Add CANINE (#12024)

* First pass

* More progress

* Add support for local attention

* More improvements

* More improvements

* Conversion script working

* Add CanineTokenizer

* Make style & quality

* First draft of integration test

* Remove decoder test

* Improve tests

* Add documentation

* Mostly docs improvements

* Add CanineTokenizer tests

* Fix most tests on GPU, improve upsampling projection

* Address most comments by @dhgarrette

* Remove decoder logic

* Improve Canine tests, improve docs of CanineConfig

* All tokenizer tests passing

* Make fix-copies and fix tokenizer tests

* Fix test_model_outputs_equivalence test

* Apply suggestions from @sgugger's review
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Address some more comments

* Add support for hidden_states and attentions of shallow encoders

* Define custom CanineModelOutputWithPooling, tests pass

* First pass

* More progress

* Add support for local attention

* More improvements

* More improvements

* Conversion script working

* Add CanineTokenizer

* Make style & quality

* First draft of integration test

* Remove decoder test

* Improve tests

* Add documentation

* Mostly docs improvements

* Add CanineTokenizer tests

* Fix most tests on GPU, improve upsampling projection

* Address most comments by @dhgarrette

* Remove decoder logic

* Improve Canine tests, improve docs of CanineConfig

* All tokenizer tests passing

* Make fix-copies and fix tokenizer tests

* Fix test_model_outputs_equivalence test

* Apply suggestions from @sgugger's review
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Address some more comments

* Make conversion script work for Canine-c too

* Fix tokenizer tests

* Remove file
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

README.md

@@ -212,7 +212,8 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[BORT](https://huggingface.co/transformers/model_doc/bort.html)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
 1. **[ByT5](https://huggingface.co/transformers/model_doc/byt5.html)** (from Google Research) released with the paper [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) by Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel.
 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. **[CLIP](https://huggingface.co/transformers/model_doc/clip.html)** from (OpenAI) released with the paper [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) by Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever.
+1. **[CANINE](https://huggingface.co/transformers/model_doc/canine.html)** (from Google Research) released with the paper [CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation](https://arxiv.org/abs/2103.06874) by Jonathan H. Clark, Dan Garrette, Iulia Turc, John Wieting.
+1. **[CLIP](https://huggingface.co/transformers/model_doc/clip.html)** (from OpenAI) released with the paper [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) by Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever.
 1. **[ConvBERT](https://huggingface.co/transformers/model_doc/convbert.html)** (from YituTech) released with the paper [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
 1. **[CPM](https://huggingface.co/transformers/model_doc/cpm.html)** (from Tsinghua University) released with the paper [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun.
 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.

docs/source/index.rst

@@ -131,158 +131,161 @@ Supported models
 12. :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.
-13. :doc:`CLIP <model_doc/clip>` from (OpenAI) released with the paper `Learning Transferable Visual Models From
+13. :doc:`CANINE <model_doc/canine>` (from Google Research) released with the paper `CANINE: Pre-training an Efficient
+    Tokenization-Free Encoder for Language Representation <https://arxiv.org/abs/2103.06874>`__ by Jonathan H. Clark,
+    Dan Garrette, Iulia Turc, John Wieting.
+14. :doc:`CLIP <model_doc/clip>` (from OpenAI) released with the paper `Learning Transferable Visual Models From
     Natural Language Supervision <https://arxiv.org/abs/2103.00020>`__ by Alec Radford, Jong Wook Kim, Chris Hallacy,
     Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen
     Krueger, Ilya Sutskever.
-14. :doc:`ConvBERT <model_doc/convbert>` (from YituTech) released with the paper `ConvBERT: Improving BERT with
+15. :doc:`ConvBERT <model_doc/convbert>` (from YituTech) released with the paper `ConvBERT: Improving BERT with
     Span-based Dynamic Convolution <https://arxiv.org/abs/2008.02496>`__ by Zihang Jiang, Weihao Yu, Daquan Zhou,
     Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
-15. :doc:`CPM <model_doc/cpm>` (from Tsinghua University) released with the paper `CPM: A Large-scale Generative
+16. :doc:`CPM <model_doc/cpm>` (from Tsinghua University) released with the paper `CPM: A Large-scale Generative
     Chinese Pre-trained Language Model <https://arxiv.org/abs/2012.00413>`__ by Zhengyan Zhang, Xu Han, Hao Zhou, Pei
     Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng,
     Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang,
     Juanzi Li, Xiaoyan Zhu, Maosong Sun.
-16. :doc:`CTRL <model_doc/ctrl>` (from Salesforce) released with the paper `CTRL: A Conditional Transformer Language
+17. :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.
-17. :doc:`DeBERTa <model_doc/deberta>` (from Microsoft) released with the paper `DeBERTa: Decoding-enhanced BERT with
+18. :doc:`DeBERTa <model_doc/deberta>` (from Microsoft) 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.
-18. :doc:`DeBERTa-v2 <model_doc/deberta_v2>` (from Microsoft) released with the paper `DeBERTa: Decoding-enhanced BERT
+19. :doc:`DeBERTa-v2 <model_doc/deberta_v2>` (from Microsoft) 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.
-19. :doc:`DeiT <model_doc/deit>` (from Facebook) released with the paper `Training data-efficient image transformers &
+20. :doc:`DeiT <model_doc/deit>` (from Facebook) released with the paper `Training data-efficient image transformers &
     distillation through attention <https://arxiv.org/abs/2012.12877>`__ by Hugo Touvron, Matthieu Cord, Matthijs
     Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
-20. :doc:`DETR <model_doc/detr>` (from Facebook) released with the paper `End-to-End Object Detection with Transformers
+21. :doc:`DETR <model_doc/detr>` (from Facebook) released with the paper `End-to-End Object Detection with Transformers
     <https://arxiv.org/abs/2005.12872>`__ by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier,
     Alexander Kirillov, Sergey Zagoruyko.
-21. :doc:`DialoGPT <model_doc/dialogpt>` (from Microsoft Research) released with the paper `DialoGPT: Large-Scale
+22. :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.
-22. :doc:`DistilBERT <model_doc/distilbert>` (from HuggingFace), released together with the paper `DistilBERT, a
+23. :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.
-23. :doc:`DPR <model_doc/dpr>` (from Facebook) released with the paper `Dense Passage Retrieval for Open-Domain
+24. :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.
-24. :doc:`ELECTRA <model_doc/electra>` (from Google Research/Stanford University) released with the paper `ELECTRA:
+25. :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.
-25. :doc:`FlauBERT <model_doc/flaubert>` (from CNRS) released with the paper `FlauBERT: Unsupervised Language Model
+26. :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.
-26. :doc:`Funnel Transformer <model_doc/funnel>` (from CMU/Google Brain) released with the paper `Funnel-Transformer:
+27. :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.
-27. :doc:`GPT <model_doc/gpt>` (from OpenAI) released with the paper `Improving Language Understanding by Generative
+28. :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.
-28. :doc:`GPT-2 <model_doc/gpt2>` (from OpenAI) released with the paper `Language Models are Unsupervised Multitask
+29. :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**.
-29. :doc:`GPT Neo <model_doc/gpt_neo>` (from EleutherAI) released in the repository `EleutherAI/gpt-neo
+30. :doc:`GPT Neo <model_doc/gpt_neo>` (from EleutherAI) released in the repository `EleutherAI/gpt-neo
     <https://github.com/EleutherAI/gpt-neo>`__ by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
-30. :doc:`Hubert <model_doc/hubert>` (from Facebook) released with the paper `HuBERT: Self-Supervised Speech
+31. :doc:`Hubert <model_doc/hubert>` (from Facebook) released with the paper `HuBERT: Self-Supervised Speech
     Representation Learning by Masked Prediction of Hidden Units <https://arxiv.org/abs/2106.07447>`__ by Wei-Ning Hsu,
     Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed.
-31. :doc:`I-BERT <model_doc/ibert>` (from Berkeley) released with the paper `I-BERT: Integer-only BERT Quantization
+32. :doc:`I-BERT <model_doc/ibert>` (from Berkeley) released with the paper `I-BERT: Integer-only BERT Quantization
     <https://arxiv.org/abs/2101.01321>`__ by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer
-32. :doc:`LayoutLM <model_doc/layoutlm>` (from Microsoft Research Asia) released with the paper `LayoutLM: Pre-training
+33. :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.
-33. :doc:`LED <model_doc/led>` (from AllenAI) released with the paper `Longformer: The Long-Document Transformer
+34. :doc:`LED <model_doc/led>` (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.
-34. :doc:`Longformer <model_doc/longformer>` (from AllenAI) released with the paper `Longformer: The Long-Document
+35. :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.
-35. :doc:`LUKE <model_doc/luke>` (from Studio Ousia) released with the paper `LUKE: Deep Contextualized Entity
+36. :doc:`LUKE <model_doc/luke>` (from Studio Ousia) released with the paper `LUKE: Deep Contextualized Entity
     Representations with Entity-aware Self-attention <https://arxiv.org/abs/2010.01057>`__ by Ikuya Yamada, Akari Asai,
     Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
-36. :doc:`LXMERT <model_doc/lxmert>` (from UNC Chapel Hill) released with the paper `LXMERT: Learning Cross-Modality
+37. :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.
-37. :doc:`M2M100 <model_doc/m2m_100>` (from Facebook) released with the paper `Beyond English-Centric Multilingual
+38. :doc:`M2M100 <model_doc/m2m_100>` (from Facebook) released with the paper `Beyond English-Centric Multilingual
     Machine Translation <https://arxiv.org/abs/2010.11125>`__ by by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi
     Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman
     Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
-38. :doc:`MarianMT <model_doc/marian>` Machine translation models trained using `OPUS <http://opus.nlpl.eu/>`__ data by
+39. :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.
-39. :doc:`MBart <model_doc/mbart>` (from Facebook) released with the paper `Multilingual Denoising Pre-training for
+40. :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.
-40. :doc:`MBart-50 <model_doc/mbart>` (from Facebook) released with the paper `Multilingual Translation with Extensible
+41. :doc:`MBart-50 <model_doc/mbart>` (from Facebook) released with the paper `Multilingual Translation with Extensible
     Multilingual Pretraining and Finetuning <https://arxiv.org/abs/2008.00401>`__ by Yuqing Tang, Chau Tran, Xian Li,
     Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan.
-41. :doc:`Megatron-BERT <model_doc/megatron_bert>` (from NVIDIA) released with the paper `Megatron-LM: Training
+42. :doc:`Megatron-BERT <model_doc/megatron_bert>` (from NVIDIA) released with the paper `Megatron-LM: Training
     Multi-Billion Parameter Language Models Using Model Parallelism <https://arxiv.org/abs/1909.08053>`__ by Mohammad
     Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
-42. :doc:`Megatron-GPT2 <model_doc/megatron_gpt2>` (from NVIDIA) released with the paper `Megatron-LM: Training
+43. :doc:`Megatron-GPT2 <model_doc/megatron_gpt2>` (from NVIDIA) released with the paper `Megatron-LM: Training
     Multi-Billion Parameter Language Models Using Model Parallelism <https://arxiv.org/abs/1909.08053>`__ by Mohammad
     Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
-43. :doc:`MPNet <model_doc/mpnet>` (from Microsoft Research) released with the paper `MPNet: Masked and Permuted
+44. :doc:`MPNet <model_doc/mpnet>` (from Microsoft Research) released with the paper `MPNet: Masked and Permuted
     Pre-training for Language Understanding <https://arxiv.org/abs/2004.09297>`__ by Kaitao Song, Xu Tan, Tao Qin,
     Jianfeng Lu, Tie-Yan Liu.
-44. :doc:`MT5 <model_doc/mt5>` (from Google AI) released with the paper `mT5: A massively multilingual pre-trained
+45. :doc:`MT5 <model_doc/mt5>` (from Google AI) released with the paper `mT5: A massively multilingual pre-trained
     text-to-text transformer <https://arxiv.org/abs/2010.11934>`__ by Linting Xue, Noah Constant, Adam Roberts, Mihir
     Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
-45. :doc:`Pegasus <model_doc/pegasus>` (from Google) released with the paper `PEGASUS: Pre-training with Extracted
+46. :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.
-46. :doc:`ProphetNet <model_doc/prophetnet>` (from Microsoft Research) released with the paper `ProphetNet: Predicting
+47. :doc:`ProphetNet <model_doc/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.
-47. :doc:`Reformer <model_doc/reformer>` (from Google Research) released with the paper `Reformer: The Efficient
+48. :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.
-48. :doc:`RoBERTa <model_doc/roberta>` (from Facebook), released together with the paper a `Robustly Optimized BERT
+49. :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.
-49. :doc:`RoFormer <model_doc/roformer>` (from ZhuiyiTechnology), released together with the paper a `RoFormer:
+50. :doc:`RoFormer <model_doc/roformer>` (from ZhuiyiTechnology), released together with the paper a `RoFormer:
     Enhanced Transformer with Rotary Position Embedding <https://arxiv.org/pdf/2104.09864v1.pdf>`__ by Jianlin Su and
     Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
-50. :doc:`SpeechToTextTransformer <model_doc/speech_to_text>` (from Facebook), released together with the paper
+51. :doc:`SpeechToTextTransformer <model_doc/speech_to_text>` (from Facebook), released together with the paper
     `fairseq S2T: Fast Speech-to-Text Modeling with fairseq <https://arxiv.org/abs/2010.05171>`__ by Changhan Wang, Yun
     Tang, Xutai Ma, Anne Wu, Dmytro Okhonko, Juan Pino.
-51. :doc:`SqueezeBert <model_doc/squeezebert>` released with the paper `SqueezeBERT: What can computer vision teach NLP
+52. :doc:`SqueezeBert <model_doc/squeezebert>` 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.
-52. :doc:`T5 <model_doc/t5>` (from Google AI) released with the paper `Exploring the Limits of Transfer Learning with a
+53. :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.
-53. :doc:`TAPAS <model_doc/tapas>` (from Google AI) released with the paper `TAPAS: Weakly Supervised Table Parsing via
+54. :doc:`TAPAS <model_doc/tapas>` (from Google AI) released with the paper `TAPAS: Weakly Supervised Table Parsing via
     Pre-training <https://arxiv.org/abs/2004.02349>`__ by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller,
     Francesco Piccinno and Julian Martin Eisenschlos.
-54. :doc:`Transformer-XL <model_doc/transformerxl>` (from Google/CMU) released with the paper `Transformer-XL:
+55. :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.
-55. :doc:`Vision Transformer (ViT) <model_doc/vit>` (from Google AI) released with the paper `An Image is Worth 16x16
+56. :doc:`Vision Transformer (ViT) <model_doc/vit>` (from Google AI) released with the paper `An Image is Worth 16x16
     Words: Transformers for Image Recognition at Scale <https://arxiv.org/abs/2010.11929>`__ by Alexey Dosovitskiy,
     Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias
     Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
-56. :doc:`VisualBERT <model_doc/visual_bert>` (from UCLA NLP) released with the paper `VisualBERT: A Simple and
+57. :doc:`VisualBERT <model_doc/visual_bert>` (from UCLA NLP) released with the paper `VisualBERT: A Simple and
     Performant Baseline for Vision and Language <https://arxiv.org/pdf/1908.03557>`__ by Liunian Harold Li, Mark
     Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
-57. :doc:`Wav2Vec2 <model_doc/wav2vec2>` (from Facebook AI) released with the paper `wav2vec 2.0: A Framework for
+58. :doc:`Wav2Vec2 <model_doc/wav2vec2>` (from Facebook AI) released with the paper `wav2vec 2.0: A Framework for
     Self-Supervised Learning of Speech Representations <https://arxiv.org/abs/2006.11477>`__ by Alexei Baevski, Henry
     Zhou, Abdelrahman Mohamed, Michael Auli.
-58. :doc:`XLM <model_doc/xlm>` (from Facebook) released together with the paper `Cross-lingual Language Model
+59. :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.
-59. :doc:`XLM-ProphetNet <model_doc/xlmprophetnet>` (from Microsoft Research) released with the paper `ProphetNet:
+60. :doc:`XLM-ProphetNet <model_doc/xlmprophetnet>` (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.
-60. :doc:`XLM-RoBERTa <model_doc/xlmroberta>` (from Facebook AI), released together with the paper `Unsupervised
+61. :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.
-61. :doc:`XLNet <model_doc/xlnet>` (from Google/CMU) released with the paper `​XLNet: Generalized Autoregressive
+62. :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.
-62. :doc:`XLSR-Wav2Vec2 <model_doc/xlsr_wav2vec2>` (from Facebook AI) released with the paper `Unsupervised
+63. :doc:`XLSR-Wav2Vec2 <model_doc/xlsr_wav2vec2>` (from Facebook AI) released with the paper `Unsupervised
     Cross-Lingual Representation Learning For Speech Recognition <https://arxiv.org/abs/2006.13979>`__ by Alexis
     Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
 
@@ -324,6 +327,8 @@ Flax), PyTorch, and/or TensorFlow.
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |          CamemBERT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
+|           Canine            |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
++-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |          ConvBERT           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |            DETR             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
@@ -508,6 +513,7 @@ Flax), PyTorch, and/or TensorFlow.
     model_doc/bort
     model_doc/byt5
     model_doc/camembert
+    model_doc/canine
     model_doc/clip
     model_doc/convbert
     model_doc/cpm

docs/source/model_doc/canine.rst

+.. 
+    Copyright 2021 The HuggingFace Team. All rights reserved.
+
+    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.
+
+CANINE
+-----------------------------------------------------------------------------------------------------------------------
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The CANINE model was proposed in `CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language
+Representation <https://arxiv.org/abs/2103.06874>`__ by Jonathan H. Clark, Dan Garrette, Iulia Turc, John Wieting. It's
+among the first papers that trains a Transformer without using an explicit tokenization step (such as Byte Pair
+Encoding (BPE), WordPiece or SentencePiece). Instead, the model is trained directly at a Unicode character-level.
+Training at a character-level inevitably comes with a longer sequence length, which CANINE solves with an efficient
+downsampling strategy, before applying a deep Transformer encoder.
+
+The abstract from the paper is the following:
+
+*Pipelined NLP systems have largely been superseded by end-to-end neural modeling, yet nearly all commonly-used models
+still require an explicit tokenization step. While recent tokenization approaches based on data-derived subword
+lexicons are less brittle than manually engineered tokenizers, these techniques are not equally suited to all
+languages, and the use of any fixed vocabulary may limit a model's ability to adapt. In this paper, we present CANINE,
+a neural encoder that operates directly on character sequences, without explicit tokenization or vocabulary, and a
+pre-training strategy that operates either directly on characters or optionally uses subwords as a soft inductive bias.
+To use its finer-grained input effectively and efficiently, CANINE combines downsampling, which reduces the input
+sequence length, with a deep transformer stack, which encodes context. CANINE outperforms a comparable mBERT model by
+2.8 F1 on TyDi QA, a challenging multilingual benchmark, despite having 28% fewer model parameters.*
+
+Tips:
+
+- CANINE uses no less than 3 Transformer encoders internally: 2 "shallow" encoders (which only consist of a single
+  layer) and 1 "deep" encoder (which is a regular BERT encoder). First, a "shallow" encoder is used to contextualize
+  the character embeddings, using local attention. Next, after downsampling, a "deep" encoder is applied. Finally,
+  after upsampling, a "shallow" encoder is used to create the final character embeddings. Details regarding up- and
+  downsampling can be found in the paper.
+- CANINE uses a max sequence length of 2048 characters by default. One can use :class:`~transformers.CanineTokenizer`
+  to prepare text for the model.
+- Classification can be done by placing a linear layer on top of the final hidden state of the special [CLS] token
+  (which has a predefined Unicode code point). For token classification tasks however, the downsampled sequence of
+  tokens needs to be upsampled again to match the length of the original character sequence (which is 2048). The
+  details for this can be found in the paper.
+
+This model was contributed by `nielsr <https://huggingface.co/nielsr>`__. The original code can be found `here
+<https://github.com/google-research/language/tree/master/language/canine>`__.
+
+
+Example
+_______________________________________________________________________________________________________________________
+
+CANINE works on raw characters, so it can be used without a tokenizer:
+
+.. code-block::
+
+    from transformers import CanineModel
+    import torch
+
+    model = CanineModel.from_pretrained('google/canine-s') # model pre-trained with autoregressive character loss
+
+    text = "hello world"
+    # use Python's built-in ord() function to turn each character into its unicode code point id
+    input_ids = torch.tensor([[ord(char) for char in text]])
+
+    outputs = model(input_ids) # forward pass
+    pooled_output = outputs.pooler_output
+    sequence_output = outputs.last_hidden_state
+
+
+For batched inference and training, it is however recommended to make use of the tokenizer (to pad/truncate all
+sequences to the same length):
+
+.. code-block::
+
+    from transformers import CanineTokenizer, CanineModel
+
+    model = CanineModel.from_pretrained('google/canine-s')
+    tokenizer = CanineTokenizer.from_pretrained('google/canine-s')
+
+    inputs = ["Life is like a box of chocolates.", "You never know what you gonna get."]
+    encoding = tokenizer(inputs, padding="longest", truncation=True, return_tensors="pt")
+
+    outputs = model(**encoding) # forward pass
+    pooled_output = outputs.pooler_output
+    sequence_output = outputs.last_hidden_state
+
+
+CANINE specific outputs
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.models.canine.modeling_canine.CanineModelOutputWithPooling
+    :members:
+
+
+CanineConfig
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CanineConfig
+    :members:
+
+
+CanineTokenizer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CanineTokenizer
+    :members: build_inputs_with_special_tokens, get_special_tokens_mask,
+        create_token_type_ids_from_sequences
+
+
+CanineModel
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CanineModel
+    :members: forward
+
+
+CanineForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CanineForSequenceClassification
+    :members: forward
+
+
+CanineForMultipleChoice
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CanineForMultipleChoice
+    :members: forward
+
+
+CanineForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CanineForTokenClassification
+    :members: forward
+
+
+CanineForQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.CanineForQuestionAnswering
+    :members: forward

src/transformers/__init__.py

@@ -170,6 +170,7 @@ _import_structure = {
     ],
     "models.byt5": ["ByT5Tokenizer"],
     "models.camembert": ["CAMEMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP", "CamembertConfig"],
+    "models.canine": ["CANINE_PRETRAINED_CONFIG_ARCHIVE_MAP", "CanineConfig", "CanineTokenizer"],
     "models.clip": [
         "CLIP_PRETRAINED_CONFIG_ARCHIVE_MAP",
         "CLIPConfig",
@@ -505,7 +506,6 @@ if is_torch_available():
             "load_tf_weights_in_albert",
         ]
     )
-
     _import_structure["models.auto"].extend(
         [
             "MODEL_FOR_CAUSAL_LM_MAPPING",
@@ -632,6 +632,19 @@ if is_torch_available():
             "CamembertModel",
         ]
     )
+    _import_structure["models.canine"].extend(
+        [
+            "CANINE_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "CanineForMultipleChoice",
+            "CanineForQuestionAnswering",
+            "CanineForSequenceClassification",
+            "CanineForTokenClassification",
+            "CanineLayer",
+            "CanineModel",
+            "CaninePreTrainedModel",
+            "load_tf_weights_in_canine",
+        ]
+    )
     _import_structure["models.clip"].extend(
         [
             "CLIP_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -1756,6 +1769,7 @@ if TYPE_CHECKING:
     )
     from .models.byt5 import ByT5Tokenizer
     from .models.camembert import CAMEMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, CamembertConfig
+    from .models.canine import CANINE_PRETRAINED_CONFIG_ARCHIVE_MAP, CanineConfig, CanineTokenizer
     from .models.clip import (
         CLIP_PRETRAINED_CONFIG_ARCHIVE_MAP,
         CLIPConfig,
@@ -2156,6 +2170,17 @@ if TYPE_CHECKING:
             CamembertForTokenClassification,
             CamembertModel,
         )
+        from .models.canine import (
+            CANINE_PRETRAINED_MODEL_ARCHIVE_LIST,
+            CanineForMultipleChoice,
+            CanineForQuestionAnswering,
+            CanineForSequenceClassification,
+            CanineForTokenClassification,
+            CanineLayer,
+            CanineModel,
+            CaninePreTrainedModel,
+            load_tf_weights_in_canine,
+        )
         from .models.clip import (
             CLIP_PRETRAINED_MODEL_ARCHIVE_LIST,
             CLIPModel,

src/transformers/models/__init__.py

@@ -30,6 +30,7 @@ from . import (
     blenderbot,
     blenderbot_small,
     camembert,
+    canine,
     clip,
     convbert,
     cpm,

src/transformers/models/auto/configuration_auto.py

@@ -33,6 +33,7 @@ from ..blenderbot_small.configuration_blenderbot_small import (
     BlenderbotSmallConfig,
 )
 from ..camembert.configuration_camembert import CAMEMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, CamembertConfig
+from ..canine.configuration_canine import CANINE_PRETRAINED_CONFIG_ARCHIVE_MAP, CanineConfig
 from ..clip.configuration_clip import CLIP_PRETRAINED_CONFIG_ARCHIVE_MAP, CLIPConfig
 from ..convbert.configuration_convbert import CONVBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, ConvBertConfig
 from ..ctrl.configuration_ctrl import CTRL_PRETRAINED_CONFIG_ARCHIVE_MAP, CTRLConfig
@@ -96,6 +97,7 @@ ALL_PRETRAINED_CONFIG_ARCHIVE_MAP = dict(
     for pretrained_map in [
         # Add archive maps here
         VISUAL_BERT_PRETRAINED_CONFIG_ARCHIVE_MAP,
+        CANINE_PRETRAINED_CONFIG_ARCHIVE_MAP,
         ROFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP,
         CLIP_PRETRAINED_CONFIG_ARCHIVE_MAP,
         BIGBIRD_PEGASUS_PRETRAINED_CONFIG_ARCHIVE_MAP,
@@ -155,6 +157,7 @@ CONFIG_MAPPING = OrderedDict(
     [
         # Add configs here
         ("visual_bert", VisualBertConfig),
+        ("canine", CanineConfig),
         ("roformer", RoFormerConfig),
         ("clip", CLIPConfig),
         ("bigbird_pegasus", BigBirdPegasusConfig),
@@ -220,6 +223,7 @@ MODEL_NAMES_MAPPING = OrderedDict(
     [
         # Add full (and cased) model names here
         ("visual_bert", "VisualBert"),
+        ("canine", "Canine"),
         ("roformer", "RoFormer"),
         ("clip", "CLIP"),
         ("bigbird_pegasus", "BigBirdPegasus"),

src/transformers/models/auto/modeling_auto.py

@@ -81,6 +81,13 @@ from ..camembert.modeling_camembert import (
     CamembertForTokenClassification,
     CamembertModel,
 )
+from ..canine.modeling_canine import (
+    CanineForMultipleChoice,
+    CanineForQuestionAnswering,
+    CanineForSequenceClassification,
+    CanineForTokenClassification,
+    CanineModel,
+)
 from ..clip.modeling_clip import CLIPModel
 from ..convbert.modeling_convbert import (
     ConvBertForMaskedLM,
@@ -312,6 +319,7 @@ from .configuration_auto import (
     BlenderbotConfig,
     BlenderbotSmallConfig,
     CamembertConfig,
+    CanineConfig,
     CLIPConfig,
     ConvBertConfig,
     CTRLConfig,
@@ -371,6 +379,7 @@ MODEL_MAPPING = OrderedDict(
     [
         # Base model mapping
         (VisualBertConfig, VisualBertModel),
+        (CanineConfig, CanineModel),
         (RoFormerConfig, RoFormerModel),
         (CLIPConfig, CLIPModel),
         (BigBirdPegasusConfig, BigBirdPegasusModel),
@@ -624,6 +633,7 @@ MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING = OrderedDict(
 MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING = OrderedDict(
     [
         # Model for Sequence Classification mapping
+        (CanineConfig, CanineForSequenceClassification),
         (RoFormerConfig, RoFormerForSequenceClassification),
         (BigBirdPegasusConfig, BigBirdPegasusForSequenceClassification),
         (BigBirdConfig, BigBirdForSequenceClassification),
@@ -664,6 +674,7 @@ MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING = OrderedDict(
 MODEL_FOR_QUESTION_ANSWERING_MAPPING = OrderedDict(
     [
         # Model for Question Answering mapping
+        (CanineConfig, CanineForQuestionAnswering),
         (RoFormerConfig, RoFormerForQuestionAnswering),
         (BigBirdPegasusConfig, BigBirdPegasusForQuestionAnswering),
         (BigBirdConfig, BigBirdForQuestionAnswering),
@@ -705,6 +716,7 @@ MODEL_FOR_TABLE_QUESTION_ANSWERING_MAPPING = OrderedDict(
 MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING = OrderedDict(
     [
         # Model for Token Classification mapping
+        (CanineConfig, CanineForTokenClassification),
         (RoFormerConfig, RoFormerForTokenClassification),
         (BigBirdConfig, BigBirdForTokenClassification),
         (ConvBertConfig, ConvBertForTokenClassification),
@@ -735,6 +747,7 @@ MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING = OrderedDict(
 MODEL_FOR_MULTIPLE_CHOICE_MAPPING = OrderedDict(
     [
         # Model for Multiple Choice mapping
+        (CanineConfig, CanineForMultipleChoice),
         (RoFormerConfig, RoFormerForMultipleChoice),
         (BigBirdConfig, BigBirdForMultipleChoice),
         (ConvBertConfig, ConvBertForMultipleChoice),

src/transformers/models/auto/tokenization_auto.py

@@ -37,6 +37,7 @@ from ..bertweet.tokenization_bertweet import BertweetTokenizer
 from ..blenderbot.tokenization_blenderbot import BlenderbotTokenizer
 from ..blenderbot_small.tokenization_blenderbot_small import BlenderbotSmallTokenizer
 from ..byt5.tokenization_byt5 import ByT5Tokenizer
+from ..canine.tokenization_canine import CanineTokenizer
 from ..convbert.tokenization_convbert import ConvBertTokenizer
 from ..ctrl.tokenization_ctrl import CTRLTokenizer
 from ..deberta.tokenization_deberta import DebertaTokenizer
@@ -78,6 +79,7 @@ from .configuration_auto import (
     BlenderbotConfig,
     BlenderbotSmallConfig,
     CamembertConfig,
+    CanineConfig,
     ConvBertConfig,
     CTRLConfig,
     DebertaConfig,
@@ -294,6 +296,7 @@ TOKENIZER_MAPPING = OrderedDict(
         (GPTNeoConfig, (GPT2Tokenizer, GPT2TokenizerFast)),
         (LukeConfig, (LukeTokenizer, None)),
         (BigBirdPegasusConfig, (PegasusTokenizer, PegasusTokenizerFast)),
+        (CanineConfig, (CanineTokenizer, None)),
     ]
 )
 

src/transformers/models/canine/__init__.py

+# flake8: noqa
+# There's no way to ignore "F401 '...' imported but unused" warnings in this
+# module, but to preserve other warnings. So, don't check this module at all.
+
+# Copyright 2020 The HuggingFace Team. All rights reserved.
+#
+# 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.
+from typing import TYPE_CHECKING
+
+from ...file_utils import _BaseLazyModule, is_tokenizers_available, is_torch_available
+
+
+_import_structure = {
+    "configuration_canine": ["CANINE_PRETRAINED_CONFIG_ARCHIVE_MAP", "CanineConfig"],
+    "tokenization_canine": ["CanineTokenizer"],
+}
+
+if is_torch_available():
+    _import_structure["modeling_canine"] = [
+        "CANINE_PRETRAINED_MODEL_ARCHIVE_LIST",
+        "CanineForMultipleChoice",
+        "CanineForQuestionAnswering",
+        "CanineForSequenceClassification",
+        "CanineForTokenClassification",
+        "CanineLayer",
+        "CanineModel",
+        "CaninePreTrainedModel",
+        "load_tf_weights_in_canine",
+    ]
+
+
+if TYPE_CHECKING:
+    from .configuration_canine import CANINE_PRETRAINED_CONFIG_ARCHIVE_MAP, CanineConfig
+    from .tokenization_canine import CanineTokenizer
+
+    if is_torch_available():
+        from .modeling_canine import (
+            CANINE_PRETRAINED_MODEL_ARCHIVE_LIST,
+            CanineForMultipleChoice,
+            CanineForQuestionAnswering,
+            CanineForSequenceClassification,
+            CanineForTokenClassification,
+            CanineLayer,
+            CanineModel,
+            CaninePreTrainedModel,
+            load_tf_weights_in_canine,
+        )
+
+
+else:
+    import importlib
+    import os
+    import sys
+
+    class _LazyModule(_BaseLazyModule):
+        """
+        Module class that surfaces all objects but only performs associated imports when the objects are requested.
+        """
+
+        __file__ = globals()["__file__"]
+        __path__ = [os.path.dirname(__file__)]
+
+        def _get_module(self, module_name: str):
+            return importlib.import_module("." + module_name, self.__name__)
+
+    sys.modules[__name__] = _LazyModule(__name__, _import_structure)

src/transformers/models/canine/configuration_canine.py

+# coding=utf-8
+# Copyright Google AI and The HuggingFace Inc. team. All rights reserved.
+#
+# 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.
+""" CANINE model configuration """
+
+from ...configuration_utils import PretrainedConfig
+from ...utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+CANINE_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "google/canine-s": "https://huggingface.co/google/canine-s/resolve/main/config.json",
+    # See all CANINE models at https://huggingface.co/models?filter=canine
+}
+
+
+class CanineConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a :class:`~transformers.CanineModel`. It is used to
+    instantiate an CANINE model according to the specified arguments, defining the model architecture. Instantiating a
+    configuration with the defaults will yield a similar configuration to that of the CANINE `google/canine-s
+    <https://huggingface.co/google/canine-s>`__ architecture.
+
+    Configuration objects inherit from :class:`~transformers.PretrainedConfig` and can be used to control the model
+    outputs. Read the documentation from :class:`~transformers.PretrainedConfig` for more information.
+
+
+    Args:
+        hidden_size (:obj:`int`, `optional`, defaults to 768):
+            Dimension of the encoder layers and the pooler layer.
+        num_hidden_layers (:obj:`int`, `optional`, defaults to 12):
+            Number of hidden layers in the deep Transformer encoder.
+        num_attention_heads (:obj:`int`, `optional`, defaults to 12):
+            Number of attention heads for each attention layer in the Transformer encoders.
+        intermediate_size (:obj:`int`, `optional`, defaults to 3072):
+            Dimension of the "intermediate" (i.e., feed-forward) layer in the Transformer encoders.
+        hidden_act (:obj:`str` or :obj:`function`, `optional`, defaults to :obj:`"gelu"`):
+            The non-linear activation function (function or string) in the encoder and pooler. If string,
+            :obj:`"gelu"`, :obj:`"relu"`, :obj:`"selu"` and :obj:`"gelu_new"` are supported.
+        hidden_dropout_prob (:obj:`float`, `optional`, defaults to 0.1):
+            The dropout probabilitiy for all fully connected layers in the embeddings, encoders, and pooler.
+        attention_probs_dropout_prob (:obj:`float`, `optional`, defaults to 0.1):
+            The dropout ratio for the attention probabilities.
+        max_position_embeddings (:obj:`int`, `optional`, defaults to 16384):
+            The maximum sequence length that this model might ever be used with.
+        type_vocab_size (:obj:`int`, `optional`, defaults to 16):
+            The vocabulary size of the :obj:`token_type_ids` passed when calling :class:`~transformers.CanineModel`.
+        initializer_range (:obj:`float`, `optional`, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        layer_norm_eps (:obj:`float`, `optional`, defaults to 1e-12):
+            The epsilon used by the layer normalization layers.
+        gradient_checkpointing (:obj:`bool`, `optional`, defaults to :obj:`False`):
+            If :obj:`True`, use gradient checkpointing to save memory at the expense of slower backward pass.
+        downsampling_rate (:obj:`int`, `optional`, defaults to 4):
+            The rate at which to downsample the original character sequence length before applying the deep Transformer
+            encoder.
+        upsampling_kernel_size (:obj:`int`, `optional`, defaults to 4):
+            The kernel size (i.e. the number of characters in each window) of the convolutional projection layer when
+            projecting back from :obj:`hidden_size`*2 to :obj:`hidden_size`.
+        num_hash_functions (:obj:`int`, `optional`, defaults to 8):
+            The number of hash functions to use. Each hash function has its own embedding matrix.
+        num_hash_buckets (:obj:`int`, `optional`, defaults to 16384):
+            The number of hash buckets to use.
+        local_transformer_stride (:obj:`int`, `optional`, defaults to 128):
+            The stride of the local attention of the first shallow Transformer encoder. Defaults to 128 for good
+            TPU/XLA memory alignment.
+
+    Example::
+
+        >>> from transformers import CanineModel, CanineConfig
+
+        >>> # Initializing a CANINE google/canine-s style configuration
+        >>> configuration = CanineConfig()
+
+        >>> # Initializing a model from the google/canine-s style configuration
+        >>> model = CanineModel(configuration)
+
+        >>> # Accessing the model configuration
+        >>> configuration = model.config
+    """
+    model_type = "canine"
+
+    def __init__(
+        self,
+        hidden_size=768,
+        num_hidden_layers=12,
+        num_attention_heads=12,
+        intermediate_size=3072,
+        hidden_act="gelu",
+        hidden_dropout_prob=0.1,
+        attention_probs_dropout_prob=0.1,
+        max_position_embeddings=16384,
+        type_vocab_size=16,
+        initializer_range=0.02,
+        layer_norm_eps=1e-12,
+        use_cache=True,
+        is_encoder_decoder=False,
+        pad_token_id=0,
+        bos_token_id=0xE000,
+        eos_token_id=0xE001,
+        downsampling_rate=4,
+        upsampling_kernel_size=4,
+        num_hash_functions=8,
+        num_hash_buckets=16384,
+        local_transformer_stride=128,  # Good TPU/XLA memory alignment.
+        **kwargs
+    ):
+        super().__init__(pad_token_id=pad_token_id, bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs)
+
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.intermediate_size = intermediate_size
+        self.hidden_act = hidden_act
+        self.hidden_dropout_prob = hidden_dropout_prob
+        self.attention_probs_dropout_prob = attention_probs_dropout_prob
+        self.initializer_range = initializer_range
+        self.type_vocab_size = type_vocab_size
+        self.layer_norm_eps = layer_norm_eps
+        self.use_cache = use_cache
+
+        # Character config:
+        self.downsampling_rate = downsampling_rate
+        self.upsampling_kernel_size = upsampling_kernel_size
+        self.num_hash_functions = num_hash_functions
+        self.num_hash_buckets = num_hash_buckets
+        self.local_transformer_stride = local_transformer_stride

src/transformers/models/canine/convert_canine_original_tf_checkpoint_to_pytorch.py

+# coding=utf-8
+# Copyright 2021 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 CANINE checkpoint."""
+
+
+import argparse
+
+from transformers import CanineConfig, CanineModel, CanineTokenizer, load_tf_weights_in_canine
+from transformers.utils import logging
+
+
+logging.set_verbosity_info()
+
+
+def convert_tf_checkpoint_to_pytorch(tf_checkpoint_path, pytorch_dump_path):
+
+    # Initialize PyTorch model
+    config = CanineConfig()
+    model = CanineModel(config)
+    model.eval()
+
+    print(f"Building PyTorch model from configuration: {config}")
+
+    # Load weights from tf checkpoint
+    load_tf_weights_in_canine(model, config, tf_checkpoint_path)
+
+    # Save pytorch-model (weights and configuration)
+    print(f"Save PyTorch model to {pytorch_dump_path}")
+    model.save_pretrained(pytorch_dump_path)
+
+    # Save tokenizer files
+    tokenizer = CanineTokenizer()
+    print(f"Save tokenizer files to {pytorch_dump_path}")
+    tokenizer.save_pretrained(pytorch_dump_path)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    # Required parameters
+    parser.add_argument(
+        "--tf_checkpoint_path",
+        default=None,
+        type=str,
+        required=True,
+        help="Path to the TensorFlow checkpoint. Should end with model.ckpt",
+    )
+    parser.add_argument(
+        "--pytorch_dump_path",
+        default=None,
+        type=str,
+        required=True,
+        help="Path to a folder where the PyTorch model will be placed.",
+    )
+    args = parser.parse_args()
+    convert_tf_checkpoint_to_pytorch(args.tf_checkpoint_path, args.pytorch_dump_path)

src/transformers/models/canine/modeling_canine.py

+# coding=utf-8
+# Copyright 2021 Google AI The HuggingFace Inc. team. All rights reserved.
+#
+# 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.
+""" PyTorch CANINE model. """
+
+
+import copy
+import math
+import os
+from dataclasses import dataclass
+from typing import Optional, Tuple
+
+import torch
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
+
+from ...activations import ACT2FN
+from ...file_utils import add_code_sample_docstrings, add_start_docstrings, add_start_docstrings_to_model_forward
+from ...modeling_outputs import (
+    BaseModelOutput,
+    ModelOutput,
+    MultipleChoiceModelOutput,
+    QuestionAnsweringModelOutput,
+    SequenceClassifierOutput,
+    TokenClassifierOutput,
+)
+from ...modeling_utils import (
+    PreTrainedModel,
+    apply_chunking_to_forward,
+    find_pruneable_heads_and_indices,
+    prune_linear_layer,
+)
+from ...utils import logging
+from .configuration_canine import CanineConfig
+
+
+logger = logging.get_logger(__name__)
+
+_CHECKPOINT_FOR_DOC = "google/canine-s"
+_CONFIG_FOR_DOC = "CanineConfig"
+_TOKENIZER_FOR_DOC = "CanineTokenizer"
+
+CANINE_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "google/canine-s",
+    "google/canine-r"
+    # See all CANINE models at https://huggingface.co/models?filter=canine
+]
+
+# Support up to 16 hash functions.
+_PRIMES = [31, 43, 59, 61, 73, 97, 103, 113, 137, 149, 157, 173, 181, 193, 211, 223]
+
+
+@dataclass
+class CanineModelOutputWithPooling(ModelOutput):
+    """
+    Output type of :class:`~transformers.CanineModel`. Based on
+    :class:`~transformers.modeling_outputs.BaseModelOutputWithPooling`, but with slightly different
+    :obj:`hidden_states` and :obj:`attentions`, as these also include the hidden states and attentions of the shallow
+    Transformer encoders.
+
+    Args:
+        last_hidden_state (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`):
+            Sequence of hidden-states at the output of the last layer of the model (i.e. the output of the final
+            shallow Transformer encoder).
+        pooler_output (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, hidden_size)`):
+            Hidden-state of the first token of the sequence (classification token) at the last layer of the deep
+            Transformer encoder, further processed by a Linear layer and a Tanh activation function. The Linear layer
+            weights are trained from the next sentence prediction (classification) objective during pretraining.
+        hidden_states (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``output_hidden_states=True`` is passed or when ``config.output_hidden_states=True``):
+            Tuple of :obj:`torch.FloatTensor` (one for the input to each encoder + one for the output of each layer of
+            each encoder) of shape :obj:`(batch_size, sequence_length, hidden_size)` and :obj:`(batch_size,
+            sequence_length // config.downsampling_rate, hidden_size)`. Hidden-states of the model at the output of
+            each layer plus the initial input to each Transformer encoder. The hidden states of the shallow encoders
+            have length :obj:`sequence_length`, but the hidden states of the deep encoder have length
+            :obj:`sequence_length` // :obj:`config.downsampling_rate`.
+        attentions (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``output_attentions=True`` is passed or when ``config.output_attentions=True``):
+            Tuple of :obj:`torch.FloatTensor` (one for each layer) of the 3 Transformer encoders of shape
+            :obj:`(batch_size, num_heads, sequence_length, sequence_length)` and :obj:`(batch_size, num_heads,
+            sequence_length // config.downsampling_rate, sequence_length // config.downsampling_rate)`. Attentions
+            weights after the attention softmax, used to compute the weighted average in the self-attention heads.
+    """
+
+    last_hidden_state: torch.FloatTensor = None
+    pooler_output: torch.FloatTensor = None
+    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
+    attentions: Optional[Tuple[torch.FloatTensor]] = None
+
+
+def load_tf_weights_in_canine(model, config, tf_checkpoint_path):
+    """Load tf checkpoints in a pytorch model."""
+    try:
+        import re
+
+        import numpy as np
+        import tensorflow as tf
+    except ImportError:
+        logger.error(
+            "Loading a TensorFlow model in PyTorch, requires TensorFlow to be installed. Please see "
+            "https://www.tensorflow.org/install/ for installation instructions."
+        )
+        raise
+    tf_path = os.path.abspath(tf_checkpoint_path)
+    logger.info(f"Converting TensorFlow checkpoint from {tf_path}")
+    # Load weights from TF model
+    init_vars = tf.train.list_variables(tf_path)
+    names = []
+    arrays = []
+    for name, shape in init_vars:
+        logger.info(f"Loading TF weight {name} with shape {shape}")
+        array = tf.train.load_variable(tf_path, name)
+        names.append(name)
+        arrays.append(array)
+
+    for name, array in zip(names, arrays):
+        name = name.split("/")
+        # adam_v and adam_m are variables used in AdamWeightDecayOptimizer to calculated m and v
+        # which are not required for using pretrained model
+        # also discard the cls weights (which were used for the next sentence prediction pre-training task)
+        if any(
+            n
+            in [
+                "adam_v",
+                "adam_m",
+                "AdamWeightDecayOptimizer",
+                "AdamWeightDecayOptimizer_1",
+                "global_step",
+                "cls",
+                "autoregressive_decoder",
+                "char_output_weights",
+            ]
+            for n in name
+        ):
+            logger.info(f"Skipping {'/'.join(name)}")
+            continue
+        # if first scope name starts with "bert", change it to "encoder"
+        if name[0] == "bert":
+            name[0] = "encoder"
+        # remove "embeddings" middle name of HashBucketCodepointEmbedders
+        elif name[1] == "embeddings":
+            name.remove(name[1])
+        # rename segment_embeddings to token_type_embeddings
+        elif name[1] == "segment_embeddings":
+            name[1] = "token_type_embeddings"
+        # rename initial convolutional projection layer
+        elif name[1] == "initial_char_encoder":
+            name = ["chars_to_molecules"] + name[-2:]
+        # rename final convolutional projection layer
+        elif name[0] == "final_char_encoder" and name[1] in ["LayerNorm", "conv"]:
+            name = ["projection"] + name[1:]
+        pointer = model
+        for m_name in name:
+            if (re.fullmatch(r"[A-Za-z]+_\d+", m_name)) and "Embedder" not in m_name:
+                scope_names = re.split(r"_(\d+)", m_name)
+            else:
+                scope_names = [m_name]
+            if scope_names[0] == "kernel" or scope_names[0] == "gamma":
+                pointer = getattr(pointer, "weight")
+            elif scope_names[0] == "output_bias" or scope_names[0] == "beta":
+                pointer = getattr(pointer, "bias")
+            elif scope_names[0] == "output_weights":
+                pointer = getattr(pointer, "weight")
+            else:
+                try:
+                    pointer = getattr(pointer, scope_names[0])
+                except AttributeError:
+                    logger.info(f"Skipping {'/'.join(name)}")
+                    continue
+            if len(scope_names) >= 2:
+                num = int(scope_names[1])
+                pointer = pointer[num]
+        if m_name[-11:] == "_embeddings":
+            pointer = getattr(pointer, "weight")
+        elif m_name[-10:] in [f"Embedder_{i}" for i in range(8)]:
+            pointer = getattr(pointer, "weight")
+        elif m_name == "kernel":
+            array = np.transpose(array)
+        try:
+            assert (
+                pointer.shape == array.shape
+            ), f"Pointer shape {pointer.shape} and array shape {array.shape} mismatched"
+        except AssertionError as e:
+            e.args += (pointer.shape, array.shape)
+            raise
+        logger.info(f"Initialize PyTorch weight {name}")
+        pointer.data = torch.from_numpy(array)
+    return model
+
+
+class CanineEmbeddings(nn.Module):
+    """Construct the character, position and token_type embeddings."""
+
+    def __init__(self, config):
+        super().__init__()
+
+        self.config = config
+
+        # character embeddings
+        shard_embedding_size = config.hidden_size // config.num_hash_functions
+        for i in range(config.num_hash_functions):
+            name = f"HashBucketCodepointEmbedder_{i}"
+            setattr(self, name, nn.Embedding(config.num_hash_buckets, shard_embedding_size))
+        self.char_position_embeddings = nn.Embedding(config.num_hash_buckets, config.hidden_size)
+        self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.hidden_size)
+
+        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
+        # any TensorFlow checkpoint file
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+        # position_ids (1, len position emb) is contiguous in memory and exported when serialized
+        self.register_buffer("position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)))
+        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
+
+    def _hash_bucket_tensors(self, input_ids, num_hashes: int, num_buckets: int):
+        """
+        Converts ids to hash bucket ids via multiple hashing.
+
+        Args:
+            input_ids: The codepoints or other IDs to be hashed.
+            num_hashes: The number of hash functions to use.
+            num_buckets: The number of hash buckets (i.e. embeddings in each table).
+
+        Returns:
+            A list of tensors, each of which is the hash bucket IDs from one hash function.
+        """
+        if num_hashes > len(_PRIMES):
+            raise ValueError(f"`num_hashes` must be <= {len(_PRIMES)}")
+
+        primes = _PRIMES[:num_hashes]
+
+        result_tensors = []
+        for prime in primes:
+            hashed = ((input_ids + 1) * prime) % num_buckets
+            result_tensors.append(hashed)
+        return result_tensors
+
+    def _embed_hash_buckets(self, input_ids, embedding_size: int, num_hashes: int, num_buckets: int):
+        """Converts IDs (e.g. codepoints) into embeddings via multiple hashing."""
+        if embedding_size % num_hashes != 0:
+            raise ValueError(f"Expected `embedding_size` ({embedding_size}) % `num_hashes` ({num_hashes}) == 0")
+
+        hash_bucket_tensors = self._hash_bucket_tensors(input_ids, num_hashes=num_hashes, num_buckets=num_buckets)
+        embedding_shards = []
+        for i, hash_bucket_ids in enumerate(hash_bucket_tensors):
+            name = f"HashBucketCodepointEmbedder_{i}"
+            shard_embeddings = getattr(self, name)(hash_bucket_ids)
+            embedding_shards.append(shard_embeddings)
+
+        return torch.cat(embedding_shards, dim=-1)
+
+    def forward(
+        self,
+        input_ids=None,
+        token_type_ids=None,
+        position_ids=None,
+        inputs_embeds=None,
+    ):
+        if input_ids is not None:
+            input_shape = input_ids.size()
+        else:
+            input_shape = inputs_embeds.size()[:-1]
+
+        seq_length = input_shape[1]
+
+        if position_ids is None:
+            position_ids = self.position_ids[:, :seq_length]
+
+        if token_type_ids is None:
+            token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=self.position_ids.device)
+
+        if inputs_embeds is None:
+            inputs_embeds = self._embed_hash_buckets(
+                input_ids, self.config.hidden_size, self.config.num_hash_functions, self.config.num_hash_buckets
+            )
+
+        token_type_embeddings = self.token_type_embeddings(token_type_ids)
+
+        embeddings = inputs_embeds + token_type_embeddings
+
+        if self.position_embedding_type == "absolute":
+            position_embeddings = self.char_position_embeddings(position_ids)
+            embeddings += position_embeddings
+        embeddings = self.LayerNorm(embeddings)
+        embeddings = self.dropout(embeddings)
+        return embeddings
+
+
+class CharactersToMolecules(nn.Module):
+    """Convert character sequence to initial molecule sequence (i.e. downsample) using strided convolutions."""
+
+    def __init__(self, config):
+        super().__init__()
+
+        self.conv = nn.Conv1d(
+            in_channels=config.hidden_size,
+            out_channels=config.hidden_size,
+            kernel_size=config.downsampling_rate,
+            stride=config.downsampling_rate,
+        )
+        self.activation = ACT2FN[config.hidden_act]
+
+        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
+        # any TensorFlow checkpoint file
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+
+    def forward(self, char_encoding: torch.Tensor) -> torch.Tensor:
+
+        # `cls_encoding`: [batch, 1, hidden_size]
+        cls_encoding = char_encoding[:, 0:1, :]
+
+        # char_encoding has shape [batch, char_seq, hidden_size]
+        # We transpose it to be [batch, hidden_size, char_seq]
+        char_encoding = torch.transpose(char_encoding, 1, 2)
+        downsampled = self.conv(char_encoding)
+        downsampled = torch.transpose(downsampled, 1, 2)
+        downsampled = self.activation(downsampled)
+
+        # Truncate the last molecule in order to reserve a position for [CLS].
+        # Often, the last position is never used (unless we completely fill the
+        # text buffer). This is important in order to maintain alignment on TPUs
+        # (i.e. a multiple of 128).
+        downsampled_truncated = downsampled[:, 0:-1, :]
+
+        # We also keep [CLS] as a separate sequence position since we always
+        # want to reserve a position (and the model capacity that goes along
+        # with that) in the deep BERT stack.
+        # `result`: [batch, molecule_seq, molecule_dim]
+        result = torch.cat([cls_encoding, downsampled_truncated], dim=1)
+
+        result = self.LayerNorm(result)
+
+        return result
+
+
+class ConvProjection(nn.Module):
+    """
+    Project representations from hidden_size*2 back to hidden_size across a window of w = config.upsampling_kernel_size
+    characters.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.conv = nn.Conv1d(
+            in_channels=config.hidden_size * 2,
+            out_channels=config.hidden_size,
+            kernel_size=config.upsampling_kernel_size,
+            stride=1,
+        )
+        self.activation = ACT2FN[config.hidden_act]
+        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be able to load
+        # any TensorFlow checkpoint file
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+    def forward(self, inputs, final_seq_char_positions=None):
+        # inputs has shape [batch, mol_seq, molecule_hidden_size+char_hidden_final]
+        # we transpose it to be [batch, molecule_hidden_size+char_hidden_final, mol_seq]
+        inputs = torch.transpose(inputs, 1, 2)
+
+        # PyTorch < 1.9 does not support padding="same" (which is used in the original implementation),
+        # so we pad the tensor manually before passing it to the conv layer
+        # based on https://github.com/google-research/big_transfer/blob/49afe42338b62af9fbe18f0258197a33ee578a6b/bit_tf2/models.py#L36-L38
+        pad_total = self.config.upsampling_kernel_size - 1
+        pad_beg = pad_total // 2
+        pad_end = pad_total - pad_beg
+
+        pad = nn.ConstantPad1d((pad_beg, pad_end), 0)
+        # `result`: shape (batch_size, char_seq_len, hidden_size)
+        result = self.conv(pad(inputs))
+        result = torch.transpose(result, 1, 2)
+        result = self.activation(result)
+        result = self.LayerNorm(result)
+        result = self.dropout(result)
+        final_char_seq = result
+
+        if final_seq_char_positions is not None:
+            # Limit transformer query seq and attention mask to these character
+            # positions to greatly reduce the compute cost. Typically, this is just
+            # done for the MLM training task.
+            # TODO add support for MLM
+            raise NotImplementedError("CanineForMaskedLM is currently not supported")
+        else:
+            query_seq = final_char_seq
+
+        return query_seq
+
+
+class CanineSelfAttention(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
+            raise ValueError(
+                f"The hidden size ({config.hidden_size}) is not a multiple of the number of attention "
+                f"heads ({config.num_attention_heads})"
+            )
+
+        self.num_attention_heads = config.num_attention_heads
+        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
+        self.all_head_size = self.num_attention_heads * self.attention_head_size
+
+        self.query = nn.Linear(config.hidden_size, self.all_head_size)
+        self.key = nn.Linear(config.hidden_size, self.all_head_size)
+        self.value = nn.Linear(config.hidden_size, self.all_head_size)
+
+        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
+        self.position_embedding_type = getattr(config, "position_embedding_type", "absolute")
+        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
+            self.max_position_embeddings = config.max_position_embeddings
+            self.distance_embedding = nn.Embedding(2 * config.max_position_embeddings - 1, self.attention_head_size)
+
+    def transpose_for_scores(self, x):
+        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
+        x = x.view(*new_x_shape)
+        return x.permute(0, 2, 1, 3)
+
+    def forward(
+        self,
+        from_tensor,
+        to_tensor,
+        attention_mask=None,
+        head_mask=None,
+        output_attentions=False,
+    ):
+        mixed_query_layer = self.query(from_tensor)
+
+        # If this is instantiated as a cross-attention module, the keys
+        # and values come from an encoder; the attention mask needs to be
+        # such that the encoder's padding tokens are not attended to.
+
+        key_layer = self.transpose_for_scores(self.key(to_tensor))
+        value_layer = self.transpose_for_scores(self.value(to_tensor))
+
+        query_layer = self.transpose_for_scores(mixed_query_layer)
+
+        # Take the dot product between "query" and "key" to get the raw attention scores.
+        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
+
+        if self.position_embedding_type == "relative_key" or self.position_embedding_type == "relative_key_query":
+            seq_length = from_tensor.size()[1]
+            position_ids_l = torch.arange(seq_length, dtype=torch.long, device=from_tensor.device).view(-1, 1)
+            position_ids_r = torch.arange(seq_length, dtype=torch.long, device=from_tensor.device).view(1, -1)
+            distance = position_ids_l - position_ids_r
+            positional_embedding = self.distance_embedding(distance + self.max_position_embeddings - 1)
+            positional_embedding = positional_embedding.to(dtype=query_layer.dtype)  # fp16 compatibility
+
+            if self.position_embedding_type == "relative_key":
+                relative_position_scores = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
+                attention_scores = attention_scores + relative_position_scores
+            elif self.position_embedding_type == "relative_key_query":
+                relative_position_scores_query = torch.einsum("bhld,lrd->bhlr", query_layer, positional_embedding)
+                relative_position_scores_key = torch.einsum("bhrd,lrd->bhlr", key_layer, positional_embedding)
+                attention_scores = attention_scores + relative_position_scores_query + relative_position_scores_key
+
+        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
+        if attention_mask is not None:
+            if attention_mask.ndim == 3:
+                # if attention_mask is 3D, do the following:
+                attention_mask = torch.unsqueeze(attention_mask, dim=1)
+                # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
+                # masked positions, this operation will create a tensor which is 0.0 for
+                # positions we want to attend and -10000.0 for masked positions.
+                attention_mask = (1.0 - attention_mask.float()) * -10000.0
+            # Apply the attention mask (precomputed for all layers in CanineModel forward() function)
+            attention_scores = attention_scores + attention_mask
+
+        # Normalize the attention scores to probabilities.
+        attention_probs = nn.Softmax(dim=-1)(attention_scores)
+
+        # This is actually dropping out entire tokens to attend to, which might
+        # seem a bit unusual, but is taken from the original Transformer paper.
+        attention_probs = self.dropout(attention_probs)
+
+        # Mask heads if we want to
+        if head_mask is not None:
+            attention_probs = attention_probs * head_mask
+
+        context_layer = torch.matmul(attention_probs, value_layer)
+
+        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
+        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
+        context_layer = context_layer.view(*new_context_layer_shape)
+
+        outputs = (context_layer, attention_probs) if output_attentions else (context_layer,)
+
+        return outputs
+
+
+class CanineSelfOutput(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+    def forward(self, hidden_states, input_tensor):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states + input_tensor)
+        return hidden_states
+
+
+class CanineAttention(nn.Module):
+    """
+    Additional arguments related to local attention:
+
+        - **local** (:obj:`bool`, `optional`, defaults to :obj:`False`) -- Whether to apply local attention.
+        - **always_attend_to_first_position** (:obj:`bool`, `optional`, defaults to :obj:`False`) -- Should all blocks
+          be able to attend
+        to the :obj:`to_tensor`'s first position (e.g. a [CLS] position)? - **first_position_attends_to_all**
+        (:obj:`bool`, `optional`, defaults to :obj:`False`) -- Should the `from_tensor`'s first position be able to
+        attend to all positions within the `from_tensor`? - **attend_from_chunk_width** (:obj:`int`, `optional`,
+        defaults to 128) -- The width of each block-wise chunk in :obj:`from_tensor`. - **attend_from_chunk_stride**
+        (:obj:`int`, `optional`, defaults to 128) -- The number of elements to skip when moving to the next block in
+        :obj:`from_tensor`. - **attend_to_chunk_width** (:obj:`int`, `optional`, defaults to 128) -- The width of each
+        block-wise chunk in `to_tensor`. - **attend_to_chunk_stride** (:obj:`int`, `optional`, defaults to 128) -- The
+        number of elements to skip when moving to the next block in :obj:`to_tensor`.
+    """
+
+    def __init__(
+        self,
+        config,
+        local=False,
+        always_attend_to_first_position: bool = False,
+        first_position_attends_to_all: bool = False,
+        attend_from_chunk_width: int = 128,
+        attend_from_chunk_stride: int = 128,
+        attend_to_chunk_width: int = 128,
+        attend_to_chunk_stride: int = 128,
+    ):
+        super().__init__()
+        self.self = CanineSelfAttention(config)
+        self.output = CanineSelfOutput(config)
+        self.pruned_heads = set()
+
+        # additional arguments related to local attention
+        self.local = local
+        if attend_from_chunk_width < attend_from_chunk_stride:
+            raise ValueError(
+                "`attend_from_chunk_width` < `attend_from_chunk_stride`"
+                "would cause sequence positions to get skipped."
+            )
+        if attend_to_chunk_width < attend_to_chunk_stride:
+            raise ValueError(
+                "`attend_to_chunk_width` < `attend_to_chunk_stride`" "would cause sequence positions to get skipped."
+            )
+        self.always_attend_to_first_position = always_attend_to_first_position
+        self.first_position_attends_to_all = first_position_attends_to_all
+        self.attend_from_chunk_width = attend_from_chunk_width
+        self.attend_from_chunk_stride = attend_from_chunk_stride
+        self.attend_to_chunk_width = attend_to_chunk_width
+        self.attend_to_chunk_stride = attend_to_chunk_stride
+
+    def prune_heads(self, heads):
+        if len(heads) == 0:
+            return
+        heads, index = find_pruneable_heads_and_indices(
+            heads, self.self.num_attention_heads, self.self.attention_head_size, self.pruned_heads
+        )
+
+        # Prune linear layers
+        self.self.query = prune_linear_layer(self.self.query, index)
+        self.self.key = prune_linear_layer(self.self.key, index)
+        self.self.value = prune_linear_layer(self.self.value, index)
+        self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)
+
+        # Update hyper params and store pruned heads
+        self.self.num_attention_heads = self.self.num_attention_heads - len(heads)
+        self.self.all_head_size = self.self.attention_head_size * self.self.num_attention_heads
+        self.pruned_heads = self.pruned_heads.union(heads)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        output_attentions=False,
+    ):
+        if not self.local:
+            self_outputs = self.self(hidden_states, hidden_states, attention_mask, head_mask, output_attentions)
+            attention_output = self_outputs[0]
+        else:
+            from_seq_length = to_seq_length = hidden_states.shape[1]
+            from_tensor = to_tensor = hidden_states
+
+            # Create chunks (windows) that we will attend *from* and then concatenate them.
+            from_chunks = []
+            if self.first_position_attends_to_all:
+                from_chunks.append((0, 1))
+                # We must skip this first position so that our output sequence is the
+                # correct length (this matters in the *from* sequence only).
+                from_start = 1
+            else:
+                from_start = 0
+            for chunk_start in range(from_start, from_seq_length, self.attend_from_chunk_stride):
+                chunk_end = min(from_seq_length, chunk_start + self.attend_from_chunk_width)
+                from_chunks.append((chunk_start, chunk_end))
+
+            # Determine the chunks (windows) that will will attend *to*.
+            to_chunks = []
+            if self.first_position_attends_to_all:
+                to_chunks.append((0, to_seq_length))
+            for chunk_start in range(0, to_seq_length, self.attend_to_chunk_stride):
+                chunk_end = min(to_seq_length, chunk_start + self.attend_to_chunk_width)
+                to_chunks.append((chunk_start, chunk_end))
+
+            if len(from_chunks) != len(to_chunks):
+                raise ValueError(
+                    f"Expected to have same number of `from_chunks` ({from_chunks}) and "
+                    f"`to_chunks` ({from_chunks}). Check strides."
+                )
+
+            # next, compute attention scores for each pair of windows and concatenate
+            attention_output_chunks = []
+            attention_probs_chunks = []
+            for (from_start, from_end), (to_start, to_end) in zip(from_chunks, to_chunks):
+                from_tensor_chunk = from_tensor[:, from_start:from_end, :]
+                to_tensor_chunk = to_tensor[:, to_start:to_end, :]
+                # `attention_mask`: <float>[batch_size, from_seq, to_seq]
+                # `attention_mask_chunk`: <float>[batch_size, from_seq_chunk, to_seq_chunk]
+                attention_mask_chunk = attention_mask[:, from_start:from_end, to_start:to_end]
+                if self.always_attend_to_first_position:
+                    cls_attention_mask = attention_mask[:, from_start:from_end, 0:1]
+                    attention_mask_chunk = torch.cat([cls_attention_mask, attention_mask_chunk], dim=2)
+
+                    cls_position = to_tensor[:, 0:1, :]
+                    to_tensor_chunk = torch.cat([cls_position, to_tensor_chunk], dim=1)
+
+                attention_outputs_chunk = self.self(
+                    from_tensor_chunk, to_tensor_chunk, attention_mask_chunk, head_mask, output_attentions
+                )
+                attention_output_chunks.append(attention_outputs_chunk[0])
+                if output_attentions:
+                    attention_probs_chunks.append(attention_outputs_chunk[1])
+
+            attention_output = torch.cat(attention_output_chunks, dim=1)
+
+        attention_output = self.output(attention_output, hidden_states)
+        outputs = (attention_output,)
+        if not self.local:
+            outputs = outputs + self_outputs[1:]  # add attentions if we output them
+        else:
+            outputs = outputs + tuple(attention_probs_chunks)  # add attentions if we output them
+        return outputs
+
+
+class CanineIntermediate(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
+        if isinstance(config.hidden_act, str):
+            self.intermediate_act_fn = ACT2FN[config.hidden_act]
+        else:
+            self.intermediate_act_fn = config.hidden_act
+
+    def forward(self, hidden_states):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.intermediate_act_fn(hidden_states)
+        return hidden_states
+
+
+class CanineOutput(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+    def forward(self, hidden_states, input_tensor):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states + input_tensor)
+        return hidden_states
+
+
+class CanineLayer(nn.Module):
+    def __init__(
+        self,
+        config,
+        local,
+        always_attend_to_first_position,
+        first_position_attends_to_all,
+        attend_from_chunk_width,
+        attend_from_chunk_stride,
+        attend_to_chunk_width,
+        attend_to_chunk_stride,
+    ):
+        super().__init__()
+        self.chunk_size_feed_forward = config.chunk_size_feed_forward
+        self.seq_len_dim = 1
+        self.attention = CanineAttention(
+            config,
+            local,
+            always_attend_to_first_position,
+            first_position_attends_to_all,
+            attend_from_chunk_width,
+            attend_from_chunk_stride,
+            attend_to_chunk_width,
+            attend_to_chunk_stride,
+        )
+        self.intermediate = CanineIntermediate(config)
+        self.output = CanineOutput(config)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        output_attentions=False,
+    ):
+        self_attention_outputs = self.attention(
+            hidden_states,
+            attention_mask,
+            head_mask,
+            output_attentions=output_attentions,
+        )
+        attention_output = self_attention_outputs[0]
+
+        outputs = self_attention_outputs[1:]  # add self attentions if we output attention weights
+
+        layer_output = apply_chunking_to_forward(
+            self.feed_forward_chunk, self.chunk_size_feed_forward, self.seq_len_dim, attention_output
+        )
+        outputs = (layer_output,) + outputs
+
+        return outputs
+
+    def feed_forward_chunk(self, attention_output):
+        intermediate_output = self.intermediate(attention_output)
+        layer_output = self.output(intermediate_output, attention_output)
+        return layer_output
+
+
+class CanineEncoder(nn.Module):
+    def __init__(
+        self,
+        config,
+        local=False,
+        always_attend_to_first_position=False,
+        first_position_attends_to_all=False,
+        attend_from_chunk_width=128,
+        attend_from_chunk_stride=128,
+        attend_to_chunk_width=128,
+        attend_to_chunk_stride=128,
+    ):
+        super().__init__()
+        self.config = config
+        self.layer = nn.ModuleList(
+            [
+                CanineLayer(
+                    config,
+                    local,
+                    always_attend_to_first_position,
+                    first_position_attends_to_all,
+                    attend_from_chunk_width,
+                    attend_from_chunk_stride,
+                    attend_to_chunk_width,
+                    attend_to_chunk_stride,
+                )
+                for _ in range(config.num_hidden_layers)
+            ]
+        )
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        head_mask=None,
+        output_attentions=False,
+        output_hidden_states=False,
+        return_dict=True,
+    ):
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attentions = () if output_attentions else None
+
+        for i, layer_module in enumerate(self.layer):
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            layer_head_mask = head_mask[i] if head_mask is not None else None
+
+            if getattr(self.config, "gradient_checkpointing", False) and self.training:
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        return module(*inputs, output_attentions)
+
+                    return custom_forward
+
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(layer_module),
+                    hidden_states,
+                    attention_mask,
+                    layer_head_mask,
+                )
+            else:
+                layer_outputs = layer_module(hidden_states, attention_mask, layer_head_mask, output_attentions)
+
+            hidden_states = layer_outputs[0]
+            if output_attentions:
+                all_self_attentions = all_self_attentions + (layer_outputs[1],)
+
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(v for v in [hidden_states, all_hidden_states, all_self_attentions] if v is not None)
+        return BaseModelOutput(
+            last_hidden_state=hidden_states,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+        )
+
+
+class CaninePooler(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        self.activation = nn.Tanh()
+
+    def forward(self, hidden_states):
+        # We "pool" the model by simply taking the hidden state corresponding
+        # to the first token.
+        first_token_tensor = hidden_states[:, 0]
+        pooled_output = self.dense(first_token_tensor)
+        pooled_output = self.activation(pooled_output)
+        return pooled_output
+
+
+class CaninePredictionHeadTransform(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        if isinstance(config.hidden_act, str):
+            self.transform_act_fn = ACT2FN[config.hidden_act]
+        else:
+            self.transform_act_fn = config.hidden_act
+        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
+
+    def forward(self, hidden_states):
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.transform_act_fn(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states)
+        return hidden_states
+
+
+class CanineLMPredictionHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.transform = CaninePredictionHeadTransform(config)
+
+        # The output weights are the same as the input embeddings, but there is
+        # an output-only bias for each token.
+        self.decoder = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        self.bias = nn.Parameter(torch.zeros(config.vocab_size))
+
+        # Need a link between the two variables so that the bias is correctly resized with `resize_token_embeddings`
+        self.decoder.bias = self.bias
+
+    def forward(self, hidden_states):
+        hidden_states = self.transform(hidden_states)
+        hidden_states = self.decoder(hidden_states)
+        return hidden_states
+
+
+class CanineOnlyMLMHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.predictions = CanineLMPredictionHead(config)
+
+    def forward(self, sequence_output):
+        prediction_scores = self.predictions(sequence_output)
+        return prediction_scores
+
+
+class CaninePreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = CanineConfig
+    load_tf_weights = load_tf_weights_in_canine
+    base_model_prefix = "canine"
+    _keys_to_ignore_on_load_missing = [r"position_ids"]
+
+    def _init_weights(self, module):
+        """Initialize the weights"""
+        if isinstance(module, (nn.Linear, nn.Conv1d)):
+            # Slightly different from the TF version which uses truncated_normal for initialization
+            # cf https://github.com/pytorch/pytorch/pull/5617
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+        elif isinstance(module, nn.LayerNorm):
+            module.bias.data.zero_()
+            module.weight.data.fill_(1.0)
+
+
+CANINE_START_DOCSTRING = r"""
+    This model is a PyTorch `torch.nn.Module <https://pytorch.org/docs/stable/nn.html#torch.nn.Module>`_ sub-class. Use
+    it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and
+    behavior.
+
+    Parameters:
+        config (:class:`~transformers.CanineConfig`): Model configuration class with all the parameters of the model.
+            Initializing with a config file does not load the weights associated with the model, only the
+            configuration. Check out the :meth:`~transformers.PreTrainedModel.from_pretrained` method to load the model
+            weights.
+"""
+
+CANINE_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`):
+            Indices of input sequence tokens in the vocabulary.
+
+            Indices can be obtained using :class:`transformers.CanineTokenizer`. See
+            :func:`transformers.PreTrainedTokenizer.encode` and :func:`transformers.PreTrainedTokenizer.__call__` for
+            details.
+
+            `What are input IDs? <../glossary.html#input-ids>`__
+        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`{0}`, `optional`):
+            Mask to avoid performing attention on padding token indices. Mask values selected in ``[0, 1]``:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            `What are attention masks? <../glossary.html#attention-mask>`__
+        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `optional`):
+            Segment token indices to indicate first and second portions of the inputs. Indices are selected in ``[0,
+            1]``:
+
+            - 0 corresponds to a `sentence A` token,
+            - 1 corresponds to a `sentence B` token.
+
+            `What are token type IDs? <../glossary.html#token-type-ids>`_
+        position_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `optional`):
+            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range ``[0,
+            config.max_position_embeddings - 1]``.
+
+            `What are position IDs? <../glossary.html#position-ids>`_
+        head_mask (:obj:`torch.FloatTensor` of shape :obj:`(num_heads,)` or :obj:`(num_layers, num_heads)`, `optional`):
+            Mask to nullify selected heads of the self-attention modules. Mask values selected in ``[0, 1]``:
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+
+        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+            Optionally, instead of passing :obj:`input_ids` you can choose to directly pass an embedded representation.
+            This is useful if you want more control over how to convert `input_ids` indices into associated vectors
+            than the model's internal embedding lookup matrix.
+        output_attentions (:obj:`bool`, `optional`):
+            Whether or not to return the attentions tensors of all attention layers. See ``attentions`` under returned
+            tensors for more detail.
+        output_hidden_states (:obj:`bool`, `optional`):
+            Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors for
+            more detail.
+        return_dict (:obj:`bool`, `optional`):
+            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
+"""
+
+
+@add_start_docstrings(
+    "The bare CANINE Model transformer outputting raw hidden-states without any specific head on top.",
+    CANINE_START_DOCSTRING,
+)
+class CanineModel(CaninePreTrainedModel):
+    def __init__(self, config, add_pooling_layer=True):
+        super().__init__(config)
+        self.config = config
+        shallow_config = copy.deepcopy(config)
+        shallow_config.num_hidden_layers = 1
+
+        self.char_embeddings = CanineEmbeddings(config)
+        # shallow/low-dim transformer encoder to get a initial character encoding
+        self.initial_char_encoder = CanineEncoder(
+            shallow_config,
+            local=True,
+            always_attend_to_first_position=False,
+            first_position_attends_to_all=False,
+            attend_from_chunk_width=config.local_transformer_stride,
+            attend_from_chunk_stride=config.local_transformer_stride,
+            attend_to_chunk_width=config.local_transformer_stride,
+            attend_to_chunk_stride=config.local_transformer_stride,
+        )
+        self.chars_to_molecules = CharactersToMolecules(config)
+        # deep transformer encoder
+        self.encoder = CanineEncoder(config)
+        self.projection = ConvProjection(config)
+        # shallow/low-dim transformer encoder to get a final character encoding
+        self.final_char_encoder = CanineEncoder(shallow_config)
+
+        self.pooler = CaninePooler(config) if add_pooling_layer else None
+
+        self.init_weights()
+
+    def _prune_heads(self, heads_to_prune):
+        """
+        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} See base
+        class PreTrainedModel
+        """
+        for layer, heads in heads_to_prune.items():
+            self.encoder.layer[layer].attention.prune_heads(heads)
+
+    def _create_3d_attention_mask_from_input_mask(self, from_tensor, to_mask):
+        """
+        Create 3D attention mask from a 2D tensor mask.
+
+        Args:
+            from_tensor: 2D or 3D Tensor of shape [batch_size, from_seq_length, ...].
+            to_mask: int32 Tensor of shape [batch_size, to_seq_length].
+
+        Returns:
+            float Tensor of shape [batch_size, from_seq_length, to_seq_length].
+        """
+        batch_size, from_seq_length = from_tensor.shape[0], from_tensor.shape[1]
+
+        to_seq_length = to_mask.shape[1]
+
+        to_mask = torch.reshape(to_mask, (batch_size, 1, to_seq_length)).float()
+
+        # We don't assume that `from_tensor` is a mask (although it could be). We
+        # don't actually care if we attend *from* padding tokens (only *to* padding)
+        # tokens so we create a tensor of all ones.
+        broadcast_ones = torch.ones(size=(batch_size, from_seq_length, 1), dtype=torch.float32, device=to_mask.device)
+
+        # Here we broadcast along two dimensions to create the mask.
+        mask = broadcast_ones * to_mask
+
+        return mask
+
+    def _downsample_attention_mask(self, char_attention_mask: torch.Tensor, downsampling_rate: int):
+        """Downsample 2D character attention mask to 2D molecule attention mask using MaxPool1d layer."""
+
+        # first, make char_attention_mask 3D by adding a channel dim
+        batch_size, char_seq_len = char_attention_mask.shape
+        poolable_char_mask = torch.reshape(char_attention_mask, (batch_size, 1, char_seq_len))
+
+        # next, apply MaxPool1d to get pooled_molecule_mask of shape (batch_size, 1, mol_seq_len)
+        pooled_molecule_mask = torch.nn.MaxPool1d(kernel_size=downsampling_rate, stride=downsampling_rate)(
+            poolable_char_mask.float()
+        )
+
+        # finally, squeeze to get tensor of shape (batch_size, mol_seq_len)
+        molecule_attention_mask = torch.squeeze(pooled_molecule_mask, dim=-1)
+
+        return molecule_attention_mask
+
+    def _repeat_molecules(self, molecules: torch.Tensor, char_seq_length: torch.Tensor) -> torch.Tensor:
+        """Repeats molecules to make them the same length as the char sequence."""
+
+        rate = self.config.downsampling_rate
+
+        molecules_without_extra_cls = molecules[:, 1:, :]
+        # `repeated`: [batch_size, almost_char_seq_len, molecule_hidden_size]
+        repeated = torch.repeat_interleave(molecules_without_extra_cls, repeats=rate, dim=-2)
+
+        # So far, we've repeated the elements sufficient for any `char_seq_length`
+        # that's a multiple of `downsampling_rate`. Now we account for the last
+        # n elements (n < `downsampling_rate`), i.e. the remainder of floor
+        # division. We do this by repeating the last molecule a few extra times.
+        last_molecule = molecules[:, -1:, :]
+        remainder_length = torch.fmod(torch.tensor(char_seq_length), torch.tensor(rate)).item()
+        remainder_repeated = torch.repeat_interleave(
+            last_molecule,
+            # +1 molecule to compensate for truncation.
+            repeats=remainder_length + rate,
+            dim=-2,
+        )
+
+        # `repeated`: [batch_size, char_seq_len, molecule_hidden_size]
+        return torch.cat([repeated, remainder_repeated], dim=-2)
+
+    @add_start_docstrings_to_model_forward(CANINE_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_code_sample_docstrings(
+        tokenizer_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=CanineModelOutputWithPooling,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attentions = () if output_attentions else None
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
+        elif input_ids is not None:
+            input_shape = input_ids.size()
+            batch_size, seq_length = input_shape
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.size()[:-1]
+            batch_size, seq_length = input_shape
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        device = input_ids.device if input_ids is not None else inputs_embeds.device
+
+        if attention_mask is None:
+            attention_mask = torch.ones(((batch_size, seq_length)), device=device)
+        if token_type_ids is None:
+            token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=device)
+
+        # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
+        # ourselves in which case we just need to make it broadcastable to all heads.
+        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, device)
+        molecule_attention_mask = self._downsample_attention_mask(
+            attention_mask, downsampling_rate=self.config.downsampling_rate
+        )
+        extended_molecule_attention_mask: torch.Tensor = self.get_extended_attention_mask(
+            molecule_attention_mask, (batch_size, molecule_attention_mask.shape[-1]), device
+        )
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x n_heads x N x N
+        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
+        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
+        head_mask = self.get_head_mask(head_mask, self.config.num_hidden_layers)
+
+        # `input_char_embeddings`: shape (batch_size, char_seq, char_dim)
+        input_char_embeddings = self.char_embeddings(
+            input_ids=input_ids,
+            position_ids=position_ids,
+            token_type_ids=token_type_ids,
+            inputs_embeds=inputs_embeds,
+        )
+
+        # Contextualize character embeddings using shallow Transformer.
+        # We use a 3D attention mask for the local attention.
+        # `input_char_encoding`: shape (batch_size, char_seq_len, char_dim)
+        char_attention_mask = self._create_3d_attention_mask_from_input_mask(input_ids, attention_mask)
+        init_chars_encoder_outputs = self.initial_char_encoder(
+            input_char_embeddings,
+            attention_mask=char_attention_mask,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+        )
+        input_char_encoding = init_chars_encoder_outputs.last_hidden_state
+
+        # Downsample chars to molecules.
+        # The following lines have dimensions: [batch, molecule_seq, molecule_dim].
+        # In this transformation, we change the dimensionality from `char_dim` to
+        # `molecule_dim`, but do *NOT* add a resnet connection. Instead, we rely on
+        # the resnet connections (a) from the final char transformer stack back into
+        # the original char transformer stack and (b) the resnet connections from
+        # the final char transformer stack back into the deep BERT stack of
+        # molecules.
+        #
+        # Empirically, it is critical to use a powerful enough transformation here:
+        # mean pooling causes training to diverge with huge gradient norms in this
+        # region of the model; using a convolution here resolves this issue. From
+        # this, it seems that molecules and characters require a very different
+        # feature space; intuitively, this makes sense.
+        init_molecule_encoding = self.chars_to_molecules(input_char_encoding)
+
+        # Deep BERT encoder
+        # `molecule_sequence_output`: shape (batch_size, mol_seq_len, mol_dim)
+        encoder_outputs = self.encoder(
+            init_molecule_encoding,
+            attention_mask=extended_molecule_attention_mask,
+            head_mask=head_mask,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        molecule_sequence_output = encoder_outputs[0]
+        pooled_output = self.pooler(molecule_sequence_output) if self.pooler is not None else None
+
+        # Upsample molecules back to characters.
+        # `repeated_molecules`: shape (batch_size, char_seq_len, mol_hidden_size)
+        repeated_molecules = self._repeat_molecules(molecule_sequence_output, char_seq_length=input_shape[-1])
+
+        # Concatenate representations (contextualized char embeddings and repeated molecules):
+        # `concat`: shape [batch_size, char_seq_len, molecule_hidden_size+char_hidden_final]
+        concat = torch.cat([input_char_encoding, repeated_molecules], dim=-1)
+
+        # Project representation dimension back to hidden_size
+        # `sequence_output`: shape (batch_size, char_seq_len, hidden_size])
+        sequence_output = self.projection(concat)
+
+        # Apply final shallow Transformer
+        # `sequence_output`: shape (batch_size, char_seq_len, hidden_size])
+        final_chars_encoder_outputs = self.final_char_encoder(
+            sequence_output,
+            attention_mask=extended_attention_mask,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+        )
+        sequence_output = final_chars_encoder_outputs.last_hidden_state
+
+        if output_hidden_states:
+            deep_encoder_hidden_states = encoder_outputs.hidden_states if return_dict else encoder_outputs[1]
+            all_hidden_states = (
+                all_hidden_states
+                + init_chars_encoder_outputs.hidden_states
+                + deep_encoder_hidden_states
+                + final_chars_encoder_outputs.hidden_states
+            )
+
+        if output_attentions:
+            deep_encoder_self_attentions = encoder_outputs.attentions if return_dict else encoder_outputs[-1]
+            all_self_attentions = (
+                all_self_attentions
+                + init_chars_encoder_outputs.attentions
+                + deep_encoder_self_attentions
+                + final_chars_encoder_outputs.attentions
+            )
+
+        if not return_dict:
+            output = (sequence_output, pooled_output)
+            output += tuple(v for v in [all_hidden_states, all_self_attentions] if v is not None)
+            return output
+
+        return CanineModelOutputWithPooling(
+            last_hidden_state=sequence_output,
+            pooler_output=pooled_output,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+        )
+
+
+@add_start_docstrings(
+    """
+    CANINE Model transformer with a sequence classification/regression head on top (a linear layer on top of the pooled
+    output) e.g. for GLUE tasks.
+    """,
+    CANINE_START_DOCSTRING,
+)
+class CanineForSequenceClassification(CaninePreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+
+        self.canine = CanineModel(config)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+        self.classifier = nn.Linear(config.hidden_size, config.num_labels)
+
+        self.init_weights()
+
+    @add_start_docstrings_to_model_forward(CANINE_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_code_sample_docstrings(
+        tokenizer_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=SequenceClassifierOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        labels=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        r"""
+        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`):
+            Labels for computing the sequence classification/regression loss. Indices should be in :obj:`[0, ...,
+            config.num_labels - 1]`. If :obj:`config.num_labels == 1` a regression loss is computed (Mean-Square loss),
+            If :obj:`config.num_labels > 1` a classification loss is computed (Cross-Entropy).
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.canine(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        pooled_output = outputs[1]
+
+        pooled_output = self.dropout(pooled_output)
+        logits = self.classifier(pooled_output)
+
+        loss = None
+        if labels is not None:
+            if self.config.problem_type is None:
+                if self.num_labels == 1:
+                    self.config.problem_type = "regression"
+                elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
+                    self.config.problem_type = "single_label_classification"
+                else:
+                    self.config.problem_type = "multi_label_classification"
+
+            if self.config.problem_type == "regression":
+                loss_fct = MSELoss()
+                if self.num_labels == 1:
+                    loss = loss_fct(logits.squeeze(), labels.squeeze())
+                else:
+                    loss = loss_fct(logits, labels)
+            elif self.config.problem_type == "single_label_classification":
+                loss_fct = CrossEntropyLoss()
+                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
+            elif self.config.problem_type == "multi_label_classification":
+                loss_fct = BCEWithLogitsLoss()
+                loss = loss_fct(logits, labels)
+        if not return_dict:
+            output = (logits,) + outputs[2:]
+            return ((loss,) + output) if loss is not None else output
+
+        return SequenceClassifierOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+@add_start_docstrings(
+    """
+    CANINE Model with a multiple choice classification head on top (a linear layer on top of the pooled output and a
+    softmax) e.g. for RocStories/SWAG tasks.
+    """,
+    CANINE_START_DOCSTRING,
+)
+class CanineForMultipleChoice(CaninePreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.canine = CanineModel(config)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+        self.classifier = nn.Linear(config.hidden_size, 1)
+
+        self.init_weights()
+
+    @add_start_docstrings_to_model_forward(CANINE_INPUTS_DOCSTRING.format("batch_size, num_choices, sequence_length"))
+    @add_code_sample_docstrings(
+        tokenizer_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=MultipleChoiceModelOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        labels=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        r"""
+        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`):
+            Labels for computing the multiple choice classification loss. Indices should be in ``[0, ...,
+            num_choices-1]`` where :obj:`num_choices` is the size of the second dimension of the input tensors. (See
+            :obj:`input_ids` above)
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        num_choices = input_ids.shape[1] if input_ids is not None else inputs_embeds.shape[1]
+
+        input_ids = input_ids.view(-1, input_ids.size(-1)) if input_ids is not None else None
+        attention_mask = attention_mask.view(-1, attention_mask.size(-1)) if attention_mask is not None else None
+        token_type_ids = token_type_ids.view(-1, token_type_ids.size(-1)) if token_type_ids is not None else None
+        position_ids = position_ids.view(-1, position_ids.size(-1)) if position_ids is not None else None
+        inputs_embeds = (
+            inputs_embeds.view(-1, inputs_embeds.size(-2), inputs_embeds.size(-1))
+            if inputs_embeds is not None
+            else None
+        )
+
+        outputs = self.canine(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        pooled_output = outputs[1]
+
+        pooled_output = self.dropout(pooled_output)
+        logits = self.classifier(pooled_output)
+        reshaped_logits = logits.view(-1, num_choices)
+
+        loss = None
+        if labels is not None:
+            loss_fct = CrossEntropyLoss()
+            loss = loss_fct(reshaped_logits, labels)
+
+        if not return_dict:
+            output = (reshaped_logits,) + outputs[2:]
+            return ((loss,) + output) if loss is not None else output
+
+        return MultipleChoiceModelOutput(
+            loss=loss,
+            logits=reshaped_logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+@add_start_docstrings(
+    """
+    CANINE Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for
+    Named-Entity-Recognition (NER) tasks.
+    """,
+    CANINE_START_DOCSTRING,
+)
+class CanineForTokenClassification(CaninePreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+
+        self.canine = CanineModel(config)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+        self.classifier = nn.Linear(config.hidden_size, config.num_labels)
+
+        self.init_weights()
+
+    @add_start_docstrings_to_model_forward(CANINE_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_code_sample_docstrings(
+        tokenizer_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=TokenClassifierOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        labels=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        r"""
+        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
+            Labels for computing the token classification loss. Indices should be in ``[0, ..., config.num_labels -
+            1]``.
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.canine(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        sequence_output = outputs[0]
+
+        sequence_output = self.dropout(sequence_output)
+        logits = self.classifier(sequence_output)
+
+        loss = None
+        if labels is not None:
+            loss_fct = CrossEntropyLoss()
+            # Only keep active parts of the loss
+            if attention_mask is not None:
+                active_loss = attention_mask.view(-1) == 1
+                active_logits = logits.view(-1, self.num_labels)
+                active_labels = torch.where(
+                    active_loss, labels.view(-1), torch.tensor(loss_fct.ignore_index).type_as(labels)
+                )
+                loss = loss_fct(active_logits, active_labels)
+            else:
+                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
+
+        if not return_dict:
+            output = (logits,) + outputs[2:]
+            return ((loss,) + output) if loss is not None else output
+
+        return TokenClassifierOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+@add_start_docstrings(
+    """
+    CANINE Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear
+    layers on top of the hidden-states output to compute `span start logits` and `span end logits`).
+    """,
+    CANINE_START_DOCSTRING,
+)
+class CanineForQuestionAnswering(CaninePreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+
+        self.canine = CanineModel(config)
+        self.qa_outputs = nn.Linear(config.hidden_size, config.num_labels)
+
+        self.init_weights()
+
+    @add_start_docstrings_to_model_forward(CANINE_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_code_sample_docstrings(
+        tokenizer_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=QuestionAnsweringModelOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        start_positions=None,
+        end_positions=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        r"""
+        start_positions (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`):
+            Labels for position (index) of the start of the labelled span for computing the token classification loss.
+            Positions are clamped to the length of the sequence (:obj:`sequence_length`). Position outside of the
+            sequence are not taken into account for computing the loss.
+        end_positions (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`):
+            Labels for position (index) of the end of the labelled span for computing the token classification loss.
+            Positions are clamped to the length of the sequence (:obj:`sequence_length`). Position outside of the
+            sequence are not taken into account for computing the loss.
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.canine(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        sequence_output = outputs[0]
+
+        logits = self.qa_outputs(sequence_output)
+        start_logits, end_logits = logits.split(1, dim=-1)
+        start_logits = start_logits.squeeze(-1)
+        end_logits = end_logits.squeeze(-1)
+
+        total_loss = None
+        if start_positions is not None and end_positions is not None:
+            # If we are on multi-GPU, split add a dimension
+            if len(start_positions.size()) > 1:
+                start_positions = start_positions.squeeze(-1)
+            if len(end_positions.size()) > 1:
+                end_positions = end_positions.squeeze(-1)
+            # sometimes the start/end positions are outside our model inputs, we ignore these terms
+            ignored_index = start_logits.size(1)
+            start_positions.clamp_(0, ignored_index)
+            end_positions.clamp_(0, ignored_index)
+
+            loss_fct = CrossEntropyLoss(ignore_index=ignored_index)
+            start_loss = loss_fct(start_logits, start_positions)
+            end_loss = loss_fct(end_logits, end_positions)
+            total_loss = (start_loss + end_loss) / 2
+
+        if not return_dict:
+            output = (start_logits, end_logits) + outputs[2:]
+            return ((total_loss,) + output) if total_loss is not None else output
+
+        return QuestionAnsweringModelOutput(
+            loss=total_loss,
+            start_logits=start_logits,
+            end_logits=end_logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )

src/transformers/models/canine/tokenization_canine.py

+# coding=utf-8
+# Copyright Google AI and The HuggingFace Inc. team. All rights reserved.
+#
+# 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.
+"""Tokenization classes for CANINE."""
+
+from typing import Dict, List, Optional
+
+from ...tokenization_utils import AddedToken, PreTrainedTokenizer
+from ...utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    "nielsr/canine-s": 2048,
+}
+
+# Unicode defines 1,114,112 total “codepoints”
+UNICODE_VOCAB_SIZE = 1114112
+
+# Below: Constants defining canonical codepoints for special, pseudo-characters.
+# Copied from https://github.com/google-research/language/blob/master/language/canine/special_codepoints.py
+PAD = 0
+
+CLS = 0xE000
+SEP = 0xE001
+BOS = 0xE002
+MASK = 0xE003
+RESERVED = 0xE004
+
+# Maps special codepoints to human-readable names.
+SPECIAL_CODEPOINTS: Dict[int, str] = {
+    # Special symbols are represented using codepoints values that are valid,
+    # but designated as "Private Use", meaning that they will never be assigned
+    # characters by the Unicode Consortium, and are thus safe for use here.
+    #
+    # NOTE: Do *NOT* add any sort of [UNK_CHAR] here. They are explicitly
+    # excluded and should fail with a hard error.
+    CLS: "[CLS]",
+    SEP: "[SEP]",
+    BOS: "[BOS]",
+    MASK: "[MASK]",
+    PAD: "[PAD]",
+    RESERVED: "[RESERVED]",
+}
+
+# Maps special codepoint human-readable names to their codepoint values.
+SPECIAL_CODEPOINTS_BY_NAME: Dict[str, int] = {name: codepoint for codepoint, name in SPECIAL_CODEPOINTS.items()}
+
+
+class CanineTokenizer(PreTrainedTokenizer):
+    r"""
+    Construct a CANINE tokenizer (i.e. a character splitter). It turns text into a sequence of characters, and then
+    converts each character into its Unicode code point.
+
+    :class:`~transformers.CanineTokenizer` inherits from :class:`~transformers.PreTrainedTokenizer`.
+
+    Refer to superclass :class:`~transformers.PreTrainedTokenizer` for usage examples and documentation concerning
+    parameters.
+
+    Args:
+        model_max_length (:obj:`int`, `optional`, defaults to 2048):
+                The maximum sentence length the model accepts.
+    """
+
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+
+    def __init__(
+        self,
+        bos_token=chr(CLS),
+        eos_token=chr(SEP),
+        sep_token=chr(SEP),
+        cls_token=chr(CLS),
+        pad_token=chr(PAD),
+        mask_token=chr(MASK),
+        add_prefix_space=False,
+        model_max_length=2048,
+        **kwargs
+    ):
+        bos_token = AddedToken(bos_token, lstrip=False, rstrip=False) if isinstance(bos_token, str) else bos_token
+        eos_token = AddedToken(eos_token, lstrip=False, rstrip=False) if isinstance(eos_token, str) else eos_token
+        sep_token = AddedToken(sep_token, lstrip=False, rstrip=False) if isinstance(sep_token, str) else sep_token
+        cls_token = AddedToken(cls_token, lstrip=False, rstrip=False) if isinstance(cls_token, str) else cls_token
+        pad_token = AddedToken(pad_token, lstrip=False, rstrip=False) if isinstance(pad_token, str) else pad_token
+
+        # Mask token behave like a normal word, i.e. include the space before it
+        mask_token = AddedToken(mask_token, lstrip=True, rstrip=False) if isinstance(mask_token, str) else mask_token
+
+        super().__init__(
+            bos_token=bos_token,
+            eos_token=eos_token,
+            sep_token=sep_token,
+            cls_token=cls_token,
+            pad_token=pad_token,
+            mask_token=mask_token,
+            add_prefix_space=add_prefix_space,
+            model_max_length=model_max_length,
+            **kwargs,
+        )
+
+        # Creates a mapping for looking up the IDs of special symbols.
+        self._special_codepoints: Dict[str, int] = {}
+        for codepoint, name in SPECIAL_CODEPOINTS.items():
+            self._special_codepoints[name] = codepoint
+
+        # Creates a mapping for looking up the string forms of special symbol IDs.
+        self._special_codepoint_strings: Dict[int, str] = {
+            codepoint: name for name, codepoint in self._special_codepoints.items()
+        }
+
+        self._unicode_vocab_size = UNICODE_VOCAB_SIZE
+        self._num_special_tokens = len(self._special_codepoints)
+
+    @property
+    def vocab_size(self) -> int:
+        return self._unicode_vocab_size
+
+    def _tokenize(self, text: str) -> List[str]:
+        """Tokenize a string (i.e. perform character splitting)."""
+        return list(text)
+
+    def _convert_token_to_id(self, token: str) -> int:
+        """Converts a token (i.e. a Unicode character) in an id (i.e. its integer Unicode code point value)."""
+        try:
+            return ord(token)
+        except TypeError:
+            raise ValueError(f"invalid token: '{token}'")
+
+    def _convert_id_to_token(self, index: int) -> str:
+        """
+        Converts a Unicode code point (integer) in a token (str). In case it's a special code point, convert to
+        human-readable format.
+        """
+        try:
+            if index in SPECIAL_CODEPOINTS:
+                return SPECIAL_CODEPOINTS[index]
+            return chr(index)
+        except TypeError:
+            raise ValueError(f"invalid id: {index}")
+
+    def convert_tokens_to_string(self, tokens):
+        return "".join(tokens)
+
+    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 CANINE 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.
+        """
+        sep = [self.sep_token_id]
+        cls = [self.cls_token_id]
+
+        result = cls + token_ids_0 + sep
+        if token_ids_1 is not None:
+            result += token_ids_1 + sep
+        return result
+
+    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:
+            return super().get_special_tokens_mask(
+                token_ids_0=token_ids_0, token_ids_1=token_ids_1, already_has_special_tokens=True
+            )
+
+        result = [1] + ([0] * len(token_ids_0)) + [1]
+        if token_ids_1 is not None:
+            result += ([0] * len(token_ids_1)) + [1]
+        return result
+
+    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 CANINE
+        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]
+        cls = [self.cls_token_id]
+
+        result = len(cls + token_ids_0 + sep) * [0]
+        if token_ids_1 is not None:
+            result += len(token_ids_1 + sep) * [1]
+        return result
+
+    # CanineTokenizer has no vocab file
+    def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None):
+        return ()

src/transformers/utils/dummy_pt_objects.py

@@ -995,6 +995,72 @@ class CamembertModel:
         requires_backends(cls, ["torch"])
 
 
+CANINE_PRETRAINED_MODEL_ARCHIVE_LIST = None
+
+
+class CanineForMultipleChoice:
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch"])
+
+
+class CanineForQuestionAnswering:
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch"])
+
+
+class CanineForSequenceClassification:
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch"])
+
+
+class CanineForTokenClassification:
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch"])
+
+
+class CanineLayer:
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class CanineModel:
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch"])
+
+
+class CaninePreTrainedModel:
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch"])
+
+
+def load_tf_weights_in_canine(*args, **kwargs):
+    requires_backends(load_tf_weights_in_canine, ["torch"])
+
+
 CLIP_PRETRAINED_MODEL_ARCHIVE_LIST = None
 
 

src/transformers/utils/modeling_auto_mapping.py

@@ -6,6 +6,7 @@ from collections import OrderedDict
 
 MODEL_FOR_QUESTION_ANSWERING_MAPPING_NAMES = OrderedDict(
     [
+        ("CanineConfig", "CanineForQuestionAnswering"),
         ("RoFormerConfig", "RoFormerForQuestionAnswering"),
         ("BigBirdPegasusConfig", "BigBirdPegasusForQuestionAnswering"),
         ("BigBirdConfig", "BigBirdForQuestionAnswering"),
@@ -112,6 +113,7 @@ MODEL_FOR_MASKED_LM_MAPPING_NAMES = OrderedDict(
 
 MODEL_FOR_MULTIPLE_CHOICE_MAPPING_NAMES = OrderedDict(
     [
+        ("CanineConfig", "CanineForMultipleChoice"),
         ("RoFormerConfig", "RoFormerForMultipleChoice"),
         ("BigBirdConfig", "BigBirdForMultipleChoice"),
         ("ConvBertConfig", "ConvBertForMultipleChoice"),
@@ -175,6 +177,7 @@ MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES = OrderedDict(
 
 MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES = OrderedDict(
     [
+        ("CanineConfig", "CanineForSequenceClassification"),
         ("RoFormerConfig", "RoFormerForSequenceClassification"),
         ("BigBirdPegasusConfig", "BigBirdPegasusForSequenceClassification"),
         ("BigBirdConfig", "BigBirdForSequenceClassification"),
@@ -222,6 +225,7 @@ MODEL_FOR_TABLE_QUESTION_ANSWERING_MAPPING_NAMES = OrderedDict(
 
 MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING_NAMES = OrderedDict(
     [
+        ("CanineConfig", "CanineForTokenClassification"),
         ("RoFormerConfig", "RoFormerForTokenClassification"),
         ("BigBirdConfig", "BigBirdForTokenClassification"),
         ("ConvBertConfig", "ConvBertForTokenClassification"),
@@ -252,6 +256,7 @@ MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING_NAMES = OrderedDict(
 MODEL_MAPPING_NAMES = OrderedDict(
     [
         ("VisualBertConfig", "VisualBertModel"),
+        ("CanineConfig", "CanineModel"),
         ("RoFormerConfig", "RoFormerModel"),
         ("CLIPConfig", "CLIPModel"),
         ("BigBirdPegasusConfig", "BigBirdPegasusModel"),

tests/test_modeling_canine.py

+# coding=utf-8
+# Copyright 2021 The HuggingFace Inc. team. All rights reserved.
+#
+# 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.
+""" Testing suite for the PyTorch CANINE model. """
+
+
+import unittest
+from typing import List, Tuple
+
+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, _config_zero_init, global_rng, ids_tensor, random_attention_mask
+
+
+if is_torch_available():
+    import torch
+
+    from transformers import (
+        CanineConfig,
+        CanineForMultipleChoice,
+        CanineForQuestionAnswering,
+        CanineForSequenceClassification,
+        CanineForTokenClassification,
+        CanineModel,
+    )
+    from transformers.models.canine.modeling_canine import CANINE_PRETRAINED_MODEL_ARCHIVE_LIST
+
+
+class CanineModelTester:
+    def __init__(
+        self,
+        parent,
+        batch_size=13,
+        seq_length=7,
+        is_training=True,
+        use_input_mask=True,
+        use_token_type_ids=True,
+        use_labels=True,
+        hidden_size=32,
+        num_hidden_layers=5,
+        num_attention_heads=4,
+        intermediate_size=37,
+        hidden_act="gelu",
+        hidden_dropout_prob=0.1,
+        attention_probs_dropout_prob=0.1,
+        max_position_embeddings=512,
+        type_vocab_size=16,
+        type_sequence_label_size=2,
+        initializer_range=0.02,
+        num_labels=3,
+        num_choices=4,
+        scope=None,
+    ):
+        self.parent = parent
+        self.batch_size = batch_size
+        self.seq_length = seq_length
+        self.is_training = is_training
+        self.use_input_mask = use_input_mask
+        self.use_token_type_ids = use_token_type_ids
+        self.use_labels = use_labels
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.intermediate_size = intermediate_size
+        self.hidden_act = hidden_act
+        self.hidden_dropout_prob = hidden_dropout_prob
+        self.attention_probs_dropout_prob = attention_probs_dropout_prob
+        self.max_position_embeddings = max_position_embeddings
+        self.type_vocab_size = type_vocab_size
+        self.type_sequence_label_size = type_sequence_label_size
+        self.initializer_range = initializer_range
+        self.num_labels = num_labels
+        self.num_choices = num_choices
+        self.scope = scope
+
+    def prepare_config_and_inputs(self):
+        # let's use a vocab size that's way bigger than BERT's one
+        input_ids = ids_tensor([self.batch_size, self.seq_length], 100000)
+
+        input_mask = None
+        if self.use_input_mask:
+            input_mask = random_attention_mask([self.batch_size, self.seq_length])
+
+        token_type_ids = None
+        if self.use_token_type_ids:
+            token_type_ids = ids_tensor(input_ids.shape, self.type_vocab_size)
+
+        sequence_labels = None
+        token_labels = None
+        choice_labels = None
+        if self.use_labels:
+            sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
+            token_labels = ids_tensor([self.batch_size, self.seq_length], self.num_labels)
+            choice_labels = ids_tensor([self.batch_size], self.num_choices)
+
+        config = CanineConfig(
+            hidden_size=self.hidden_size,
+            num_hidden_layers=self.num_hidden_layers,
+            num_attention_heads=self.num_attention_heads,
+            intermediate_size=self.intermediate_size,
+            hidden_act=self.hidden_act,
+            hidden_dropout_prob=self.hidden_dropout_prob,
+            attention_probs_dropout_prob=self.attention_probs_dropout_prob,
+            max_position_embeddings=self.max_position_embeddings,
+            type_vocab_size=self.type_vocab_size,
+            is_decoder=False,
+            initializer_range=self.initializer_range,
+        )
+
+        return config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+
+    def create_and_check_model(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = CanineModel(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids)
+        result = model(input_ids, token_type_ids=token_type_ids)
+        result = model(input_ids)
+        self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
+
+    def create_and_check_for_question_answering(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = CanineForQuestionAnswering(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_ids,
+            attention_mask=input_mask,
+            token_type_ids=token_type_ids,
+            start_positions=sequence_labels,
+            end_positions=sequence_labels,
+        )
+        self.parent.assertEqual(result.start_logits.shape, (self.batch_size, self.seq_length))
+        self.parent.assertEqual(result.end_logits.shape, (self.batch_size, self.seq_length))
+
+    def create_and_check_for_sequence_classification(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        config.num_labels = self.num_labels
+        model = CanineForSequenceClassification(config)
+        model.to(torch_device)
+        model.eval()
+        result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=sequence_labels)
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels))
+
+    def create_and_check_for_token_classification(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        config.num_labels = self.num_labels
+        model = CanineForTokenClassification(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=token_labels)
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.num_labels))
+
+    def create_and_check_for_multiple_choice(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        config.num_choices = self.num_choices
+        model = CanineForMultipleChoice(config=config)
+        model.to(torch_device)
+        model.eval()
+        multiple_choice_inputs_ids = input_ids.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
+        multiple_choice_token_type_ids = token_type_ids.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
+        multiple_choice_input_mask = input_mask.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
+        result = model(
+            multiple_choice_inputs_ids,
+            attention_mask=multiple_choice_input_mask,
+            token_type_ids=multiple_choice_token_type_ids,
+            labels=choice_labels,
+        )
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_choices))
+
+    def prepare_config_and_inputs_for_common(self):
+        config_and_inputs = self.prepare_config_and_inputs()
+        (
+            config,
+            input_ids,
+            token_type_ids,
+            input_mask,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+        ) = config_and_inputs
+        inputs_dict = {"input_ids": input_ids, "token_type_ids": token_type_ids, "attention_mask": input_mask}
+        return config, inputs_dict
+
+
+@require_torch
+class CanineModelTest(ModelTesterMixin, unittest.TestCase):
+
+    all_model_classes = (
+        (
+            CanineModel,
+            CanineForMultipleChoice,
+            CanineForQuestionAnswering,
+            CanineForSequenceClassification,
+            CanineForTokenClassification,
+        )
+        if is_torch_available()
+        else ()
+    )
+
+    test_torchscript = False
+    test_resize_embeddings = False
+    test_pruning = False
+
+    def setUp(self):
+        self.model_tester = CanineModelTester(self)
+        # we set has_text_modality to False as the config has no vocab_size attribute
+        self.config_tester = ConfigTester(self, config_class=CanineConfig, has_text_modality=False, hidden_size=37)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    def test_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_model(*config_and_inputs)
+
+    def test_for_multiple_choice(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_for_multiple_choice(*config_and_inputs)
+
+    def test_for_question_answering(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_for_question_answering(*config_and_inputs)
+
+    def test_for_sequence_classification(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_for_sequence_classification(*config_and_inputs)
+
+    def test_for_token_classification(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_for_token_classification(*config_and_inputs)
+
+    def test_hidden_states_output(self):
+        def check_hidden_states_output(inputs_dict, config, model_class):
+            model = model_class(config)
+            model.to(torch_device)
+            model.eval()
+
+            with torch.no_grad():
+                outputs = model(**self._prepare_for_class(inputs_dict, model_class))
+
+            hidden_states = outputs.hidden_states
+            # expected_num_layers equals num_hidden_layers of the deep encoder + 1, + 2 for the first shallow encoder, + 2
+            # for the final shallow encoder
+            expected_num_layers = self.model_tester.num_hidden_layers + 1 + 2 + 2
+            self.assertEqual(len(hidden_states), expected_num_layers)
+
+            seq_length = self.model_tester.seq_length
+            for i in range(expected_num_layers):
+                if (i < 2) or ((expected_num_layers - i) < 3):
+                    # the expected length of the hidden_states of the first and final shallow encoders
+                    # is equal to the seq_length
+                    self.assertListEqual(
+                        list(hidden_states[i].shape[-2:]),
+                        [seq_length, self.model_tester.hidden_size],
+                    )
+                else:
+                    # the expected length of the hidden_states of the deep encoder need to be updated
+                    # for CANINE since the seq length is downsampled
+                    self.assertListEqual(
+                        list(hidden_states[i].shape[-2:]),
+                        [seq_length // config.downsampling_rate, self.model_tester.hidden_size],
+                    )
+
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+
+        for model_class in self.all_model_classes:
+            inputs_dict["output_hidden_states"] = True
+            check_hidden_states_output(inputs_dict, config, model_class)
+
+            # check that output_hidden_states also work using config
+            del inputs_dict["output_hidden_states"]
+            config.output_hidden_states = True
+
+            check_hidden_states_output(inputs_dict, config, model_class)
+
+    def test_attention_outputs(self):
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+        config.return_dict = True
+
+        seq_len = getattr(self.model_tester, "seq_length", None)
+
+        for model_class in self.all_model_classes:
+            inputs_dict["output_attentions"] = True
+            inputs_dict["output_hidden_states"] = False
+            config.return_dict = True
+            model = model_class(config)
+            model.to(torch_device)
+            model.eval()
+            with torch.no_grad():
+                outputs = model(**self._prepare_for_class(inputs_dict, model_class))
+            attentions = outputs.attentions
+            # we add + 2 due to the 2 shallow encoders
+            self.assertEqual(len(attentions), self.model_tester.num_hidden_layers + 2)
+
+            # check that output_attentions also work using config
+            del inputs_dict["output_attentions"]
+            config.output_attentions = True
+            model = model_class(config)
+            model.to(torch_device)
+            model.eval()
+            with torch.no_grad():
+                outputs = model(**self._prepare_for_class(inputs_dict, model_class))
+            attentions = outputs.attentions
+            # we add + 2 due to the 2 shallow encoders
+            self.assertEqual(len(attentions), self.model_tester.num_hidden_layers + 2)
+
+            self.assertListEqual(
+                list(attentions[0].shape[-3:]),
+                [self.model_tester.num_attention_heads, seq_len, seq_len],
+            )
+            out_len = len(outputs)
+
+            # Check attention is always last and order is fine
+            inputs_dict["output_attentions"] = True
+            inputs_dict["output_hidden_states"] = True
+            model = model_class(config)
+            model.to(torch_device)
+            model.eval()
+            with torch.no_grad():
+                outputs = model(**self._prepare_for_class(inputs_dict, model_class))
+
+            if hasattr(self.model_tester, "num_hidden_states_types"):
+                added_hidden_states = self.model_tester.num_hidden_states_types
+            else:
+                added_hidden_states = 1
+            self.assertEqual(out_len + added_hidden_states, len(outputs))
+
+            self_attentions = outputs.attentions
+
+            self.assertEqual(len(self_attentions), self.model_tester.num_hidden_layers + 2)
+            self.assertListEqual(
+                list(self_attentions[0].shape[-3:]),
+                [self.model_tester.num_attention_heads, seq_len, seq_len],
+            )
+
+    def test_model_outputs_equivalence(self):
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+
+        def set_nan_tensor_to_zero(t):
+            t[t != t] = 0
+            return t
+
+        def check_equivalence(model, tuple_inputs, dict_inputs, additional_kwargs={}):
+            with torch.no_grad():
+                tuple_output = model(**tuple_inputs, return_dict=False, **additional_kwargs)
+                dict_output = model(**dict_inputs, return_dict=True, **additional_kwargs).to_tuple()
+
+                def recursive_check(tuple_object, dict_object):
+                    if isinstance(tuple_object, (List, Tuple)):
+                        for tuple_iterable_value, dict_iterable_value in zip(tuple_object, dict_object):
+                            recursive_check(tuple_iterable_value, dict_iterable_value)
+                    elif tuple_object is None:
+                        return
+                    else:
+                        self.assertTrue(
+                            torch.allclose(
+                                set_nan_tensor_to_zero(tuple_object), set_nan_tensor_to_zero(dict_object), atol=1e-5
+                            ),
+                            msg=f"Tuple and dict output are not equal. Difference: {torch.max(torch.abs(tuple_object - dict_object))}. Tuple has `nan`: {torch.isnan(tuple_object).any()} and `inf`: {torch.isinf(tuple_object)}. Dict has `nan`: {torch.isnan(dict_object).any()} and `inf`: {torch.isinf(dict_object)}.",
+                        )
+
+                recursive_check(tuple_output, dict_output)
+
+        for model_class in self.all_model_classes:
+            print(model_class)
+            model = model_class(config)
+            model.to(torch_device)
+            model.eval()
+
+            tuple_inputs = self._prepare_for_class(inputs_dict, model_class)
+            dict_inputs = self._prepare_for_class(inputs_dict, model_class)
+            check_equivalence(model, tuple_inputs, dict_inputs)
+
+            tuple_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
+            dict_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
+            check_equivalence(model, tuple_inputs, dict_inputs)
+
+            tuple_inputs = self._prepare_for_class(inputs_dict, model_class)
+            dict_inputs = self._prepare_for_class(inputs_dict, model_class)
+            check_equivalence(model, tuple_inputs, dict_inputs, {"output_hidden_states": True})
+
+            tuple_inputs = self._prepare_for_class(inputs_dict, model_class)
+            dict_inputs = self._prepare_for_class(inputs_dict, model_class)
+            check_equivalence(model, tuple_inputs, dict_inputs, {"output_attentions": True})
+
+            tuple_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
+            dict_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
+            check_equivalence(model, tuple_inputs, dict_inputs, {"output_hidden_states": True})
+
+            tuple_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
+            dict_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
+            check_equivalence(model, tuple_inputs, dict_inputs, {"output_attentions": True})
+
+            tuple_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
+            dict_inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
+            check_equivalence(
+                model, tuple_inputs, dict_inputs, {"output_hidden_states": True, "output_attentions": True}
+            )
+
+    def test_headmasking(self):
+        if not self.test_head_masking:
+            return
+
+        global_rng.seed(42)
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+        global_rng.seed()
+
+        inputs_dict["output_attentions"] = True
+        config.output_hidden_states = True
+        configs_no_init = _config_zero_init(config)  # To be sure we have no Nan
+        for model_class in self.all_model_classes:
+            model = model_class(config=configs_no_init)
+            model.to(torch_device)
+            model.eval()
+
+            # Prepare head_mask
+            # Set require_grad after having prepared the tensor to avoid error (leaf variable has been moved into the graph interior)
+            head_mask = torch.ones(
+                self.model_tester.num_hidden_layers,
+                self.model_tester.num_attention_heads,
+                device=torch_device,
+            )
+            head_mask[0, 0] = 0
+            head_mask[-1, :-1] = 0
+            head_mask.requires_grad_(requires_grad=True)
+            inputs = self._prepare_for_class(inputs_dict, model_class).copy()
+            inputs["head_mask"] = head_mask
+
+            outputs = model(**inputs, return_dict=True)
+
+            # Test that we can get a gradient back for importance score computation
+            output = sum(t.sum() for t in outputs[0])
+            output = output.sum()
+            output.backward()
+            multihead_outputs = head_mask.grad
+
+            self.assertIsNotNone(multihead_outputs)
+            self.assertEqual(len(multihead_outputs), self.model_tester.num_hidden_layers)
+
+            def check_attentions_validity(attentions):
+                # Remove Nan
+                for t in attentions:
+                    self.assertLess(
+                        torch.sum(torch.isnan(t)), t.numel() / 4
+                    )  # Check we don't have more than 25% nans (arbitrary)
+                attentions = [
+                    t.masked_fill(torch.isnan(t), 0.0) for t in attentions
+                ]  # remove them (the test is less complete)
+
+                self.assertAlmostEqual(attentions[1][..., 0, :, :].flatten().sum().item(), 0.0)
+                self.assertNotEqual(attentions[1][..., -1, :, :].flatten().sum().item(), 0.0)
+                self.assertAlmostEqual(attentions[-2][..., -2, :, :].flatten().sum().item(), 0.0)
+                self.assertNotEqual(attentions[-2][..., -1, :, :].flatten().sum().item(), 0.0)
+
+            check_attentions_validity(outputs.attentions)
+
+    @unittest.skip("CANINE does not have a get_input_embeddings() method.")
+    def test_inputs_embeds(self):
+        # ViT does not use inputs_embeds
+        pass
+
+    @unittest.skip("CANINE does not have a get_input_embeddings() method.")
+    def test_model_common_attributes(self):
+        pass
+
+    @slow
+    def test_model_from_pretrained(self):
+        for model_name in CANINE_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
+            model = CanineModel.from_pretrained(model_name)
+            self.assertIsNotNone(model)
+
+
+@require_torch
+class CanineModelIntegrationTest(unittest.TestCase):
+    @slow
+    def test_inference_no_head(self):
+        # TODO replace nielsr by google
+        model = CanineModel.from_pretrained("nielsr/canine-s")
+        # this one corresponds to the first example of the TydiQA dev set (in Swahili)
+        # fmt: off
+        input_ids = [57344, 57349, 85, 107, 117, 98, 119, 97, 32, 119, 97, 32, 82, 105, 106, 105, 108, 105, 32, 75, 97, 110, 116, 111, 114, 105, 32, 110, 105, 32, 107, 105, 97, 115, 105, 32, 103, 97, 110, 105, 63, 57345, 57350, 32, 82, 105, 106, 105, 108, 105, 32, 75, 97, 110, 116, 111, 114, 105, 32, 44, 32, 82, 105, 106, 105, 108, 105, 32, 75, 97, 110, 116, 97, 114, 117, 115, 105, 32, 97, 117, 32, 105, 110, 103, 46, 32, 65, 108, 112, 104, 97, 32, 67, 101, 110, 116, 97, 117, 114, 105, 32, 40, 112, 105, 97, 58, 32, 84, 111, 108, 105, 109, 97, 110, 32, 97, 117, 32, 82, 105, 103, 105, 108, 32, 75, 101, 110, 116, 97, 117, 114, 117, 115, 41, 32, 110, 105, 32, 110, 121, 111, 116, 97, 32, 105, 110, 97, 121, 111, 110, 103, 39, 97, 97, 32, 115, 97, 110, 97, 32, 107, 97, 116, 105, 107, 97, 32, 97, 110, 103, 97, 32, 121, 97, 32, 107, 117, 115, 105, 110, 105, 32, 107, 119, 101, 110, 121, 101, 32, 107, 117, 110, 100, 105, 110, 121, 111, 116, 97, 32, 121, 97, 32, 75, 97, 110, 116, 97, 114, 117, 115, 105, 32, 40, 112, 105, 97, 58, 32, 105, 110, 103, 46, 32, 67, 101, 110, 116, 97, 117, 114, 117, 115, 41, 46, 32, 78, 105, 32, 110, 121, 111, 116, 97, 32, 121, 97, 32, 107, 117, 110, 103, 97, 97, 32, 115, 97, 110, 97, 32, 121, 97, 32, 110, 110, 101, 32, 97, 110, 103, 97, 110, 105, 32, 108, 97, 107, 105, 110, 105, 32, 104, 97, 105, 111, 110, 101, 107, 97, 110, 105, 32, 107, 119, 101, 110, 121, 101, 32, 110, 117, 115, 117, 100, 117, 110, 105, 97, 32, 121, 97, 32, 107, 97, 115, 107, 97, 122, 105, 110, 105, 46, 32, 57351, 32, 65, 108, 112, 104, 97, 32, 67, 101, 110, 116, 97, 117, 114, 105, 32, 110, 105, 32, 110, 121, 111, 116, 97, 32, 121, 97, 32, 112, 101, 107, 101, 101, 32, 107, 119, 97, 32, 115, 97, 98, 97, 98, 117, 32, 110, 105, 32, 110, 121, 111, 116, 97, 32, 121, 101, 116, 117, 32, 106, 105, 114, 97, 110, 105, 32, 107, 97, 116, 105, 107, 97, 32, 97, 110, 103, 97, 32, 105, 110, 97, 32, 117, 109, 98, 97, 108, 105, 32, 119, 97, 32, 109, 105, 97, 107, 97, 32, 121, 97, 32, 110, 117, 114, 117, 32, 52, 46, 50, 46, 32, 73, 110, 97, 111, 110, 101, 107, 97, 110, 97, 32, 97, 110, 103, 97, 110, 105, 32, 107, 97, 114, 105, 98, 117, 32, 110, 97, 32, 107, 117, 110, 100, 105, 110, 121, 111, 116, 97, 32, 121, 97, 32, 83, 97, 108, 105, 98, 117, 32, 40, 67, 114, 117, 120, 41, 46, 32, 57352, 32, 82, 105, 106, 105, 108, 105, 32, 75, 97, 110, 116, 97, 114, 117, 115, 105, 32, 40, 65, 108, 112, 104, 97, 32, 67, 101, 110, 116, 97, 117, 114, 105, 41, 32, 105, 110, 97, 111, 110, 101, 107, 97, 110, 97, 32, 107, 97, 109, 97, 32, 110, 121, 111, 116, 97, 32, 109, 111, 106, 97, 32, 108, 97, 107, 105, 110, 105, 32, 107, 119, 97, 32, 100, 97, 114, 117, 98, 105, 110, 105, 32, 107, 117, 98, 119, 97, 32, 105, 110, 97, 111, 110, 101, 107, 97, 110, 97, 32, 107, 117, 119, 97, 32, 109, 102, 117, 109, 111, 32, 119, 97, 32, 110, 121, 111, 116, 97, 32, 116, 97, 116, 117, 32, 122, 105, 110, 97, 122, 111, 107, 97, 97, 32, 107, 97, 114, 105, 98, 117, 32, 110, 97, 32, 107, 117, 115, 104, 105, 107, 97, 109, 97, 110, 97, 32, 107, 97, 116, 105, 32, 121, 97, 111, 46, 32, 78, 121, 111, 116, 97, 32, 109, 97, 112, 97, 99, 104, 97, 32, 122, 97, 32, 65, 108, 112, 104, 97, 32, 67, 101, 110, 116, 97, 117, 114, 105, 32, 65, 32, 110, 97, 32, 65, 108, 112, 104, 97, 32, 67, 101, 110, 116, 97, 117, 114, 105, 32, 66, 32, 122, 105, 107, 111, 32, 109, 105, 97, 107, 97, 32, 121, 97, 32, 110, 117, 114, 117, 32, 52, 46, 51, 54, 32, 107, 117, 116, 111, 107, 97, 32, 107, 119, 101, 116, 117, 32, 110, 97, 32, 110, 121, 111, 116, 97, 32, 121, 97, 32, 116, 97, 116, 117, 32, 65, 108, 112, 104, 97, 32, 67, 101, 110, 116, 97, 117, 114, 105, 32, 67, 32, 97, 117, 32, 80, 114, 111, 120, 105, 109, 97, 32, 67, 101, 110, 116, 97, 117, 114, 105, 32, 105, 110, 97, 32, 117, 109, 98, 97, 108, 105, 32, 119, 97, 32, 109, 105, 97, 107, 97, 32, 121, 97, 32, 110, 117, 114, 117, 32, 52, 46, 50, 50, 46, 32, 57353, 32, 80, 114, 111, 120, 105, 109, 97, 32, 67, 101, 110, 116, 97, 117, 114, 105, 32, 40, 121, 97, 97, 110, 105, 32, 110, 121, 111, 116, 97, 32, 121, 97, 32, 75, 97, 110, 116, 97, 114, 117, 115, 105, 32, 105, 108, 105, 121, 111, 32, 107, 97, 114, 105, 98, 117, 32, 122, 97, 105, 100, 105, 32, 110, 97, 115, 105, 41, 32, 105, 109, 101, 103, 117, 110, 100, 117, 108, 105, 119, 97, 32, 107, 117, 119, 97, 32, 110, 97, 32, 115, 97, 121, 97, 114, 105, 32, 109, 111, 106, 97, 46, 32, 86, 105, 112, 105, 109, 111, 32, 118, 105, 110, 97, 118, 121, 111, 112, 97, 116, 105, 107, 97, 110, 97, 32, 104, 97, 100, 105, 32, 115, 97, 115, 97, 32, 122, 105, 110, 97, 111, 110, 121, 101, 115, 104, 97, 32, 117, 119, 101, 122, 101, 107, 97, 110, 111, 32, 109, 107, 117, 98, 119, 97, 32, 121, 97, 32, 107, 119, 97, 109, 98, 97, 32, 115, 97, 121, 97, 114, 105, 32, 104, 105, 105, 32, 110, 105, 32, 121, 97, 32, 109, 119, 97, 109, 98, 97, 32, 40, 107, 97, 109, 97, 32, 100, 117, 110, 105, 97, 32, 121, 101, 116, 117, 44, 32, 77, 105, 114, 105, 104, 105, 32, 97, 117, 32, 90, 117, 104, 117, 114, 97, 41, 32, 110, 97, 32, 105, 110, 97, 119, 101, 122, 97, 32, 107, 117, 119, 97, 32, 110, 97, 32, 97, 110, 103, 97, 104, 101, 119, 97, 44, 32, 116, 101, 110, 97, 32, 107, 97, 116, 105, 107, 97, 32, 117, 112, 101, 111, 32, 119, 97, 32, 106, 111, 116, 111, 32, 117, 110, 97, 111, 114, 117, 104, 117, 115, 117, 32, 107, 117, 119, 101, 112, 111, 32, 107, 119, 97, 32, 117, 104, 97, 105, 46, 32, 91, 49, 93, 57345, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
+        attention_mask = [1 if x != 0 else 0 for x in input_ids]
+        token_type_ids = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
+        # fmt: on
+        input_ids = torch.tensor([input_ids])
+        attention_mask = torch.tensor([attention_mask])
+        token_type_ids = torch.tensor([token_type_ids])
+        outputs = model(input_ids, attention_mask, token_type_ids)
+
+        # verify sequence output
+        expected_shape = torch.Size((1, 2048, 768))
+        self.assertEqual(outputs.last_hidden_state.shape, expected_shape)
+
+        expected_slice = torch.tensor(
+            [
+                [-0.161433131, 0.395568609, 0.0407391489],
+                [-0.108025983, 0.362060368, -0.544592619],
+                [-0.141537309, 0.180541009, 0.076907],
+            ]
+        )
+
+        self.assertTrue(torch.allclose(outputs.last_hidden_state[0, :3, :3], expected_slice, atol=1e-2))
+
+        # verify pooled output
+        expected_shape = torch.Size((1, 768))
+        self.assertEqual(outputs.pooler_output.shape, expected_shape)
+
+        expected_slice = torch.tensor([-0.884311497, -0.529064834, 0.723164916])
+
+        self.assertTrue(torch.allclose(outputs.pooler_output[0, :3], expected_slice, atol=1e-2))

tests/test_tokenization_canine.py

+# coding=utf-8
+# Copyright 2021 Google AI and 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 shutil
+import tempfile
+import unittest
+
+from transformers import BatchEncoding, CanineTokenizer
+from transformers.file_utils import cached_property
+from transformers.testing_utils import require_tokenizers, require_torch
+from transformers.tokenization_utils import AddedToken
+
+from .test_tokenization_common import TokenizerTesterMixin
+
+
+class CanineTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
+
+    tokenizer_class = CanineTokenizer
+    test_rust_tokenizer = False
+
+    def setUp(self):
+        super().setUp()
+        tokenizer = CanineTokenizer()
+        tokenizer.save_pretrained(self.tmpdirname)
+
+    @cached_property
+    def canine_tokenizer(self):
+        # TODO replace nielsr by google
+        return CanineTokenizer.from_pretrained("nielsr/canine-s")
+
+    def get_tokenizer(self, **kwargs) -> CanineTokenizer:
+        return self.tokenizer_class.from_pretrained(self.tmpdirname, **kwargs)
+
+    @require_torch
+    def test_prepare_batch_integration(self):
+        tokenizer = self.canine_tokenizer
+        src_text = ["Life is like a box of chocolates.", "You never know what you're gonna get."]
+        # fmt: off
+        expected_src_tokens = [57344, 76, 105, 102, 101, 32, 105, 115, 32, 108, 105, 107, 101, 32, 97, 32, 98, 111, 120, 32, 111, 102, 32, 99, 104, 111, 99, 111, 108, 97, 116, 101, 115, 46, 57345, 0, 0, 0, 0]
+        # fmt: on
+        batch = tokenizer(src_text, padding=True, return_tensors="pt")
+        self.assertIsInstance(batch, BatchEncoding)
+
+        result = list(batch.input_ids.numpy()[0])
+
+        self.assertListEqual(expected_src_tokens, result)
+
+        self.assertEqual((2, 39), batch.input_ids.shape)
+        self.assertEqual((2, 39), batch.attention_mask.shape)
+
+    @require_torch
+    def test_encoding_keys(self):
+        tokenizer = self.canine_tokenizer
+        src_text = ["Once there was a man.", "He wrote a test in HuggingFace Tranformers."]
+        batch = tokenizer(src_text, padding=True, return_tensors="pt")
+        # check if input_ids, attention_mask and token_type_ids are returned
+        self.assertIn("input_ids", batch)
+        self.assertIn("attention_mask", batch)
+        self.assertIn("token_type_ids", batch)
+
+    @require_torch
+    def test_max_length_integration(self):
+        tokenizer = self.canine_tokenizer
+        tgt_text = [
+            "What's the weater?",
+            "It's about 25 degrees.",
+        ]
+        with tokenizer.as_target_tokenizer():
+            targets = tokenizer(tgt_text, max_length=32, padding="max_length", truncation=True, return_tensors="pt")
+        self.assertEqual(32, targets["input_ids"].shape[1])
+
+    # cannot use default save_and_load_tokenzier test method because tokenzier has no vocab
+    def test_save_and_load_tokenizer(self):
+        # safety check on max_len default value so we are sure the test works
+        tokenizers = self.get_tokenizers()
+        for tokenizer in tokenizers:
+            with self.subTest(f"{tokenizer.__class__.__name__}"):
+                self.assertNotEqual(tokenizer.model_max_length, 42)
+
+        # Now let's start the test
+        tokenizers = self.get_tokenizers()
+        for tokenizer in tokenizers:
+            with self.subTest(f"{tokenizer.__class__.__name__}"):
+                # Isolate this from the other tests because we save additional tokens/etc
+                tmpdirname = tempfile.mkdtemp()
+
+                sample_text = " He is very happy, UNwant\u00E9d,running"
+                before_tokens = tokenizer.encode(sample_text, add_special_tokens=False)
+                tokenizer.save_pretrained(tmpdirname)
+
+                after_tokenizer = tokenizer.__class__.from_pretrained(tmpdirname)
+                after_tokens = after_tokenizer.encode(sample_text, add_special_tokens=False)
+                self.assertListEqual(before_tokens, after_tokens)
+
+                shutil.rmtree(tmpdirname)
+
+        tokenizers = self.get_tokenizers(model_max_length=42)
+        for tokenizer in tokenizers:
+            with self.subTest(f"{tokenizer.__class__.__name__}"):
+                # Isolate this from the other tests because we save additional tokens/etc
+                tmpdirname = tempfile.mkdtemp()
+
+                sample_text = " He is very happy, UNwant\u00E9d,running"
+
+                additional_special_tokens = tokenizer.additional_special_tokens
+
+                # We can add a new special token for Canine as follows:
+                new_additional_special_token = chr(0xE007)
+                additional_special_tokens.append(new_additional_special_token)
+                tokenizer.add_special_tokens({"additional_special_tokens": additional_special_tokens})
+                before_tokens = tokenizer.encode(sample_text, add_special_tokens=False)
+                tokenizer.save_pretrained(tmpdirname)
+
+                after_tokenizer = tokenizer.__class__.from_pretrained(tmpdirname)
+                after_tokens = after_tokenizer.encode(sample_text, add_special_tokens=False)
+                self.assertListEqual(before_tokens, after_tokens)
+                self.assertIn(new_additional_special_token, after_tokenizer.additional_special_tokens)
+                self.assertEqual(after_tokenizer.model_max_length, 42)
+
+                tokenizer = tokenizer.__class__.from_pretrained(tmpdirname, model_max_length=43)
+                self.assertEqual(tokenizer.model_max_length, 43)
+
+                shutil.rmtree(tmpdirname)
+
+    def test_add_special_tokens(self):
+        tokenizers = self.get_tokenizers(do_lower_case=False)
+        for tokenizer in tokenizers:
+            with self.subTest(f"{tokenizer.__class__.__name__}"):
+                input_text, ids = self.get_clean_sequence(tokenizer)
+
+                # a special token for Canine can be defined as follows:
+                SPECIAL_TOKEN = 0xE005
+                special_token = chr(SPECIAL_TOKEN)
+
+                tokenizer.add_special_tokens({"cls_token": special_token})
+                encoded_special_token = tokenizer.encode(special_token, add_special_tokens=False)
+                self.assertEqual(len(encoded_special_token), 1)
+
+                text = tokenizer.decode(ids + encoded_special_token, clean_up_tokenization_spaces=False)
+                encoded = tokenizer.encode(text, add_special_tokens=False)
+
+                input_encoded = tokenizer.encode(input_text, add_special_tokens=False)
+                special_token_id = tokenizer.encode(special_token, add_special_tokens=False)
+                self.assertEqual(encoded, input_encoded + special_token_id)
+
+                decoded = tokenizer.decode(encoded, skip_special_tokens=True)
+                self.assertTrue(special_token not in decoded)
+
+    @require_tokenizers
+    def test_added_token_serializable(self):
+        tokenizers = self.get_tokenizers(do_lower_case=False)
+        for tokenizer in tokenizers:
+            with self.subTest(f"{tokenizer.__class__.__name__}"):
+
+                # a special token for Canine can be defined as follows:
+                NEW_TOKEN = 0xE006
+                new_token = chr(NEW_TOKEN)
+
+                new_token = AddedToken(new_token, lstrip=True)
+                tokenizer.add_special_tokens({"additional_special_tokens": [new_token]})
+
+                with tempfile.TemporaryDirectory() as tmp_dir_name:
+                    tokenizer.save_pretrained(tmp_dir_name)
+                    tokenizer.from_pretrained(tmp_dir_name)
+
+    @require_tokenizers
+    def test_encode_decode_with_spaces(self):
+        tokenizers = self.get_tokenizers(do_lower_case=False)
+        for tokenizer in tokenizers:
+            with self.subTest(f"{tokenizer.__class__.__name__}"):
+
+                input = "hello world"
+                if self.space_between_special_tokens:
+                    output = "[CLS] hello world [SEP]"
+                else:
+                    output = input
+                encoded = tokenizer.encode(input, add_special_tokens=False)
+                decoded = tokenizer.decode(encoded, spaces_between_special_tokens=self.space_between_special_tokens)
+                self.assertIn(decoded, [output, output.lower()])
+
+    # tokenizer has a fixed vocab_size (namely all possible unicode code points)
+    def test_add_tokens_tokenizer(self):
+        pass
+
+    # CanineTokenizer does not support do_lower_case = True, as each character has its own Unicode code point
+    # ("b" and "B" for example have different Unicode code points)
+    def test_added_tokens_do_lower_case(self):
+        pass
+
+    # CanineModel does not support the get_input_embeddings nor the get_vocab method
+    def test_np_encode_plus_sent_to_model(self):
+        pass
+
+    # CanineModel does not support the get_input_embeddings nor the get_vocab method
+    def test_torch_encode_plus_sent_to_model(self):
+        pass
+
+    # tokenizer can be instantiated without any pretrained files, so no need for pretrained tokenizer list
+    def test_pretrained_model_lists(self):
+        pass
+
+    # tokenizer does not have vocabulary
+    def test_get_vocab(self):
+        pass
+
+    # inputs cannot be pretokenized since ids depend on whole input string and not just on single characters
+    def test_pretokenized_inputs(self):
+        pass
+
+    # tests all ids in vocab => vocab doesn't exist so unnecessary to test
+    def test_conversion_reversible(self):
+        pass