Implement Roberta PreLayerNorm (#20305)

* Copy RoBERTa

* formatting

* implement RoBERTa with prelayer normalization

* update test expectations

* add documentation

* add convertion script for DinkyTrain weights

* update checkpoint repo

Unfortunately the original checkpoints assumes a hacked roberta model

* add to RoBERTa-PreLayerNorm docs to toc

* run utils/check_copies.py

* lint files

* remove unused import

* fix check_repo reporting wrongly a test is missing

* fix import error, caused by rebase

* run make fix-copies

* add RobertaPreLayerNormConfig to ROBERTA_EMBEDDING_ADJUSMENT_CONFIGS

* Fix documentation <Facebook> -> Facebook
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* fixup: Fix documentation <Facebook> -> Facebook
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* Add missing Flax header
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* expected_slice -> EXPECTED_SLICE
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* update copies after rebase

* add missing copied from statements

* make fix-copies

* make prelayernorm explicit in code

* fix checkpoint path for the original implementation

* add flax integration tests

* improve docs

* update utils/documentation_tests.txt

* lint files

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

* make fix-copies

* Remove EXPECTED_SLICE calculation comments
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

README.md

@@ -377,7 +377,8 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/abs/2010.12821) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
 1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
 1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (from Facebook), released together with the paper [RoBERTa: 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.
-1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
+1. **[RoBERTa-PreLayerNorm](https://huggingface.co/docs/transformers/main/model_doc/roberta-prelayernorm)** (from Facebook) released with the paper [fairseq: A Fast, Extensible Toolkit for Sequence Modeling](https://arxiv.org/abs/1904.01038) by Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, Sam Gross, Nathan Ng, David Grangier, Michael Auli.
+1. **[RoCBert](https://huggingface.co/docs/transformers/main/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
 1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/abs/2104.09864) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.

README_es.md

@@ -282,7 +282,7 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 1. **[ByT5](https://huggingface.co/docs/transformers/model_doc/byt5)** (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/docs/transformers/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.
 1. **[CANINE](https://huggingface.co/docs/transformers/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.
-1. **[Chinese-CLIP](https://huggingface.co/docs/transformers/model_doc/chinese_clip)** (from OFA-Sys) released with the paper [Chinese CLIP: Contrastive Vision-Language Pretraining in Chinese](https://arxiv.org/abs/2211.01335) by An Yang, Junshu Pan, Junyang Lin, Rui Men, Yichang Zhang, Jingren Zhou, Chang Zhou. 
+1. **[Chinese-CLIP](https://huggingface.co/docs/transformers/model_doc/chinese_clip)** (from OFA-Sys) released with the paper [Chinese CLIP: Contrastive Vision-Language Pretraining in Chinese](https://arxiv.org/abs/2211.01335) by An Yang, Junshu Pan, Junyang Lin, Rui Men, Yichang Zhang, Jingren Zhou, Chang Zhou.
 1. **[CLIP](https://huggingface.co/docs/transformers/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.
 1. **[CLIPSeg](https://huggingface.co/docs/transformers/model_doc/clipseg)** (from University of Göttingen) released with the paper [Image Segmentation Using Text and Image Prompts](https://arxiv.org/abs/2112.10003) by Timo Lüddecke and Alexander Ecker.
 1. **[CodeGen](https://huggingface.co/docs/transformers/model_doc/codegen)** (from Salesforce) released with the paper [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) by Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong.
@@ -377,7 +377,8 @@ Número actual de puntos de control: ![](https://img.shields.io/endpoint?url=htt
 1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/abs/2010.12821) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
 1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
 1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (from Facebook), released together with the paper [RoBERTa: 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.
-1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
+1. **[RoBERTa-PreLayerNorm](https://huggingface.co/docs/transformers/main/model_doc/roberta-prelayernorm)** (from Facebook) released with the paper [fairseq: A Fast, Extensible Toolkit for Sequence Modeling](https://arxiv.org/abs/1904.01038) by Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, Sam Gross, Nathan Ng, David Grangier, Michael Auli.
+1. **[RoCBert](https://huggingface.co/docs/transformers/main/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
 1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/abs/2104.09864) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.

README_hd.md

@@ -350,7 +350,8 @@ conda install -c huggingface transformers
 1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (गूगल रिसर्च से) साथ वाला पेपर [पूर्व-प्रशिक्षित भाषा मॉडल में एम्बेडिंग कपलिंग पर पुनर्विचार](https://arxiv .org/pdf/2010.12821.pdf) ह्युंग वोन चुंग, थिबॉल्ट फ़ेवरी, हेनरी त्साई, एम. जॉनसन, सेबेस्टियन रुडर द्वारा।
 1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (माइक्रोसॉफ्ट रिसर्च से) [डीप रेसिडुअल लर्निंग फॉर इमेज रिकग्निशन] (https://arxiv. org/abs/1512.03385) कैमिंग हे, जियांग्यु झांग, शाओकिंग रेन, जियान सन द्वारा।
 1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (फेसबुक से), साथ में कागज [मजबूत रूप से अनुकूलित BERT प्रीट्रेनिंग दृष्टिकोण](https://arxiv.org/abs /1907.11692) यिनहान लियू, मायल ओट, नमन गोयल, जिंगफेई डू, मंदार जोशी, डैनकी चेन, ओमर लेवी, माइक लुईस, ल्यूक ज़ेटलमॉयर, वेसेलिन स्टोयानोव द्वारा।
-1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
+1. **[RoBERTa-PreLayerNorm](https://huggingface.co/docs/transformers/main/model_doc/roberta-prelayernorm)** (from Facebook) released with the paper [fairseq: A Fast, Extensible Toolkit for Sequence Modeling](https://arxiv.org/abs/1904.01038) by Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, Sam Gross, Nathan Ng, David Grangier, Michael Auli.
+1. **[RoCBert](https://huggingface.co/docs/transformers/main/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
 1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (झुईई टेक्नोलॉजी से), साथ में पेपर [रोफॉर्मर: रोटरी पोजिशन एंबेडिंग के साथ एन्हांस्ड ट्रांसफॉर्मर] (https://arxiv.org/pdf/2104.09864v1.pdf) जियानलिन सु और यू लू और शेंगफेंग पैन और बो वेन और युनफेंग लियू द्वारा प्रकाशित।
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (ASAPP से) साथ देने वाला पेपर [भाषण पहचान के लिए अनसुपरवाइज्ड प्री-ट्रेनिंग में परफॉर्मेंस-एफिशिएंसी ट्रेड-ऑफ्स](https ://arxiv.org/abs/2109.06870) फेलिक्स वू, क्वांगयुन किम, जिंग पैन, क्यू हान, किलियन क्यू. वेनबर्गर, योव आर्टज़ी द्वारा।

README_ja.md

@@ -412,7 +412,8 @@ Flax、PyTorch、TensorFlowをcondaでインストールする方法は、それ
 1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/abs/2010.12821) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
 1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
 1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (from Facebook), released together with the paper [RoBERTa: 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.
-1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
+1. **[RoBERTa-PreLayerNorm](https://huggingface.co/docs/transformers/main/model_doc/roberta-prelayernorm)** (from Facebook) released with the paper [fairseq: A Fast, Extensible Toolkit for Sequence Modeling](https://arxiv.org/abs/1904.01038) by Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, Sam Gross, Nathan Ng, David Grangier, Michael Auli.
+1. **[RoCBert](https://huggingface.co/docs/transformers/main/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
 1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/abs/2104.09864) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.

README_ko.md

@@ -327,7 +327,8 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/pdf/2010.12821.pdf) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
 1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
 1. **[RoBERTa](https://huggingface.co/docs/transformers/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.
-1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
+1. **[RoBERTa-PreLayerNorm](https://huggingface.co/docs/transformers/main/model_doc/roberta-prelayernorm)** (from Facebook) released with the paper [fairseq: A Fast, Extensible Toolkit for Sequence Modeling](https://arxiv.org/abs/1904.01038) by Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, Sam Gross, Nathan Ng, David Grangier, Michael Auli.
+1. **[RoCBert](https://huggingface.co/docs/transformers/main/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
 1. **[RoFormer](https://huggingface.co/docs/transformers/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.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
@@ -347,7 +348,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (from Microsoft Research) released with the paper [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) by Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou.
 1. **[Time Series Transformer](https://huggingface.co/docs/transformers/model_doc/time_series_transformer)** (from HuggingFace).
 1. **[TimeSformer](https://huggingface.co/docs/transformers/main/model_doc/timesformer)** (from Facebook) released with the paper [Is Space-Time Attention All You Need for Video Understanding?](https://arxiv.org/abs/2102.05095) by Gedas Bertasius, Heng Wang, Lorenzo Torresani.
-1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine 
+1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
 1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
 1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (from Microsoft), released together with the paper [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) by Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei.
 1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler

README_zh-hans.md

@@ -351,7 +351,8 @@ conda install -c huggingface transformers
 1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (来自 Google Research) 伴随论文 [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/pdf/2010.12821.pdf) 由 Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder 发布。
 1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
 1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (来自 Facebook), 伴随论文 [Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) 由 Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov 发布。
-1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (来自 WeChatAI), 伴随论文 [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) 由 HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou 发布。
+1. **[RoBERTa-PreLayerNorm](https://huggingface.co/docs/transformers/main/model_doc/roberta-prelayernorm)** (来自 Facebook) 伴随论文 [fairseq: A Fast, Extensible Toolkit for Sequence Modeling](https://arxiv.org/abs/1904.01038) 由 Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, Sam Gross, Nathan Ng, David Grangier, Michael Auli 发布。
+1. **[RoCBert](https://huggingface.co/docs/transformers/main/model_doc/roc_bert)** (来自 WeChatAI), 伴随论文 [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) 由 HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou 发布。
 1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (来自 ZhuiyiTechnology), 伴随论文 [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/pdf/2104.09864v1.pdf) 由 Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu 发布。
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (来自 NVIDIA) 伴随论文 [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) 由 Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo 发布。
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (来自 ASAPP) 伴随论文 [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) 由 Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi 发布。
@@ -371,7 +372,7 @@ conda install -c huggingface transformers
 1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (来自 Microsoft Research) 伴随论文 [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) 由 Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou 发布。
 1. **[Time Series Transformer](https://huggingface.co/docs/transformers/model_doc/time_series_transformer)** (from HuggingFace).
 1. **[TimeSformer](https://huggingface.co/docs/transformers/main/model_doc/timesformer)** (from Facebook) released with the paper [Is Space-Time Attention All You Need for Video Understanding?](https://arxiv.org/abs/2102.05095) by Gedas Bertasius, Heng Wang, Lorenzo Torresani.
-1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine 
+1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
 1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (来自 Google/CMU) 伴随论文 [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) 由 Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov 发布。
 1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (来自 Microsoft) 伴随论文 [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) 由 Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei 发布。
 1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler

README_zh-hant.md

@@ -363,7 +363,8 @@ conda install -c huggingface transformers
 1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/pdf/2010.12821.pdf) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
 1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
 1. **[RoBERTa](https://huggingface.co/docs/transformers/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.
-1. **[RoCBert](https://huggingface.co/docs/transformers/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
+1. **[RoBERTa-PreLayerNorm](https://huggingface.co/docs/transformers/main/model_doc/roberta-prelayernorm)** (from Facebook) released with the paper [fairseq: A Fast, Extensible Toolkit for Sequence Modeling](https://arxiv.org/abs/1904.01038) by Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, Sam Gross, Nathan Ng, David Grangier, Michael Auli.
+1. **[RoCBert](https://huggingface.co/docs/transformers/main/model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
 1. **[RoFormer](https://huggingface.co/docs/transformers/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.
 1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
 1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
@@ -381,9 +382,9 @@ conda install -c huggingface transformers
 1. **[Table Transformer](https://huggingface.co/docs/transformers/model_doc/table-transformer)** (from Microsoft Research) released with the paper [PubTables-1M: Towards Comprehensive Table Extraction From Unstructured Documents](https://arxiv.org/abs/2110.00061) by Brandon Smock, Rohith Pesala, Robin Abraham.
 1. **[TAPAS](https://huggingface.co/docs/transformers/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.
 1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (from Microsoft Research) released with the paper [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) by Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou.
-1. **[Time Series Transformer](https://huggingface.co/docs/transformers/model_doc/time_series_transformer)** (from HuggingFace).  
+1. **[Time Series Transformer](https://huggingface.co/docs/transformers/model_doc/time_series_transformer)** (from HuggingFace).
 1. **[TimeSformer](https://huggingface.co/docs/transformers/main/model_doc/timesformer)** (from Facebook) released with the paper [Is Space-Time Attention All You Need for Video Understanding?](https://arxiv.org/abs/2102.05095) by Gedas Bertasius, Heng Wang, Lorenzo Torresani.
-1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine 
+1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
 1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
 1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (from Microsoft) released with the paper [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) by Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei.
 1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler

docs/source/en/_toctree.yml

@@ -349,6 +349,8 @@
         title: RetriBERT
       - local: model_doc/roberta
         title: RoBERTa
+      - local: model_doc/roberta-prelayernorm
+        title: RoBERTa-PreLayerNorm
       - local: model_doc/roc_bert
         title: RoCBert
       - local: model_doc/roformer

docs/source/en/index.mdx

@@ -164,6 +164,7 @@ The documentation is organized into five sections:
 1. **[RemBERT](model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/abs/2010.12821) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
 1. **[ResNet](model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
 1. **[RoBERTa](model_doc/roberta)** (from Facebook), released together with the paper [RoBERTa: 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.
+1. **[RoBERTa-PreLayerNorm](model_doc/roberta-prelayernorm)** (from Facebook) released with the paper [fairseq: A Fast, Extensible Toolkit for Sequence Modeling](https://arxiv.org/abs/1904.01038) by Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, Sam Gross, Nathan Ng, David Grangier, Michael Auli.
 1. **[RoCBert](model_doc/roc_bert)** (from WeChatAI) released with the paper [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf) by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
 1. **[RoFormer](model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/abs/2104.09864) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
 1. **[SegFormer](model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
@@ -331,6 +332,7 @@ Flax), PyTorch, and/or TensorFlow.
 |            ResNet             |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
 |           RetriBERT           |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
 |            RoBERTa            |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
+|     RoBERTa-PreLayerNorm      |       ❌       |       ❌       |       ✅        |         ✅         |      ✅      |
 |            RoCBert            |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
 |           RoFormer            |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 |           SegFormer           |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |

docs/source/en/model_doc/roberta-prelayernorm.mdx

+<!--Copyright 2022 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.
+-->
+
+# RoBERTa-PreLayerNorm
+
+## Overview
+
+The RoBERTa-PreLayerNorm model was proposed in [fairseq: A Fast, Extensible Toolkit for Sequence Modeling](https://arxiv.org/abs/1904.01038) by Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, Sam Gross, Nathan Ng, David Grangier, Michael Auli.
+It is identical to using the `--encoder-normalize-before` flag in [fairseq](https://fairseq.readthedocs.io/).
+
+The abstract from the paper is the following:
+
+*fairseq is an open-source sequence modeling toolkit that allows researchers and developers to train custom models for translation, summarization, language modeling, and other text generation tasks. The toolkit is based on PyTorch and supports distributed training across multiple GPUs and machines. We also support fast mixed-precision training and inference on modern GPUs.*
+
+Tips:
+
+- The implementation is the same as [Roberta](roberta) except instead of using _Add and Norm_ it does _Norm and Add_. _Add_ and _Norm_ refers to the Addition and LayerNormalization as described in [Attention Is All You Need](https://arxiv.org/abs/1706.03762).
+- This is identical to using the `--encoder-normalize-before` flag in [fairseq](https://fairseq.readthedocs.io/).
+
+This model was contributed by [andreasmaden](https://huggingface.co/andreasmaden).
+The original code can be found [here](https://github.com/princeton-nlp/DinkyTrain).
+
+
+## RobertaPreLayerNormConfig
+
+[[autodoc]] RobertaPreLayerNormConfig
+
+## RobertaPreLayerNormModel
+
+[[autodoc]] RobertaPreLayerNormModel
+    - forward
+
+## RobertaPreLayerNormForCausalLM
+
+[[autodoc]] RobertaPreLayerNormForCausalLM
+    - forward
+
+## RobertaPreLayerNormForMaskedLM
+
+[[autodoc]] RobertaPreLayerNormForMaskedLM
+    - forward
+
+## RobertaPreLayerNormForSequenceClassification
+
+[[autodoc]] RobertaPreLayerNormForSequenceClassification
+    - forward
+
+## RobertaPreLayerNormForMultipleChoice
+
+[[autodoc]] RobertaPreLayerNormForMultipleChoice
+    - forward
+
+## RobertaPreLayerNormForTokenClassification
+
+[[autodoc]] RobertaPreLayerNormForTokenClassification
+    - forward
+
+## RobertaPreLayerNormForQuestionAnswering
+
+[[autodoc]] RobertaPreLayerNormForQuestionAnswering
+    - forward
+
+## TFRobertaPreLayerNormModel
+
+[[autodoc]] TFRobertaPreLayerNormModel
+    - call
+
+## TFRobertaPreLayerNormForCausalLM
+
+[[autodoc]] TFRobertaPreLayerNormForCausalLM
+    - call
+
+## TFRobertaPreLayerNormForMaskedLM
+
+[[autodoc]] TFRobertaPreLayerNormForMaskedLM
+    - call
+
+## TFRobertaPreLayerNormForSequenceClassification
+
+[[autodoc]] TFRobertaPreLayerNormForSequenceClassification
+    - call
+
+## TFRobertaPreLayerNormForMultipleChoice
+
+[[autodoc]] TFRobertaPreLayerNormForMultipleChoice
+    - call
+
+## TFRobertaPreLayerNormForTokenClassification
+
+[[autodoc]] TFRobertaPreLayerNormForTokenClassification
+    - call
+
+## TFRobertaPreLayerNormForQuestionAnswering
+
+[[autodoc]] TFRobertaPreLayerNormForQuestionAnswering
+    - call
+
+## FlaxRobertaPreLayerNormModel
+
+[[autodoc]] FlaxRobertaPreLayerNormModel
+    - __call__
+
+## FlaxRobertaPreLayerNormForCausalLM
+
+[[autodoc]] FlaxRobertaPreLayerNormForCausalLM
+    - __call__
+
+## FlaxRobertaPreLayerNormForMaskedLM
+
+[[autodoc]] FlaxRobertaPreLayerNormForMaskedLM
+    - __call__
+
+## FlaxRobertaPreLayerNormForSequenceClassification
+
+[[autodoc]] FlaxRobertaPreLayerNormForSequenceClassification
+    - __call__
+
+## FlaxRobertaPreLayerNormForMultipleChoice
+
+[[autodoc]] FlaxRobertaPreLayerNormForMultipleChoice
+    - __call__
+
+## FlaxRobertaPreLayerNormForTokenClassification
+
+[[autodoc]] FlaxRobertaPreLayerNormForTokenClassification
+    - __call__
+
+## FlaxRobertaPreLayerNormForQuestionAnswering
+
+[[autodoc]] FlaxRobertaPreLayerNormForQuestionAnswering
+    - __call__

docs/source/en/serialization.mdx

@@ -105,6 +105,7 @@ Ready-made configurations include the following architectures:
 - RemBERT
 - ResNet
 - RoBERTa
+- RoBERTa-PreLayerNorm
 - RoFormer
 - SegFormer
 - SqueezeBERT

src/transformers/__init__.py

@@ -354,6 +354,7 @@ _import_structure = {
     "models.resnet": ["RESNET_PRETRAINED_CONFIG_ARCHIVE_MAP", "ResNetConfig"],
     "models.retribert": ["RETRIBERT_PRETRAINED_CONFIG_ARCHIVE_MAP", "RetriBertConfig", "RetriBertTokenizer"],
     "models.roberta": ["ROBERTA_PRETRAINED_CONFIG_ARCHIVE_MAP", "RobertaConfig", "RobertaTokenizer"],
+    "models.roberta_prelayernorm": ["ROBERTA_PRELAYERNORM_PRETRAINED_CONFIG_ARCHIVE_MAP", "RobertaPreLayerNormConfig"],
     "models.roc_bert": ["ROC_BERT_PRETRAINED_CONFIG_ARCHIVE_MAP", "RoCBertConfig", "RoCBertTokenizer"],
     "models.roformer": ["ROFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP", "RoFormerConfig", "RoFormerTokenizer"],
     "models.segformer": ["SEGFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP", "SegformerConfig"],
@@ -1986,6 +1987,19 @@ else:
             "RobertaPreTrainedModel",
         ]
     )
+    _import_structure["models.roberta_prelayernorm"].extend(
+        [
+            "ROBERTA_PRELAYERNORM_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "RobertaPreLayerNormForCausalLM",
+            "RobertaPreLayerNormForMaskedLM",
+            "RobertaPreLayerNormForMultipleChoice",
+            "RobertaPreLayerNormForQuestionAnswering",
+            "RobertaPreLayerNormForSequenceClassification",
+            "RobertaPreLayerNormForTokenClassification",
+            "RobertaPreLayerNormModel",
+            "RobertaPreLayerNormPreTrainedModel",
+        ]
+    )
     _import_structure["models.roc_bert"].extend(
         [
             "ROC_BERT_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -2931,6 +2945,20 @@ else:
             "TFRobertaPreTrainedModel",
         ]
     )
+    _import_structure["models.roberta_prelayernorm"].extend(
+        [
+            "TF_ROBERTA_PRELAYERNORM_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "TFRobertaPreLayerNormForCausalLM",
+            "TFRobertaPreLayerNormForMaskedLM",
+            "TFRobertaPreLayerNormForMultipleChoice",
+            "TFRobertaPreLayerNormForQuestionAnswering",
+            "TFRobertaPreLayerNormForSequenceClassification",
+            "TFRobertaPreLayerNormForTokenClassification",
+            "TFRobertaPreLayerNormMainLayer",
+            "TFRobertaPreLayerNormModel",
+            "TFRobertaPreLayerNormPreTrainedModel",
+        ]
+    )
     _import_structure["models.roformer"].extend(
         [
             "TF_ROFORMER_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -3297,6 +3325,18 @@ else:
             "FlaxRobertaPreTrainedModel",
         ]
     )
+    _import_structure["models.roberta_prelayernorm"].extend(
+        [
+            "FlaxRobertaPreLayerNormForCausalLM",
+            "FlaxRobertaPreLayerNormForMaskedLM",
+            "FlaxRobertaPreLayerNormForMultipleChoice",
+            "FlaxRobertaPreLayerNormForQuestionAnswering",
+            "FlaxRobertaPreLayerNormForSequenceClassification",
+            "FlaxRobertaPreLayerNormForTokenClassification",
+            "FlaxRobertaPreLayerNormModel",
+            "FlaxRobertaPreLayerNormPreTrainedModel",
+        ]
+    )
     _import_structure["models.roformer"].extend(
         [
             "FlaxRoFormerForMaskedLM",
@@ -3624,6 +3664,10 @@ if TYPE_CHECKING:
     from .models.resnet import RESNET_PRETRAINED_CONFIG_ARCHIVE_MAP, ResNetConfig
     from .models.retribert import RETRIBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, RetriBertConfig, RetriBertTokenizer
     from .models.roberta import ROBERTA_PRETRAINED_CONFIG_ARCHIVE_MAP, RobertaConfig, RobertaTokenizer
+    from .models.roberta_prelayernorm import (
+        ROBERTA_PRELAYERNORM_PRETRAINED_CONFIG_ARCHIVE_MAP,
+        RobertaPreLayerNormConfig,
+    )
     from .models.roc_bert import ROC_BERT_PRETRAINED_CONFIG_ARCHIVE_MAP, RoCBertConfig, RoCBertTokenizer
     from .models.roformer import ROFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, RoFormerConfig, RoFormerTokenizer
     from .models.segformer import SEGFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, SegformerConfig
@@ -4979,6 +5023,17 @@ if TYPE_CHECKING:
             RobertaModel,
             RobertaPreTrainedModel,
         )
+        from .models.roberta_prelayernorm import (
+            ROBERTA_PRELAYERNORM_PRETRAINED_MODEL_ARCHIVE_LIST,
+            RobertaPreLayerNormForCausalLM,
+            RobertaPreLayerNormForMaskedLM,
+            RobertaPreLayerNormForMultipleChoice,
+            RobertaPreLayerNormForQuestionAnswering,
+            RobertaPreLayerNormForSequenceClassification,
+            RobertaPreLayerNormForTokenClassification,
+            RobertaPreLayerNormModel,
+            RobertaPreLayerNormPreTrainedModel,
+        )
         from .models.roc_bert import (
             ROC_BERT_PRETRAINED_MODEL_ARCHIVE_LIST,
             RoCBertForCausalLM,
@@ -5745,6 +5800,18 @@ if TYPE_CHECKING:
             TFRobertaModel,
             TFRobertaPreTrainedModel,
         )
+        from .models.roberta_prelayernorm import (
+            TF_ROBERTA_PRELAYERNORM_PRETRAINED_MODEL_ARCHIVE_LIST,
+            TFRobertaPreLayerNormForCausalLM,
+            TFRobertaPreLayerNormForMaskedLM,
+            TFRobertaPreLayerNormForMultipleChoice,
+            TFRobertaPreLayerNormForQuestionAnswering,
+            TFRobertaPreLayerNormForSequenceClassification,
+            TFRobertaPreLayerNormForTokenClassification,
+            TFRobertaPreLayerNormMainLayer,
+            TFRobertaPreLayerNormModel,
+            TFRobertaPreLayerNormPreTrainedModel,
+        )
         from .models.roformer import (
             TF_ROFORMER_PRETRAINED_MODEL_ARCHIVE_LIST,
             TFRoFormerForCausalLM,
@@ -6022,6 +6089,16 @@ if TYPE_CHECKING:
             FlaxRobertaModel,
             FlaxRobertaPreTrainedModel,
         )
+        from .models.roberta_prelayernorm import (
+            FlaxRobertaPreLayerNormForCausalLM,
+            FlaxRobertaPreLayerNormForMaskedLM,
+            FlaxRobertaPreLayerNormForMultipleChoice,
+            FlaxRobertaPreLayerNormForQuestionAnswering,
+            FlaxRobertaPreLayerNormForSequenceClassification,
+            FlaxRobertaPreLayerNormForTokenClassification,
+            FlaxRobertaPreLayerNormModel,
+            FlaxRobertaPreLayerNormPreTrainedModel,
+        )
         from .models.roformer import (
             FlaxRoFormerForMaskedLM,
             FlaxRoFormerForMultipleChoice,

src/transformers/models/__init__.py

@@ -137,6 +137,7 @@ from . import (
     resnet,
     retribert,
     roberta,
+    roberta_prelayernorm,
     roc_bert,
     roformer,
     segformer,

src/transformers/models/auto/configuration_auto.py

@@ -134,6 +134,7 @@ CONFIG_MAPPING_NAMES = OrderedDict(
         ("resnet", "ResNetConfig"),
         ("retribert", "RetriBertConfig"),
         ("roberta", "RobertaConfig"),
+        ("roberta-prelayernorm", "RobertaPreLayerNormConfig"),
         ("roc_bert", "RoCBertConfig"),
         ("roformer", "RoFormerConfig"),
         ("segformer", "SegformerConfig"),
@@ -282,6 +283,7 @@ CONFIG_ARCHIVE_MAP_MAPPING_NAMES = OrderedDict(
         ("resnet", "RESNET_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("retribert", "RETRIBERT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("roberta", "ROBERTA_PRETRAINED_CONFIG_ARCHIVE_MAP"),
+        ("roberta-prelayernorm", "ROBERTA_PRELAYERNORM_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("roc_bert", "ROC_BERT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("roformer", "ROFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("segformer", "SEGFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP"),
@@ -450,6 +452,7 @@ MODEL_NAMES_MAPPING = OrderedDict(
         ("resnet", "ResNet"),
         ("retribert", "RetriBERT"),
         ("roberta", "RoBERTa"),
+        ("roberta-prelayernorm", "RoBERTa-PreLayerNorm"),
         ("roc_bert", "RoCBert"),
         ("roformer", "RoFormer"),
         ("segformer", "SegFormer"),

src/transformers/models/auto/modeling_auto.py

@@ -132,6 +132,7 @@ MODEL_MAPPING_NAMES = OrderedDict(
         ("resnet", "ResNetModel"),
         ("retribert", "RetriBertModel"),
         ("roberta", "RobertaModel"),
+        ("roberta-prelayernorm", "RobertaPreLayerNormModel"),
         ("roc_bert", "RoCBertModel"),
         ("roformer", "RoFormerModel"),
         ("segformer", "SegformerModel"),
@@ -214,6 +215,7 @@ MODEL_FOR_PRETRAINING_MAPPING_NAMES = OrderedDict(
         ("openai-gpt", "OpenAIGPTLMHeadModel"),
         ("retribert", "RetriBertModel"),
         ("roberta", "RobertaForMaskedLM"),
+        ("roberta-prelayernorm", "RobertaPreLayerNormForMaskedLM"),
         ("roc_bert", "RoCBertForPreTraining"),
         ("splinter", "SplinterForPreTraining"),
         ("squeezebert", "SqueezeBertForMaskedLM"),
@@ -289,6 +291,7 @@ MODEL_WITH_LM_HEAD_MAPPING_NAMES = OrderedDict(
         ("reformer", "ReformerModelWithLMHead"),
         ("rembert", "RemBertForMaskedLM"),
         ("roberta", "RobertaForMaskedLM"),
+        ("roberta-prelayernorm", "RobertaPreLayerNormForMaskedLM"),
         ("roc_bert", "RoCBertForMaskedLM"),
         ("roformer", "RoFormerForMaskedLM"),
         ("speech_to_text", "Speech2TextForConditionalGeneration"),
@@ -344,6 +347,7 @@ MODEL_FOR_CAUSAL_LM_MAPPING_NAMES = OrderedDict(
         ("reformer", "ReformerModelWithLMHead"),
         ("rembert", "RemBertForCausalLM"),
         ("roberta", "RobertaForCausalLM"),
+        ("roberta-prelayernorm", "RobertaPreLayerNormForCausalLM"),
         ("roc_bert", "RoCBertForCausalLM"),
         ("roformer", "RoFormerForCausalLM"),
         ("speech_to_text_2", "Speech2Text2ForCausalLM"),
@@ -496,6 +500,7 @@ MODEL_FOR_MASKED_LM_MAPPING_NAMES = OrderedDict(
         ("reformer", "ReformerForMaskedLM"),
         ("rembert", "RemBertForMaskedLM"),
         ("roberta", "RobertaForMaskedLM"),
+        ("roberta-prelayernorm", "RobertaPreLayerNormForMaskedLM"),
         ("roc_bert", "RoCBertForMaskedLM"),
         ("roformer", "RoFormerForMaskedLM"),
         ("squeezebert", "SqueezeBertForMaskedLM"),
@@ -619,6 +624,7 @@ MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES = OrderedDict(
         ("reformer", "ReformerForSequenceClassification"),
         ("rembert", "RemBertForSequenceClassification"),
         ("roberta", "RobertaForSequenceClassification"),
+        ("roberta-prelayernorm", "RobertaPreLayerNormForSequenceClassification"),
         ("roc_bert", "RoCBertForSequenceClassification"),
         ("roformer", "RoFormerForSequenceClassification"),
         ("squeezebert", "SqueezeBertForSequenceClassification"),
@@ -675,6 +681,7 @@ MODEL_FOR_QUESTION_ANSWERING_MAPPING_NAMES = OrderedDict(
         ("reformer", "ReformerForQuestionAnswering"),
         ("rembert", "RemBertForQuestionAnswering"),
         ("roberta", "RobertaForQuestionAnswering"),
+        ("roberta-prelayernorm", "RobertaPreLayerNormForQuestionAnswering"),
         ("roc_bert", "RoCBertForQuestionAnswering"),
         ("roformer", "RoFormerForQuestionAnswering"),
         ("splinter", "SplinterForQuestionAnswering"),
@@ -746,6 +753,7 @@ MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING_NAMES = OrderedDict(
         ("qdqbert", "QDQBertForTokenClassification"),
         ("rembert", "RemBertForTokenClassification"),
         ("roberta", "RobertaForTokenClassification"),
+        ("roberta-prelayernorm", "RobertaPreLayerNormForTokenClassification"),
         ("roc_bert", "RoCBertForTokenClassification"),
         ("roformer", "RoFormerForTokenClassification"),
         ("squeezebert", "SqueezeBertForTokenClassification"),
@@ -785,6 +793,7 @@ MODEL_FOR_MULTIPLE_CHOICE_MAPPING_NAMES = OrderedDict(
         ("qdqbert", "QDQBertForMultipleChoice"),
         ("rembert", "RemBertForMultipleChoice"),
         ("roberta", "RobertaForMultipleChoice"),
+        ("roberta-prelayernorm", "RobertaPreLayerNormForMultipleChoice"),
         ("roc_bert", "RoCBertForMultipleChoice"),
         ("roformer", "RoFormerForMultipleChoice"),
         ("squeezebert", "SqueezeBertForMultipleChoice"),

src/transformers/models/auto/modeling_flax_auto.py

@@ -49,6 +49,7 @@ FLAX_MODEL_MAPPING_NAMES = OrderedDict(
         ("opt", "FlaxOPTModel"),
         ("pegasus", "FlaxPegasusModel"),
         ("roberta", "FlaxRobertaModel"),
+        ("roberta-prelayernorm", "FlaxRobertaPreLayerNormModel"),
         ("roformer", "FlaxRoFormerModel"),
         ("t5", "FlaxT5Model"),
         ("vision-text-dual-encoder", "FlaxVisionTextDualEncoderModel"),
@@ -71,6 +72,7 @@ FLAX_MODEL_FOR_PRETRAINING_MAPPING_NAMES = OrderedDict(
         ("mbart", "FlaxMBartForConditionalGeneration"),
         ("mt5", "FlaxMT5ForConditionalGeneration"),
         ("roberta", "FlaxRobertaForMaskedLM"),
+        ("roberta-prelayernorm", "FlaxRobertaPreLayerNormForMaskedLM"),
         ("roformer", "FlaxRoFormerForMaskedLM"),
         ("t5", "FlaxT5ForConditionalGeneration"),
         ("wav2vec2", "FlaxWav2Vec2ForPreTraining"),
@@ -89,6 +91,7 @@ FLAX_MODEL_FOR_MASKED_LM_MAPPING_NAMES = OrderedDict(
         ("electra", "FlaxElectraForMaskedLM"),
         ("mbart", "FlaxMBartForConditionalGeneration"),
         ("roberta", "FlaxRobertaForMaskedLM"),
+        ("roberta-prelayernorm", "FlaxRobertaPreLayerNormForMaskedLM"),
         ("roformer", "FlaxRoFormerForMaskedLM"),
         ("xlm-roberta", "FlaxXLMRobertaForMaskedLM"),
     ]
@@ -137,6 +140,7 @@ FLAX_MODEL_FOR_CAUSAL_LM_MAPPING_NAMES = OrderedDict(
         ("gptj", "FlaxGPTJForCausalLM"),
         ("opt", "FlaxOPTForCausalLM"),
         ("roberta", "FlaxRobertaForCausalLM"),
+        ("roberta-prelayernorm", "FlaxRobertaPreLayerNormForCausalLM"),
         ("xglm", "FlaxXGLMForCausalLM"),
     ]
 )
@@ -152,6 +156,7 @@ FLAX_MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES = OrderedDict(
         ("electra", "FlaxElectraForSequenceClassification"),
         ("mbart", "FlaxMBartForSequenceClassification"),
         ("roberta", "FlaxRobertaForSequenceClassification"),
+        ("roberta-prelayernorm", "FlaxRobertaPreLayerNormForSequenceClassification"),
         ("roformer", "FlaxRoFormerForSequenceClassification"),
         ("xlm-roberta", "FlaxXLMRobertaForSequenceClassification"),
     ]
@@ -168,6 +173,7 @@ FLAX_MODEL_FOR_QUESTION_ANSWERING_MAPPING_NAMES = OrderedDict(
         ("electra", "FlaxElectraForQuestionAnswering"),
         ("mbart", "FlaxMBartForQuestionAnswering"),
         ("roberta", "FlaxRobertaForQuestionAnswering"),
+        ("roberta-prelayernorm", "FlaxRobertaPreLayerNormForQuestionAnswering"),
         ("roformer", "FlaxRoFormerForQuestionAnswering"),
         ("xlm-roberta", "FlaxXLMRobertaForQuestionAnswering"),
     ]
@@ -182,6 +188,7 @@ FLAX_MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING_NAMES = OrderedDict(
         ("distilbert", "FlaxDistilBertForTokenClassification"),
         ("electra", "FlaxElectraForTokenClassification"),
         ("roberta", "FlaxRobertaForTokenClassification"),
+        ("roberta-prelayernorm", "FlaxRobertaPreLayerNormForTokenClassification"),
         ("roformer", "FlaxRoFormerForTokenClassification"),
         ("xlm-roberta", "FlaxXLMRobertaForTokenClassification"),
     ]
@@ -196,6 +203,7 @@ FLAX_MODEL_FOR_MULTIPLE_CHOICE_MAPPING_NAMES = OrderedDict(
         ("distilbert", "FlaxDistilBertForMultipleChoice"),
         ("electra", "FlaxElectraForMultipleChoice"),
         ("roberta", "FlaxRobertaForMultipleChoice"),
+        ("roberta-prelayernorm", "FlaxRobertaPreLayerNormForMultipleChoice"),
         ("roformer", "FlaxRoFormerForMultipleChoice"),
         ("xlm-roberta", "FlaxXLMRobertaForMultipleChoice"),
     ]

src/transformers/models/auto/modeling_tf_auto.py

@@ -73,6 +73,7 @@ TF_MODEL_MAPPING_NAMES = OrderedDict(
         ("rembert", "TFRemBertModel"),
         ("resnet", "TFResNetModel"),
         ("roberta", "TFRobertaModel"),
+        ("roberta-prelayernorm", "TFRobertaPreLayerNormModel"),
         ("roformer", "TFRoFormerModel"),
         ("segformer", "TFSegformerModel"),
         ("speech_to_text", "TFSpeech2TextModel"),
@@ -111,6 +112,7 @@ TF_MODEL_FOR_PRETRAINING_MAPPING_NAMES = OrderedDict(
         ("mpnet", "TFMPNetForMaskedLM"),
         ("openai-gpt", "TFOpenAIGPTLMHeadModel"),
         ("roberta", "TFRobertaForMaskedLM"),
+        ("roberta-prelayernorm", "TFRobertaPreLayerNormForMaskedLM"),
         ("t5", "TFT5ForConditionalGeneration"),
         ("tapas", "TFTapasForMaskedLM"),
         ("transfo-xl", "TFTransfoXLLMHeadModel"),
@@ -147,6 +149,7 @@ TF_MODEL_WITH_LM_HEAD_MAPPING_NAMES = OrderedDict(
         ("openai-gpt", "TFOpenAIGPTLMHeadModel"),
         ("rembert", "TFRemBertForMaskedLM"),
         ("roberta", "TFRobertaForMaskedLM"),
+        ("roberta-prelayernorm", "TFRobertaPreLayerNormForMaskedLM"),
         ("roformer", "TFRoFormerForMaskedLM"),
         ("speech_to_text", "TFSpeech2TextForConditionalGeneration"),
         ("t5", "TFT5ForConditionalGeneration"),
@@ -172,6 +175,7 @@ TF_MODEL_FOR_CAUSAL_LM_MAPPING_NAMES = OrderedDict(
         ("opt", "TFOPTForCausalLM"),
         ("rembert", "TFRemBertForCausalLM"),
         ("roberta", "TFRobertaForCausalLM"),
+        ("roberta-prelayernorm", "TFRobertaPreLayerNormForCausalLM"),
         ("roformer", "TFRoFormerForCausalLM"),
         ("transfo-xl", "TFTransfoXLLMHeadModel"),
         ("xglm", "TFXGLMForCausalLM"),
@@ -238,6 +242,7 @@ TF_MODEL_FOR_MASKED_LM_MAPPING_NAMES = OrderedDict(
         ("mpnet", "TFMPNetForMaskedLM"),
         ("rembert", "TFRemBertForMaskedLM"),
         ("roberta", "TFRobertaForMaskedLM"),
+        ("roberta-prelayernorm", "TFRobertaPreLayerNormForMaskedLM"),
         ("roformer", "TFRoFormerForMaskedLM"),
         ("tapas", "TFTapasForMaskedLM"),
         ("xlm", "TFXLMWithLMHeadModel"),
@@ -295,6 +300,7 @@ TF_MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES = OrderedDict(
         ("openai-gpt", "TFOpenAIGPTForSequenceClassification"),
         ("rembert", "TFRemBertForSequenceClassification"),
         ("roberta", "TFRobertaForSequenceClassification"),
+        ("roberta-prelayernorm", "TFRobertaPreLayerNormForSequenceClassification"),
         ("roformer", "TFRoFormerForSequenceClassification"),
         ("tapas", "TFTapasForSequenceClassification"),
         ("transfo-xl", "TFTransfoXLForSequenceClassification"),
@@ -324,6 +330,7 @@ TF_MODEL_FOR_QUESTION_ANSWERING_MAPPING_NAMES = OrderedDict(
         ("mpnet", "TFMPNetForQuestionAnswering"),
         ("rembert", "TFRemBertForQuestionAnswering"),
         ("roberta", "TFRobertaForQuestionAnswering"),
+        ("roberta-prelayernorm", "TFRobertaPreLayerNormForQuestionAnswering"),
         ("roformer", "TFRoFormerForQuestionAnswering"),
         ("xlm", "TFXLMForQuestionAnsweringSimple"),
         ("xlm-roberta", "TFXLMRobertaForQuestionAnswering"),
@@ -366,6 +373,7 @@ TF_MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING_NAMES = OrderedDict(
         ("mpnet", "TFMPNetForTokenClassification"),
         ("rembert", "TFRemBertForTokenClassification"),
         ("roberta", "TFRobertaForTokenClassification"),
+        ("roberta-prelayernorm", "TFRobertaPreLayerNormForTokenClassification"),
         ("roformer", "TFRoFormerForTokenClassification"),
         ("xlm", "TFXLMForTokenClassification"),
         ("xlm-roberta", "TFXLMRobertaForTokenClassification"),
@@ -389,6 +397,7 @@ TF_MODEL_FOR_MULTIPLE_CHOICE_MAPPING_NAMES = OrderedDict(
         ("mpnet", "TFMPNetForMultipleChoice"),
         ("rembert", "TFRemBertForMultipleChoice"),
         ("roberta", "TFRobertaForMultipleChoice"),
+        ("roberta-prelayernorm", "TFRobertaPreLayerNormForMultipleChoice"),
         ("roformer", "TFRoFormerForMultipleChoice"),
         ("xlm", "TFXLMForMultipleChoice"),
         ("xlm-roberta", "TFXLMRobertaForMultipleChoice"),

src/transformers/models/auto/tokenization_auto.py

@@ -252,6 +252,10 @@ else:
             ),
             ("retribert", ("RetriBertTokenizer", "RetriBertTokenizerFast" if is_tokenizers_available() else None)),
             ("roberta", ("RobertaTokenizer", "RobertaTokenizerFast" if is_tokenizers_available() else None)),
+            (
+                "roberta-prelayernorm",
+                ("RobertaTokenizer", "RobertaTokenizerFast" if is_tokenizers_available() else None),
+            ),
             ("roc_bert", ("RoCBertTokenizer", None)),
             ("roformer", ("RoFormerTokenizer", "RoFormerTokenizerFast" if is_tokenizers_available() else None)),
             ("speech_to_text", ("Speech2TextTokenizer" if is_sentencepiece_available() else None, None)),

src/transformers/models/roberta_prelayernorm/__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 2022 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 ...utils import (
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    is_flax_available,
+    is_tf_available,
+    is_torch_available,
+)
+
+
+_import_structure = {
+    "configuration_roberta_prelayernorm": [
+        "ROBERTA_PRELAYERNORM_PRETRAINED_CONFIG_ARCHIVE_MAP",
+        "RobertaPreLayerNormConfig",
+        "RobertaPreLayerNormOnnxConfig",
+    ],
+}
+
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["modeling_roberta_prelayernorm"] = [
+        "ROBERTA_PRELAYERNORM_PRETRAINED_MODEL_ARCHIVE_LIST",
+        "RobertaPreLayerNormForCausalLM",
+        "RobertaPreLayerNormForMaskedLM",
+        "RobertaPreLayerNormForMultipleChoice",
+        "RobertaPreLayerNormForQuestionAnswering",
+        "RobertaPreLayerNormForSequenceClassification",
+        "RobertaPreLayerNormForTokenClassification",
+        "RobertaPreLayerNormModel",
+        "RobertaPreLayerNormPreTrainedModel",
+    ]
+
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["modeling_tf_roberta_prelayernorm"] = [
+        "TF_ROBERTA_PRELAYERNORM_PRETRAINED_MODEL_ARCHIVE_LIST",
+        "TFRobertaPreLayerNormForCausalLM",
+        "TFRobertaPreLayerNormForMaskedLM",
+        "TFRobertaPreLayerNormForMultipleChoice",
+        "TFRobertaPreLayerNormForQuestionAnswering",
+        "TFRobertaPreLayerNormForSequenceClassification",
+        "TFRobertaPreLayerNormForTokenClassification",
+        "TFRobertaPreLayerNormMainLayer",
+        "TFRobertaPreLayerNormModel",
+        "TFRobertaPreLayerNormPreTrainedModel",
+    ]
+
+try:
+    if not is_flax_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["modeling_flax_roberta_prelayernorm"] = [
+        "FlaxRobertaPreLayerNormForCausalLM",
+        "FlaxRobertaPreLayerNormForMaskedLM",
+        "FlaxRobertaPreLayerNormForMultipleChoice",
+        "FlaxRobertaPreLayerNormForQuestionAnswering",
+        "FlaxRobertaPreLayerNormForSequenceClassification",
+        "FlaxRobertaPreLayerNormForTokenClassification",
+        "FlaxRobertaPreLayerNormModel",
+        "FlaxRobertaPreLayerNormPreTrainedModel",
+    ]
+
+
+if TYPE_CHECKING:
+    from .configuration_roberta_prelayernorm import (
+        ROBERTA_PRELAYERNORM_PRETRAINED_CONFIG_ARCHIVE_MAP,
+        RobertaPreLayerNormConfig,
+        RobertaPreLayerNormOnnxConfig,
+    )
+
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .modeling_roberta_prelayernorm import (
+            ROBERTA_PRELAYERNORM_PRETRAINED_MODEL_ARCHIVE_LIST,
+            RobertaPreLayerNormForCausalLM,
+            RobertaPreLayerNormForMaskedLM,
+            RobertaPreLayerNormForMultipleChoice,
+            RobertaPreLayerNormForQuestionAnswering,
+            RobertaPreLayerNormForSequenceClassification,
+            RobertaPreLayerNormForTokenClassification,
+            RobertaPreLayerNormModel,
+            RobertaPreLayerNormPreTrainedModel,
+        )
+
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .modeling_tf_roberta_prelayernorm import (
+            TF_ROBERTA_PRELAYERNORM_PRETRAINED_MODEL_ARCHIVE_LIST,
+            TFRobertaPreLayerNormForCausalLM,
+            TFRobertaPreLayerNormForMaskedLM,
+            TFRobertaPreLayerNormForMultipleChoice,
+            TFRobertaPreLayerNormForQuestionAnswering,
+            TFRobertaPreLayerNormForSequenceClassification,
+            TFRobertaPreLayerNormForTokenClassification,
+            TFRobertaPreLayerNormMainLayer,
+            TFRobertaPreLayerNormModel,
+            TFRobertaPreLayerNormPreTrainedModel,
+        )
+
+    try:
+        if not is_flax_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .modeling_flax_roberta_prelayernorm import (
+            FlaxRobertaPreLayerNormForCausalLM,
+            FlaxRobertaPreLayerNormForMaskedLM,
+            FlaxRobertaPreLayerNormForMultipleChoice,
+            FlaxRobertaPreLayerNormForQuestionAnswering,
+            FlaxRobertaPreLayerNormForSequenceClassification,
+            FlaxRobertaPreLayerNormForTokenClassification,
+            FlaxRobertaPreLayerNormModel,
+            FlaxRobertaPreLayerNormPreTrainedModel,
+        )
+
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)

src/transformers/models/roberta_prelayernorm/configuration_roberta_prelayernorm.py

+# coding=utf-8
+# Copyright 2022 The Google AI Language Team Authors 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.
+""" RoBERTa-PreLayerNorm configuration"""
+from collections import OrderedDict
+from typing import Mapping
+
+from ...configuration_utils import PretrainedConfig
+from ...onnx import OnnxConfig
+from ...utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+ROBERTA_PRELAYERNORM_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "andreasmadsen/efficient_mlm_m0.40": (
+        "https://huggingface.co/andreasmadsen/efficient_mlm_m0.40/resolve/main/config.json"
+    ),
+}
+
+
+# Copied from transformers.models.roberta.configuration_roberta.RobertaConfig with roberta-base->andreasmadsen/efficient_mlm_m0.40,RoBERTa->RoBERTa-PreLayerNorm,Roberta->RobertaPreLayerNorm,roberta->roberta-prelayernorm
+class RobertaPreLayerNormConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`RobertaPreLayerNormModel`] or a
+    [`TFRobertaPreLayerNormModel`]. It is used to instantiate a RoBERTa-PreLayerNorm 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 RoBERTa-PreLayerNorm
+    [andreasmadsen/efficient_mlm_m0.40](https://huggingface.co/andreasmadsen/efficient_mlm_m0.40) architecture.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 30522):
+            Vocabulary size of the RoBERTa-PreLayerNorm model. Defines the number of different tokens that can be
+            represented by the `inputs_ids` passed when calling [`RobertaPreLayerNormModel`] or
+            [`TFRobertaPreLayerNormModel`].
+        hidden_size (`int`, *optional*, defaults to 768):
+            Dimensionality of the encoder layers and the pooler layer.
+        num_hidden_layers (`int`, *optional*, defaults to 12):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 12):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        intermediate_size (`int`, *optional*, defaults to 3072):
+            Dimensionality of the "intermediate" (often named feed-forward) layer in the Transformer encoder.
+        hidden_act (`str` or `Callable`, *optional*, defaults to `"gelu"`):
+            The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
+            `"relu"`, `"silu"` and `"gelu_new"` are supported.
+        hidden_dropout_prob (`float`, *optional*, defaults to 0.1):
+            The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        attention_probs_dropout_prob (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for the attention probabilities.
+        max_position_embeddings (`int`, *optional*, defaults to 512):
+            The maximum sequence length that this model might ever be used with. Typically set this to something large
+            just in case (e.g., 512 or 1024 or 2048).
+        type_vocab_size (`int`, *optional*, defaults to 2):
+            The vocabulary size of the `token_type_ids` passed when calling [`RobertaPreLayerNormModel`] or
+            [`TFRobertaPreLayerNormModel`].
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        layer_norm_eps (`float`, *optional*, defaults to 1e-12):
+            The epsilon used by the layer normalization layers.
+        position_embedding_type (`str`, *optional*, defaults to `"absolute"`):
+            Type of position embedding. Choose one of `"absolute"`, `"relative_key"`, `"relative_key_query"`. For
+            positional embeddings use `"absolute"`. For more information on `"relative_key"`, please refer to
+            [Self-Attention with Relative Position Representations (Shaw et al.)](https://arxiv.org/abs/1803.02155).
+            For more information on `"relative_key_query"`, please refer to *Method 4* in [Improve Transformer Models
+            with Better Relative Position Embeddings (Huang et al.)](https://arxiv.org/abs/2009.13658).
+        is_decoder (`bool`, *optional*, defaults to `False`):
+            Whether the model is used as a decoder or not. If `False`, the model is used as an encoder.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`.
+        classifier_dropout (`float`, *optional*):
+            The dropout ratio for the classification head.
+
+    Examples:
+
+    ```python
+    >>> from transformers import RobertaPreLayerNormConfig, RobertaPreLayerNormModel
+
+    >>> # Initializing a RoBERTa-PreLayerNorm configuration
+    >>> configuration = RobertaPreLayerNormConfig()
+
+    >>> # Initializing a model (with random weights) from the configuration
+    >>> model = RobertaPreLayerNormModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+    model_type = "roberta-prelayernorm"
+
+    def __init__(
+        self,
+        vocab_size=30522,
+        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=512,
+        type_vocab_size=2,
+        initializer_range=0.02,
+        layer_norm_eps=1e-12,
+        pad_token_id=1,
+        bos_token_id=0,
+        eos_token_id=2,
+        position_embedding_type="absolute",
+        use_cache=True,
+        classifier_dropout=None,
+        **kwargs
+    ):
+        super().__init__(pad_token_id=pad_token_id, bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs)
+
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.hidden_act = hidden_act
+        self.intermediate_size = intermediate_size
+        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.initializer_range = initializer_range
+        self.layer_norm_eps = layer_norm_eps
+        self.position_embedding_type = position_embedding_type
+        self.use_cache = use_cache
+        self.classifier_dropout = classifier_dropout
+
+
+# Copied from transformers.models.roberta.configuration_roberta.RobertaOnnxConfig with Roberta->RobertaPreLayerNorm
+class RobertaPreLayerNormOnnxConfig(OnnxConfig):
+    @property
+    def inputs(self) -> Mapping[str, Mapping[int, str]]:
+        if self.task == "multiple-choice":
+            dynamic_axis = {0: "batch", 1: "choice", 2: "sequence"}
+        else:
+            dynamic_axis = {0: "batch", 1: "sequence"}
+        return OrderedDict(
+            [
+                ("input_ids", dynamic_axis),
+                ("attention_mask", dynamic_axis),
+            ]
+        )