    13977,
+                    2195,
+                    25962,
+                    1012,
+                    102,
+                ]
+            ]
+        ).to(torch_device)
+
+        decoder_prev_ids = torch.tensor([[102, 2129, 2116, 2372, 2024, 2006, 2169, 1997, 2122, 2048, 2780, 1029]]).to(
+            torch_device
+        )
+        output = model(
+            input_ids=encoder_ids,
+            attention_mask=None,
+            encoder_outputs=None,
+            decoder_input_ids=decoder_prev_ids,
+        )
+        output_predited_logis = output[0]
+        expected_shape = torch.Size((1, 12, 30522))
+        self.assertEqual(output_predited_logis.shape, expected_shape)
+        expected_slice = torch.tensor(
+            [[[-7.6213, -7.9008, -7.9979], [-7.6834, -7.8467, -8.2187], [-7.5326, -7.4762, -8.1914]]]
+        ).to(torch_device)
+        self.assertTrue(torch.allclose(output_predited_logis[:, :3, :3], expected_slice, atol=1e-4))
+
+        # encoder outputs
+        encoder_outputs = model.prophetnet.encoder(encoder_ids)[0]
+        expected_encoder_outputs_slice = torch.tensor(
+            [[[-0.2526, -0.1951, -0.2185], [-0.8923, 0.2992, -0.4623], [-0.4585, 0.0165, -0.6652]]]
+        ).to(torch_device)
+        expected_shape_encoder = torch.Size((1, 28, 1024))
+        self.assertEqual(encoder_outputs.shape, expected_shape_encoder)
+        self.assertTrue(torch.allclose(encoder_outputs[:, :3, :3], expected_encoder_outputs_slice, atol=1e-4))
+
+        # decoder outputs
+        decoder_outputs = model.prophetnet.decoder(decoder_prev_ids, encoder_hidden_states=encoder_outputs)
+        predicting_streams = decoder_outputs[0].view(1, model.config.ngram + 1, 12, -1)[:, 1:]
+        predicting_streams_logits = model.lm_head(predicting_streams)
+        next_first_stream_logits = predicting_streams_logits[:, 0]
+        self.assertTrue(torch.allclose(next_first_stream_logits[:, :3, :3], expected_slice, atol=1e-4))
+
+    @slow
+    def test_cnndm_inference(self):
+        model = ProphetNetForConditionalGeneration.from_pretrained("microsoft/prophetnet-large-uncased-cnndm")
+        model.to(torch_device)
+
+        tokenizer = ProphetNetTokenizer.from_pretrained("microsoft/prophetnet-large-uncased-cnndm")
+
+        ARTICLE_TO_SUMMARIZE = "USTC was founded in Beijing by the Chinese Academy of Sciences (CAS) in September 1958. The Director of CAS, Mr. Guo Moruo was appointed the first president of USTC. USTC's founding mission was to develop a high-level science and technology workforce, as deemed critical for development of China's economy, defense, and science and technology education. The establishment was hailed as \"A Major Event in the History of Chinese Education and Science.\" CAS has supported USTC by combining most of its institutes with the departments of the university. USTC is listed in the top 16 national key universities, becoming the youngest national key university.".lower()
+        input_ids = tokenizer([ARTICLE_TO_SUMMARIZE], max_length=511, return_tensors="pt").input_ids
+
+        input_ids = input_ids.to(torch_device)
+
+        summary_ids = model.generate(
+            input_ids, num_beams=4, length_penalty=1.0, no_repeat_ngram_size=3, early_stopping=True
+        )
+        EXPECTED_SUMMARIZE_512 = "us ##tc was founded by the chinese academy of sciences ( cas ) in 1958 . [X_SEP] us ##tc is listed in the top 16 national key universities ."
+        generated_titles = [
+            " ".join(tokenizer.convert_ids_to_tokens(g, skip_special_tokens=True)) for g in summary_ids
+        ]
+        self.assertListEqual(
+            [EXPECTED_SUMMARIZE_512],
+            generated_titles,
+        )
+        input_ids = tokenizer([ARTICLE_TO_SUMMARIZE], max_length=99, return_tensors="pt").input_ids
+        input_ids = input_ids.to(torch_device)
+        # actually 98 tokens are used. max_length=100 contains bos and eos.
+        summary_ids = model.generate(
+            input_ids, num_beams=4, length_penalty=1.0, no_repeat_ngram_size=3, early_stopping=True
+        )
+        EXPECTED_SUMMARIZE_100 = (
+            r"us ##tc was founded in beijing by the chinese academy of sciences ( cas ) in 1958 . [X_SEP] us ##tc "
+            "'"
+            ' s founding mission was to develop a high - level science and technology workforce . [X_SEP] establishment hailed as " a major event in the history of chinese education and science "'
+        )
+        generated_titles = [
+            " ".join(tokenizer.convert_ids_to_tokens(g, skip_special_tokens=True)) for g in summary_ids
+        ]
+        self.assertListEqual(
+            [EXPECTED_SUMMARIZE_100],
+            generated_titles,
+        )

tests/test_modeling_t5.py

@@ -44,7 +44,6 @@ class T5ModelTester:
         encoder_seq_length=7,
         decoder_seq_length=9,
         # For common tests
-        seq_length=7,
         is_training=True,
         use_attention_mask=True,
         use_labels=True,

tests/test_modeling_xlm_prophetnet.py

@@ -0,0 +1,142 @@
+# coding=utf-8
+# Copyright 2018 The Google AI Language Team Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import unittest
+
+from transformers import is_torch_available
+from transformers.testing_utils import slow, torch_device
+
+
+if is_torch_available():
+    import torch
+
+    from transformers import XLMProphetNetForConditionalGeneration, XLMProphetNetTokenizer
+
+
+class XLMProphetNetModelIntegrationTest(unittest.TestCase):
+    @slow
+    def test_pretrained_checkpoint_hidden_states(self):
+        model = XLMProphetNetForConditionalGeneration.from_pretrained(
+            "microsoft/xprophetnet-large-wiki100-cased",
+        )
+        model.to(torch_device)
+
+        # encoder-decoder outputs
+        encoder_ids = torch.tensor([[17, 96208, 103471, 2]]).to(torch_device)
+        decoder_prev_ids = torch.tensor(
+            [[2, 250, 9953, 34, 69489, 1620, 32, 118424, 624, 210, 105, 2913, 1032, 351]]
+        ).to(torch_device)
+        output = model(
+            input_ids=encoder_ids, attention_mask=None, encoder_outputs=None, decoder_input_ids=decoder_prev_ids
+        )
+        output_predited_logis = output[0]
+        expected_shape = torch.Size((1, 14, 250012))
+        self.assertEqual(output_predited_logis.shape, expected_shape)
+        expected_slice = torch.tensor(
+            [[[-6.6042, -8.3838, 12.4717], [-6.4426, -8.1994, 12.4542], [-6.0851, -7.8209, 12.9493]]]
+        ).to(torch_device)
+        self.assertTrue(torch.allclose(output_predited_logis[:, :3, :3], expected_slice, atol=1e-4))
+
+        # encoder outputs
+        encoder_outputs = model.prophetnet.encoder(encoder_ids)[0]
+        expected_encoder_outputs_slice = torch.tensor(
+            [[[-1.4260, -0.7628, 0.8453], [-1.4719, -0.1391, 0.7807], [-1.7678, 0.0114, 0.4646]]]
+        ).to(torch_device)
+        expected_shape_encoder = torch.Size((1, 4, 1024))
+        self.assertEqual(encoder_outputs.shape, expected_shape_encoder)
+        self.assertTrue(torch.allclose(encoder_outputs[:, :3, :3], expected_encoder_outputs_slice, atol=1e-4))
+
+        # decoder outputs
+        decoder_outputs = model.prophetnet.decoder(
+            decoder_prev_ids,
+            encoder_hidden_states=encoder_outputs,
+        )
+        predicting_streams = decoder_outputs[0].view(1, model.config.ngram + 1, 14, -1)[:, 1:]
+        predicting_streams_logits = model.lm_head(predicting_streams)
+        next_first_stream_logits = predicting_streams_logits[:, 0]
+        self.assertTrue(torch.allclose(next_first_stream_logits[:, :3, :3], expected_slice, atol=1e-4))
+
+    @slow
+    def test_ntg_hidden_states(self):
+        model = XLMProphetNetForConditionalGeneration.from_pretrained(
+            "microsoft/xprophetnet-large-wiki100-cased-xglue-ntg",
+        )
+        model.to(torch_device)
+
+        encoder_ids = torch.tensor([[17, 96208, 103471, 2]]).to(torch_device)
+        decoder_prev_ids = torch.tensor(
+            [[2, 250, 9953, 34, 69489, 1620, 32, 118424, 624, 210, 105, 2913, 1032, 351]]
+        ).to(torch_device)
+        output = model(
+            input_ids=encoder_ids, attention_mask=None, encoder_outputs=None, decoder_input_ids=decoder_prev_ids
+        )
+        output_predited_logis = output[0]
+        expected_shape = torch.Size((1, 14, 250012))
+        self.assertEqual(output_predited_logis.shape, expected_shape)
+        # compare the actual values for a slice.
+        expected_slice = torch.tensor(
+            [[[-8.8815, -9.2996, -4.4506], [-6.7202, -7.8944, -0.9402], [-8.6890, -7.4528, -1.9437]]]
+        ).to(torch_device)
+
+        self.assertTrue(torch.allclose(output_predited_logis[:, :3, :3], expected_slice, atol=1e-4))
+
+    @slow
+    def test_xprophetnet_ntg_inference(self):
+        model = XLMProphetNetForConditionalGeneration.from_pretrained(
+            "microsoft/xprophetnet-large-wiki100-cased-xglue-ntg",
+        )
+        model.to(torch_device)
+
+        tokenizer = XLMProphetNetTokenizer.from_pretrained("microsoft/xprophetnet-large-wiki100-cased-xglue-ntg")
+
+        EN_SENTENCE = "Microsoft Corporation intends to officially end free support for the Windows 7 operating system after January 14, 2020, according to the official portal of the organization. From that day, users of this system will not be able to receive security updates, which could make their computers vulnerable to cyber attacks."
+        RU_SENTENCE = "орпорация Microsoft намерена официально прекратить бесплатную поддержку операционной системы Windows 7 после 14 января 2020 года, сообщается на официальном портале организации . С указанного дня пользователи этой системы не смогут получать обновления безопасности, из-за чего их компьютеры могут стать уязвимыми к кибератакам."
+        ZH_SENTENCE = (
+            "根据该组织的官方门户网站，微软公司打算在2020年1月14日之后正式终止对Windows 7操作系统的免费支持。从那时起，该系统的用户将无法接收安全更新，这可能会使他们的计算机容易受到网络攻击。"
+        )
+
+        input_ids = tokenizer(
+            [EN_SENTENCE, RU_SENTENCE, ZH_SENTENCE], padding=True, max_length=255, return_tensors="pt"
+        ).input_ids
+        input_ids = input_ids.to(torch_device)
+
+        summary_ids = model.generate(
+            input_ids, num_beams=10, length_penalty=1.0, no_repeat_ngram_size=3, early_stopping=True
+        )
+        generated_titles = [tokenizer.decode(g, skip_special_tokens=True) for g in summary_ids]
+        EXPECTED_TITLE_EN = "Microsoft to end Windows 7 free support after January 14, 2020"
+        EXPECTED_TITLE_RU = "Microsoft намерена прекратить бесплатную поддержку Windows 7 после 14 января 2020 года"
+        EXPECTED_TITLE_ZH = "微软打算终止对Windows 7操作系统的免费支持"
+        self.assertListEqual(
+            [EXPECTED_TITLE_EN, EXPECTED_TITLE_RU, EXPECTED_TITLE_ZH],
+            generated_titles,
+        )
+
+        summary_ids_beam1 = model.generate(
+            input_ids, num_beams=1, length_penalty=1.0, no_repeat_ngram_size=3, early_stopping=True
+        )
+        generated_titles_beam1_tok = [
+            tokenizer.convert_ids_to_tokens(g, skip_special_tokens=True) for g in summary_ids_beam1
+        ]
+        EXPECTED_TITLE_EN_BEAM1_TOK = "▁Microsoft ▁to ▁end ▁free ▁support ▁for ▁Windows ▁7".split(" ")
+        EXPECTED_TITLE_RU_BEAM1_TOK = "▁Microsoft ▁намерен а ▁прекрати ть ▁бес плат ную ▁поддержку ▁Windows ▁7 ▁после ▁14 ▁января ▁2020 ▁года".split(
+            " "
+        )
+        EXPECTED_TITLE_ZH_BEAM1_TOK = "微软 公司 打算 终止 对 Windows ▁7 操作 系统的 免费 支持".split(" ")
+        self.assertListEqual(
+            [EXPECTED_TITLE_EN_BEAM1_TOK, EXPECTED_TITLE_RU_BEAM1_TOK, EXPECTED_TITLE_ZH_BEAM1_TOK],
+            generated_titles_beam1_tok,
+        )

tests/test_tokenization_bert.py

@@ -222,17 +222,6 @@ class BertTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
         self.assertFalse(_is_punctuation("A"))
         self.assertFalse(_is_punctuation(" "))
 
-    def test_clean_text(self):
-        tokenizer = self.get_tokenizer()
-        rust_tokenizer = self.get_rust_tokenizer()
-
-        # Example taken from the issue https://github.com/huggingface/tokenizers/issues/340
-        self.assertListEqual([tokenizer.tokenize(t) for t in ["Test", "\xad", "test"]], [["[UNK]"], [], ["[UNK]"]])
-
-        self.assertListEqual(
-            [rust_tokenizer.tokenize(t) for t in ["Test", "\xad", "test"]], [["[UNK]"], [], ["[UNK]"]]
-        )
-
     @slow
     def test_sequence_builders(self):
         tokenizer = self.tokenizer_class.from_pretrained("bert-base-uncased")

tests/test_tokenization_prophetnet.py

@@ -0,0 +1,191 @@
+# coding=utf-8
+# Copyright 2018 The Google AI Language Team Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import os
+import unittest
+
+from transformers.testing_utils import slow
+from transformers.tokenization_bert import (
+    BasicTokenizer,
+    WordpieceTokenizer,
+    _is_control,
+    _is_punctuation,
+    _is_whitespace,
+)
+from transformers.tokenization_prophetnet import VOCAB_FILES_NAMES, ProphetNetTokenizer
+
+from .test_tokenization_common import TokenizerTesterMixin
+
+
+class ProphetNetTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
+
+    tokenizer_class = ProphetNetTokenizer
+    test_rust_tokenizer = False
+
+    def setUp(self):
+        super().setUp()
+
+        vocab_tokens = [
+            "[UNK]",
+            "[CLS]",
+            "[SEP]",
+            "[PAD]",
+            "[MASK]",
+            "want",
+            "##want",
+            "##ed",
+            "wa",
+            "un",
+            "runn",
+            "##ing",
+            ",",
+            "low",
+            "lowest",
+        ]
+        self.vocab_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES["vocab_file"])
+        with open(self.vocab_file, "w", encoding="utf-8") as vocab_writer:
+            vocab_writer.write("".join([x + "\n" for x in vocab_tokens]))
+
+    def get_input_output_texts(self, tokenizer):
+        input_text = "UNwant\u00E9d,running"
+        output_text = "unwanted, running"
+        return input_text, output_text
+
+    def test_full_tokenizer(self):
+        tokenizer = self.tokenizer_class(self.vocab_file)
+
+        tokens = tokenizer.tokenize("UNwant\u00E9d,running")
+        self.assertListEqual(tokens, ["un", "##want", "##ed", ",", "runn", "##ing"])
+        self.assertListEqual(tokenizer.convert_tokens_to_ids(tokens), [9, 6, 7, 12, 10, 11])
+
+    def test_chinese(self):
+        tokenizer = BasicTokenizer()
+
+        self.assertListEqual(tokenizer.tokenize("ah\u535A\u63A8zz"), ["ah", "\u535A", "\u63A8", "zz"])
+
+    def test_basic_tokenizer_lower(self):
+        tokenizer = BasicTokenizer(do_lower_case=True)
+
+        self.assertListEqual(
+            tokenizer.tokenize(" \tHeLLo!how  \n Are yoU?  "), ["hello", "!", "how", "are", "you", "?"]
+        )
+        self.assertListEqual(tokenizer.tokenize("H\u00E9llo"), ["hello"])
+
+    def test_basic_tokenizer_lower_strip_accents_false(self):
+        tokenizer = BasicTokenizer(do_lower_case=True, strip_accents=False)
+
+        self.assertListEqual(
+            tokenizer.tokenize(" \tHäLLo!how  \n Are yoU?  "), ["hällo", "!", "how", "are", "you", "?"]
+        )
+        self.assertListEqual(tokenizer.tokenize("H\u00E9llo"), ["h\u00E9llo"])
+
+    def test_basic_tokenizer_lower_strip_accents_true(self):
+        tokenizer = BasicTokenizer(do_lower_case=True, strip_accents=True)
+
+        self.assertListEqual(
+            tokenizer.tokenize(" \tHäLLo!how  \n Are yoU?  "), ["hallo", "!", "how", "are", "you", "?"]
+        )
+        self.assertListEqual(tokenizer.tokenize("H\u00E9llo"), ["hello"])
+
+    def test_basic_tokenizer_lower_strip_accents_default(self):
+        tokenizer = BasicTokenizer(do_lower_case=True)
+
+        self.assertListEqual(
+            tokenizer.tokenize(" \tHäLLo!how  \n Are yoU?  "), ["hallo", "!", "how", "are", "you", "?"]
+        )
+        self.assertListEqual(tokenizer.tokenize("H\u00E9llo"), ["hello"])
+
+    def test_basic_tokenizer_no_lower(self):
+        tokenizer = BasicTokenizer(do_lower_case=False)
+
+        self.assertListEqual(
+            tokenizer.tokenize(" \tHeLLo!how  \n Are yoU?  "), ["HeLLo", "!", "how", "Are", "yoU", "?"]
+        )
+
+    def test_basic_tokenizer_no_lower_strip_accents_false(self):
+        tokenizer = BasicTokenizer(do_lower_case=False, strip_accents=False)
+
+        self.assertListEqual(
+            tokenizer.tokenize(" \tHäLLo!how  \n Are yoU?  "), ["HäLLo", "!", "how", "Are", "yoU", "?"]
+        )
+
+    def test_basic_tokenizer_no_lower_strip_accents_true(self):
+        tokenizer = BasicTokenizer(do_lower_case=False, strip_accents=True)
+
+        self.assertListEqual(
+            tokenizer.tokenize(" \tHäLLo!how  \n Are yoU?  "), ["HaLLo", "!", "how", "Are", "yoU", "?"]
+        )
+
+    def test_basic_tokenizer_respects_never_split_tokens(self):
+        tokenizer = BasicTokenizer(do_lower_case=False, never_split=["[UNK]"])
+
+        self.assertListEqual(
+            tokenizer.tokenize(" \tHeLLo!how  \n Are yoU? [UNK]"), ["HeLLo", "!", "how", "Are", "yoU", "?", "[UNK]"]
+        )
+
+    def test_wordpiece_tokenizer(self):
+        vocab_tokens = ["[UNK]", "[CLS]", "[SEP]", "want", "##want", "##ed", "wa", "un", "runn", "##ing"]
+
+        vocab = {}
+        for (i, token) in enumerate(vocab_tokens):
+            vocab[token] = i
+        tokenizer = WordpieceTokenizer(vocab=vocab, unk_token="[UNK]")
+
+        self.assertListEqual(tokenizer.tokenize(""), [])
+
+        self.assertListEqual(tokenizer.tokenize("unwanted running"), ["un", "##want", "##ed", "runn", "##ing"])
+
+        self.assertListEqual(tokenizer.tokenize("unwantedX running"), ["[UNK]", "runn", "##ing"])
+
+    def test_is_whitespace(self):
+        self.assertTrue(_is_whitespace(" "))
+        self.assertTrue(_is_whitespace("\t"))
+        self.assertTrue(_is_whitespace("\r"))
+        self.assertTrue(_is_whitespace("\n"))
+        self.assertTrue(_is_whitespace("\u00A0"))
+
+        self.assertFalse(_is_whitespace("A"))
+        self.assertFalse(_is_whitespace("-"))
+
+    def test_is_control(self):
+        self.assertTrue(_is_control("\u0005"))
+
+        self.assertFalse(_is_control("A"))
+        self.assertFalse(_is_control(" "))
+        self.assertFalse(_is_control("\t"))
+        self.assertFalse(_is_control("\r"))
+
+    def test_is_punctuation(self):
+        self.assertTrue(_is_punctuation("-"))
+        self.assertTrue(_is_punctuation("$"))
+        self.assertTrue(_is_punctuation("`"))
+        self.assertTrue(_is_punctuation("."))
+
+        self.assertFalse(_is_punctuation("A"))
+        self.assertFalse(_is_punctuation(" "))
+
+    @slow
+    def test_sequence_builders(self):
+        tokenizer = self.tokenizer_class.from_pretrained("microsoft/prophetnet-large-uncased")
+
+        text = tokenizer.encode("sequence builders", add_special_tokens=False)
+        text_2 = tokenizer.encode("multi-sequence build", add_special_tokens=False)
+
+        encoded_sentence = tokenizer.build_inputs_with_special_tokens(text)
+        encoded_pair = tokenizer.build_inputs_with_special_tokens(text, text_2)
+
+        assert encoded_sentence == text + [102]
+        assert encoded_pair == text + [102] + text_2 + [102]

tests/test_tokenization_xlm_prophetnet.py

@@ -0,0 +1,125 @@
+# coding=utf-8
+# Copyright 2020 The Google AI Language Team Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import os
+import unittest
+
+from transformers.file_utils import cached_property
+from transformers.testing_utils import slow
+from transformers.tokenization_xlm_prophetnet import SPIECE_UNDERLINE, XLMProphetNetTokenizer
+
+from .test_tokenization_common import TokenizerTesterMixin
+
+
+SAMPLE_VOCAB = os.path.join(os.path.dirname(os.path.abspath(__file__)), "fixtures/test_sentencepiece.model")
+
+
+class XLMRobertaTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
+
+    tokenizer_class = XLMProphetNetTokenizer
+    test_rust_tokenizer = False
+
+    def setUp(self):
+        super().setUp()
+
+        # We have a SentencePiece fixture for testing
+        tokenizer = XLMProphetNetTokenizer(SAMPLE_VOCAB, keep_accents=True)
+        tokenizer.save_pretrained(self.tmpdirname)
+
+    def test_full_tokenizer(self):
+        tokenizer = XLMProphetNetTokenizer(SAMPLE_VOCAB, keep_accents=True)
+
+        tokens = tokenizer.tokenize("This is a test")
+        self.assertListEqual(tokens, ["▁This", "▁is", "▁a", "▁t", "est"])
+
+        self.assertListEqual(
+            tokenizer.convert_tokens_to_ids(tokens),
+            [value + tokenizer.fairseq_offset for value in [285, 46, 10, 170, 382]],
+        )
+
+        tokens = tokenizer.tokenize("I was born in 92000, and this is falsé.")
+        self.assertListEqual(
+            tokens,
+            [
+                SPIECE_UNDERLINE + "I",
+                SPIECE_UNDERLINE + "was",
+                SPIECE_UNDERLINE + "b",
+                "or",
+                "n",
+                SPIECE_UNDERLINE + "in",
+                SPIECE_UNDERLINE + "",
+                "9",
+                "2",
+                "0",
+                "0",
+                "0",
+                ",",
+                SPIECE_UNDERLINE + "and",
+                SPIECE_UNDERLINE + "this",
+                SPIECE_UNDERLINE + "is",
+                SPIECE_UNDERLINE + "f",
+                "al",
+                "s",
+                "é",
+                ".",
+            ],
+        )
+        ids = tokenizer.convert_tokens_to_ids(tokens)
+        self.assertListEqual(
+            ids,
+            [
+                value + tokenizer.fairseq_offset
+                for value in [8, 21, 84, 55, 24, 19, 7, -9, 602, 347, 347, 347, 3, 12, 66, 46, 72, 80, 6, -9, 4]
+            ],
+        )
+
+        back_tokens = tokenizer.convert_ids_to_tokens(ids)
+        self.assertListEqual(
+            back_tokens,
+            [
+                SPIECE_UNDERLINE + "I",
+                SPIECE_UNDERLINE + "was",
+                SPIECE_UNDERLINE + "b",
+                "or",
+                "n",
+                SPIECE_UNDERLINE + "in",
+                SPIECE_UNDERLINE + "",
+                "[UNK]",
+                "2",
+                "0",
+                "0",
+                "0",
+                ",",
+                SPIECE_UNDERLINE + "and",
+                SPIECE_UNDERLINE + "this",
+                SPIECE_UNDERLINE + "is",
+                SPIECE_UNDERLINE + "f",
+                "al",
+                "s",
+                "[UNK]",
+                ".",
+            ],
+        )
+
+    @cached_property
+    def big_tokenizer(self):
+        return XLMProphetNetTokenizer.from_pretrained("microsoft/xprophetnet-large-wiki100-cased")
+
+    @slow
+    def test_tokenization_base_easy_symbols(self):
+        symbols = "Hello World!"
+        original_tokenizer_encodings = [35389, 6672, 49, 2]
+        self.assertListEqual(original_tokenizer_encodings, self.big_tokenizer.encode(symbols))

utils/check_copies.py

@@ -24,7 +24,6 @@ import tempfile
 # python utils/check_copies.py
 TRANSFORMERS_PATH = "src/transformers"
 PATH_TO_DOCS = "docs/source"
-REPO_PATH = "."
 
 
 def find_code_in_transformers(object_name):
@@ -176,7 +175,7 @@ def get_model_list():
     # If the introduction or the conclusion of the list change, the prompts may need to be updated.
     _start_prompt = "🤗 Transformers currently provides the following architectures"
     _end_prompt = "1. Want to contribute a new model?"
-    with open(os.path.join(REPO_PATH, "README.md"), "r", encoding="utf-8") as f:
+    with open(os.path.join("README.md"), "r", encoding="utf-8") as f:
         lines = f.readlines()
     # Find the start of the list.
     start_index = 0
@@ -220,17 +219,7 @@ def split_long_line_with_indent(line, max_per_line, indent):
 def convert_to_rst(model_list, max_per_line=None):
     """ Convert `model_list` to rst format. """
     # Convert **[description](link)** to `description <link>`__
-    def _rep_link(match):
-        title, link = match.groups()
-        # Keep hard links for the models not released yet
-        if "master" in link or not link.startswith("https://huggingface.co/transformers"):
-            return f"`{title} <{link}>`__"
-        # Convert links to relative links otherwise
-        else:
-            link = link[len("https://huggingface.co/transformers/") : -len(".html")]
-            return f":doc:`{title} <{link}>`"
-
-    model_list = re.sub(r"\*\*\[([^\]]*)\]\(([^\)]*)\)\*\*", _rep_link, model_list)
+    model_list = re.sub(r"\*\*\[([^\]]*)\]\(([^\)]*)\)\*\*", r"`\1 <\2>`__", model_list)
 
     # Convert [description](link) to `description <link>`__
     model_list = re.sub(r"\[([^\]]*)\]\(([^\)]*)\)", r"`\1 <\2>`__", model_list)