Add RocBert (#20013)

* add roc_bert

* update roc_bert readme

* code style

* change name and delete unuse file

* udpate model file

* delete unuse log file

* delete tokenizer fast

* reformat code and change model file path

* add RocBertForPreTraining

* update docs

* delete wrong notes

* fix copies

* fix make repo-consistency error

* fix files are not present in the table of contents error

* change RocBert -> RoCBert

* add doc, add detail test
Co-authored-by: weiweishi <weiweishi@tencent.com>

README.md

@@ -366,6 +366,7 @@ 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/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

@@ -366,6 +366,7 @@ 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/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

@@ -316,6 +316,7 @@ 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/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.

README_zh-hans.md

@@ -340,6 +340,7 @@ 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/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。

README_zh-hant.md

@@ -352,6 +352,7 @@ 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/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.

docs/source/en/_toctree.yml

@@ -339,6 +339,8 @@
         title: RetriBERT
       - local: model_doc/roberta
         title: RoBERTa
+      - local: model_doc/roc_bert
+        title: RoCBert
       - local: model_doc/roformer
         title: RoFormer
       - local: model_doc/splinter

docs/source/en/index.mdx

@@ -154,6 +154,7 @@ 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. **[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.
 1. **[SEW](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.
@@ -305,6 +306,7 @@ Flax), PyTorch, and/or TensorFlow.
 |           ResNet            |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
 |          RetriBERT          |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
 |           RoBERTa           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
+|           RoCBert           |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
 |          RoFormer           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 |          SegFormer          |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
 |             SEW             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |

docs/source/en/model_doc/roc_bert.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.
+-->
+
+# RoCBert
+
+## Overview
+
+The RoCBert model was proposed in [RoCBert: Robust Chinese Bert with Multimodal Contrastive Pretraining](https://aclanthology.org/2022.acl-long.65.pdf)  by HuiSu, WeiweiShi, XiaoyuShen, XiaoZhou, TuoJi, JiaruiFang, JieZhou.
+It's a pretrained Chinese language model that is robust under various forms of adversarial attacks.
+
+The abstract from the paper is the following:
+
+*Large-scale pretrained language models have achieved SOTA results on NLP tasks. However, they have been shown
+vulnerable to adversarial attacks especially for logographic languages like Chinese. In this work, we propose
+ROCBERT: a pretrained Chinese Bert that is robust to various forms of adversarial attacks like word perturbation,
+synonyms, typos, etc. It is pretrained with the contrastive learning objective which maximizes the label consistency
+under different synthesized adversarial examples. The model takes as input multimodal information including the
+semantic, phonetic and visual features. We show all these features are important to the model robustness since the
+attack can be performed in all the three forms. Across 5 Chinese NLU tasks, ROCBERT outperforms strong baselines under
+three blackbox adversarial algorithms without sacrificing the performance on clean testset. It also performs the best
+in the toxic content detection task under human-made attacks.*
+
+This model was contributed by [weiweishi](https://huggingface.co/weiweishi).
+
+## RoCBertConfig
+
+[[autodoc]] RoCBertConfig
+    - all
+
+
+## RoCBertTokenizer
+
+[[autodoc]] RoCBertTokenizer
+    - build_inputs_with_special_tokens
+    - get_special_tokens_mask
+    - create_token_type_ids_from_sequences
+    - save_vocabulary
+
+
+## RoCBertModel
+
+[[autodoc]] RoCBertModel
+    - forward
+
+
+## RoCBertForPreTraining
+
+[[autodoc]] RoCBertForPreTraining
+    - forward
+
+
+## RoCBertForCausalLM
+
+[[autodoc]] RoCBertForCausalLM
+    - forward
+
+
+## RoCBertForMaskedLM
+
+[[autodoc]] RoCBertForMaskedLM
+    - forward
+
+
+## RoCBertForSequenceClassification
+
+[[autodoc]] transformers.RoCBertForSequenceClassification
+    - forward
+
+## RoCBertForMultipleChoice
+
+[[autodoc]] transformers.RoCBertForMultipleChoice
+    - forward
+
+
+## RoCBertForTokenClassification
+
+[[autodoc]] transformers.RoCBertForTokenClassification
+    - forward
+
+
+## RoCBertForQuestionAnswering
+
+[[autodoc]] RoCBertForQuestionAnswering
+    - forward
\ No newline at end of file

src/transformers/__init__.py

@@ -322,6 +322,7 @@ _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.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"],
     "models.sew": ["SEW_PRETRAINED_CONFIG_ARCHIVE_MAP", "SEWConfig"],
@@ -848,6 +849,23 @@ else:
 
     # PyTorch models structure
 
+    _import_structure["models.roc_bert"].extend(
+        [
+            "ROC_BERT_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "RoCBertForMaskedLM",
+            "RoCBertForCausalLM",
+            "RoCBertForMultipleChoice",
+            "RoCBertForQuestionAnswering",
+            "RoCBertForSequenceClassification",
+            "RoCBertForTokenClassification",
+            "RoCBertLayer",
+            "RoCBertModel",
+            "RoCBertForPreTraining",
+            "RoCBertPreTrainedModel",
+            "load_tf_weights_in_roc_bert",
+        ]
+    )
+
     _import_structure["models.time_series_transformer"].extend(
         [
             "TIME_SERIES_TRANSFORMER_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -3383,6 +3401,7 @@ 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.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
     from .models.sew import SEW_PRETRAINED_CONFIG_ARCHIVE_MAP, SEWConfig
@@ -4656,6 +4675,20 @@ if TYPE_CHECKING:
             RobertaModel,
             RobertaPreTrainedModel,
         )
+        from .models.roc_bert import (
+            ROC_BERT_PRETRAINED_MODEL_ARCHIVE_LIST,
+            RoCBertForCausalLM,
+            RoCBertForMaskedLM,
+            RoCBertForMultipleChoice,
+            RoCBertForPreTraining,
+            RoCBertForQuestionAnswering,
+            RoCBertForSequenceClassification,
+            RoCBertForTokenClassification,
+            RoCBertLayer,
+            RoCBertModel,
+            RoCBertPreTrainedModel,
+            load_tf_weights_in_roc_bert,
+        )
         from .models.roformer import (
             ROFORMER_PRETRAINED_MODEL_ARCHIVE_LIST,
             RoFormerForCausalLM,

src/transformers/models/__init__.py

@@ -127,6 +127,7 @@ from . import (
     resnet,
     retribert,
     roberta,
+    roc_bert,
     roformer,
     segformer,
     sew,

src/transformers/models/auto/configuration_auto.py

@@ -123,6 +123,7 @@ CONFIG_MAPPING_NAMES = OrderedDict(
         ("resnet", "ResNetConfig"),
         ("retribert", "RetriBertConfig"),
         ("roberta", "RobertaConfig"),
+        ("roc_bert", "RoCBertConfig"),
         ("roformer", "RoFormerConfig"),
         ("segformer", "SegformerConfig"),
         ("sew", "SEWConfig"),
@@ -257,6 +258,7 @@ CONFIG_ARCHIVE_MAP_MAPPING_NAMES = OrderedDict(
         ("resnet", "RESNET_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("retribert", "RETRIBERT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("roberta", "ROBERTA_PRETRAINED_CONFIG_ARCHIVE_MAP"),
+        ("roc_bert", "ROC_BERT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("roformer", "ROFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("segformer", "SEGFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("sew", "SEW_PRETRAINED_CONFIG_ARCHIVE_MAP"),
@@ -409,6 +411,7 @@ MODEL_NAMES_MAPPING = OrderedDict(
         ("resnet", "ResNet"),
         ("retribert", "RetriBERT"),
         ("roberta", "RoBERTa"),
+        ("roc_bert", "RoCBert"),
         ("roformer", "RoFormer"),
         ("segformer", "SegFormer"),
         ("sew", "SEW"),

src/transformers/models/auto/modeling_auto.py

@@ -121,6 +121,7 @@ MODEL_MAPPING_NAMES = OrderedDict(
         ("resnet", "ResNetModel"),
         ("retribert", "RetriBertModel"),
         ("roberta", "RobertaModel"),
+        ("roc_bert", "RoCBertModel"),
         ("roformer", "RoFormerModel"),
         ("segformer", "SegformerModel"),
         ("sew", "SEWModel"),
@@ -197,6 +198,7 @@ MODEL_FOR_PRETRAINING_MAPPING_NAMES = OrderedDict(
         ("openai-gpt", "OpenAIGPTLMHeadModel"),
         ("retribert", "RetriBertModel"),
         ("roberta", "RobertaForMaskedLM"),
+        ("roc_bert", "RoCBertForPreTraining"),
         ("splinter", "SplinterForPreTraining"),
         ("squeezebert", "SqueezeBertForMaskedLM"),
         ("t5", "T5ForConditionalGeneration"),
@@ -269,6 +271,7 @@ MODEL_WITH_LM_HEAD_MAPPING_NAMES = OrderedDict(
         ("reformer", "ReformerModelWithLMHead"),
         ("rembert", "RemBertForMaskedLM"),
         ("roberta", "RobertaForMaskedLM"),
+        ("roc_bert", "RoCBertForMaskedLM"),
         ("roformer", "RoFormerForMaskedLM"),
         ("speech_to_text", "Speech2TextForConditionalGeneration"),
         ("squeezebert", "SqueezeBertForMaskedLM"),
@@ -320,6 +323,7 @@ MODEL_FOR_CAUSAL_LM_MAPPING_NAMES = OrderedDict(
         ("reformer", "ReformerModelWithLMHead"),
         ("rembert", "RemBertForCausalLM"),
         ("roberta", "RobertaForCausalLM"),
+        ("roc_bert", "RoCBertForCausalLM"),
         ("roformer", "RoFormerForCausalLM"),
         ("speech_to_text_2", "Speech2Text2ForCausalLM"),
         ("transfo-xl", "TransfoXLLMHeadModel"),
@@ -453,6 +457,7 @@ MODEL_FOR_MASKED_LM_MAPPING_NAMES = OrderedDict(
         ("reformer", "ReformerForMaskedLM"),
         ("rembert", "RemBertForMaskedLM"),
         ("roberta", "RobertaForMaskedLM"),
+        ("roc_bert", "RoCBertForMaskedLM"),
         ("roformer", "RoFormerForMaskedLM"),
         ("squeezebert", "SqueezeBertForMaskedLM"),
         ("tapas", "TapasForMaskedLM"),
@@ -573,6 +578,7 @@ MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES = OrderedDict(
         ("reformer", "ReformerForSequenceClassification"),
         ("rembert", "RemBertForSequenceClassification"),
         ("roberta", "RobertaForSequenceClassification"),
+        ("roc_bert", "RoCBertForSequenceClassification"),
         ("roformer", "RoFormerForSequenceClassification"),
         ("squeezebert", "SqueezeBertForSequenceClassification"),
         ("tapas", "TapasForSequenceClassification"),
@@ -628,6 +634,7 @@ MODEL_FOR_QUESTION_ANSWERING_MAPPING_NAMES = OrderedDict(
         ("reformer", "ReformerForQuestionAnswering"),
         ("rembert", "RemBertForQuestionAnswering"),
         ("roberta", "RobertaForQuestionAnswering"),
+        ("roc_bert", "RoCBertForQuestionAnswering"),
         ("roformer", "RoFormerForQuestionAnswering"),
         ("splinter", "SplinterForQuestionAnswering"),
         ("squeezebert", "SqueezeBertForQuestionAnswering"),
@@ -697,6 +704,7 @@ MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING_NAMES = OrderedDict(
         ("qdqbert", "QDQBertForTokenClassification"),
         ("rembert", "RemBertForTokenClassification"),
         ("roberta", "RobertaForTokenClassification"),
+        ("roc_bert", "RoCBertForTokenClassification"),
         ("roformer", "RoFormerForTokenClassification"),
         ("squeezebert", "SqueezeBertForTokenClassification"),
         ("xlm", "XLMForTokenClassification"),
@@ -735,6 +743,7 @@ MODEL_FOR_MULTIPLE_CHOICE_MAPPING_NAMES = OrderedDict(
         ("qdqbert", "QDQBertForMultipleChoice"),
         ("rembert", "RemBertForMultipleChoice"),
         ("roberta", "RobertaForMultipleChoice"),
+        ("roc_bert", "RoCBertForMultipleChoice"),
         ("roformer", "RoFormerForMultipleChoice"),
         ("squeezebert", "SqueezeBertForMultipleChoice"),
         ("xlm", "XLMForMultipleChoice"),

src/transformers/models/roc_bert/__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
+
+# rely on isort to merge the imports
+from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_tokenizers_available, is_torch_available
+
+
+_import_structure = {
+    "configuration_roc_bert": ["ROC_BERT_PRETRAINED_CONFIG_ARCHIVE_MAP", "RoCBertConfig"],
+    "tokenization_roc_bert": ["RoCBertTokenizer"],
+}
+
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    pass
+
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["modeling_roc_bert"] = [
+        "ROC_BERT_PRETRAINED_MODEL_ARCHIVE_LIST",
+        "RoCBertForCausalLM",
+        "RoCBertForMaskedLM",
+        "RoCBertForMultipleChoice",
+        "RoCBertForPreTraining",
+        "RoCBertForQuestionAnswering",
+        "RoCBertForSequenceClassification",
+        "RoCBertForTokenClassification",
+        "RoCBertLayer",
+        "RoCBertModel",
+        "RoCBertPreTrainedModel",
+        "load_tf_weights_in_roc_bert",
+    ]
+
+if TYPE_CHECKING:
+    from .configuration_roc_bert import ROC_BERT_PRETRAINED_CONFIG_ARCHIVE_MAP, RoCBertConfig
+    from .tokenization_roc_bert import RoCBertTokenizer
+
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        raise OptionalDependencyNotAvailable()
+
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .modeling_roc_bert import (
+            ROC_BERT_PRETRAINED_MODEL_ARCHIVE_LIST,
+            RoCBertForCausalLM,
+            RoCBertForMaskedLM,
+            RoCBertForMultipleChoice,
+            RoCBertForPreTraining,
+            RoCBertForQuestionAnswering,
+            RoCBertForSequenceClassification,
+            RoCBertForTokenClassification,
+            RoCBertLayer,
+            RoCBertModel,
+            RoCBertPreTrainedModel,
+            load_tf_weights_in_roc_bert,
+        )
+
+
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)

src/transformers/models/roc_bert/configuration_roc_bert.py

+# coding=utf-8
+# Copyright 2022 WeChatAI 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.
+""" RoCBert model configuration"""
+
+from ...configuration_utils import PretrainedConfig
+from ...utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+ROC_BERT_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "weiweishi/roc-bert-base-zh": "https://huggingface.co/weiweishi/roc-bert-base-zh/resolve/main/config.json",
+}
+
+
+class RoCBertConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`RoCBertModel`]. It is used to instantiate a
+    RoCBert 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 RoCBert
+    [weiweishi/roc-bert-base-zh](https://huggingface.co/weiweishi/roc-bert-base-zh) 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 RoCBert model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`RoCBertModel`].
+        hidden_size (`int`, *optional*, defaults to 768):
+            Dimension 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):
+            Dimension of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
+        hidden_act (`str` or `function`, *optional*, defaults to `"gelu"`):
+            The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
+            `"relu"`, `"selu"` and `"gelu_new"` are supported.
+        hidden_dropout_prob (`float`, *optional*, defaults to 0.1):
+            The dropout probabilitiy 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 [`RoCBertModel`].
+        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.
+        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`.
+        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).
+        classifier_dropout (`float`, *optional*):
+            The dropout ratio for the classification head.
+        enable_cls (`bool`, *optional*, defaults to `True`):
+            Whether or not the model use cls loss when pretrained.
+        enable_pronunciation (`bool`, *optional*, defaults to `True`):
+            Whether or not the model use pronunciation embed when training.
+        enable_shape (`bool`, *optional*, defaults to `True`):
+            Whether or not the model use shape embed when training.
+        pronunciation_embed_dim (`int`, *optional*, defaults to 768):
+            Dimension of the pronunciation_embed.
+        pronunciation_vocab_size (`int`, *optional*, defaults to 910):
+            Pronunciation Vocabulary size of the RoCBert model. Defines the number of different tokens that can be
+            represented by the `input_pronunciation_ids` passed when calling [`RoCBertModel`].
+        shape_embed_dim (`int`, *optional*, defaults to 512):
+            Dimension of the shape_embed.
+        shape_vocab_size (`int`, *optional*, defaults to 24858):
+            Shape Vocabulary size of the RoCBert model. Defines the number of different tokens that can be represented
+            by the `input_shape_ids` passed when calling [`RoCBertModel`].
+        concat_input (`bool`, *optional*, defaults to `True`):
+            Defines the way of merging the shape_embed, pronunciation_embed and word_embed, if the value is true,
+            output_embed = torch.cat((word_embed, shape_embed, pronunciation_embed), -1), else output_embed =
+            (word_embed + shape_embed + pronunciation_embed) / 3
+        Example:
+
+    ```python
+    >>> from transformers import RoCBertModel, RoCBertConfig
+
+    >>> # Initializing a RoCBert weiweishi/roc-bert-base-zh style configuration
+    >>> configuration = RoCBertConfig()
+
+    >>> # Initializing a model from the weiweishi/roc-bert-base-zh style configuration
+    >>> model = RoCBertModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+    model_type = "roc_bert"
+
+    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,
+        use_cache=True,
+        pad_token_id=0,
+        position_embedding_type="absolute",
+        classifier_dropout=None,
+        enable_cls=True,
+        enable_pronunciation=True,
+        enable_shape=True,
+        pronunciation_embed_dim=768,
+        pronunciation_vocab_size=910,
+        shape_embed_dim=512,
+        shape_vocab_size=24858,
+        concat_input=True,
+        **kwargs
+    ):
+        self.vocab_size = vocab_size
+        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
+        self.enable_cls = enable_cls
+        self.enable_pronunciation = enable_pronunciation
+        self.enable_shape = enable_shape
+        self.pronunciation_embed_dim = pronunciation_embed_dim
+        self.pronunciation_vocab_size = pronunciation_vocab_size
+        self.shape_embed_dim = shape_embed_dim
+        self.shape_vocab_size = shape_vocab_size
+        self.concat_input = concat_input
+        self.position_embedding_type = position_embedding_type
+        self.classifier_dropout = classifier_dropout
+        super().__init__(pad_token_id=pad_token_id, **kwargs)

src/transformers/models/roc_bert/modeling_roc_bert.py

+# coding=utf-8
+# Copyright 2022 WeChatAI 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 RoCBert model."""
+
+import math
+import os
+from typing import List, Optional, Tuple, Union
+
+import torch
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
+
+from ...activations import ACT2FN
+from ...modeling_outputs import (
+    BaseModelOutputWithPastAndCrossAttentions,
+    BaseModelOutputWithPoolingAndCrossAttentions,
+    CausalLMOutputWithCrossAttentions,
+    MaskedLMOutput,
+    MultipleChoiceModelOutput,
+    QuestionAnsweringModelOutput,
+    SequenceClassifierOutput,
+    TokenClassifierOutput,
+)
+from ...modeling_utils import PreTrainedModel
+from ...pytorch_utils import apply_chunking_to_forward, find_pruneable_heads_and_indices, prune_linear_layer
+from ...utils import (
+    add_code_sample_docstrings,
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+    logging,
+    replace_return_docstrings,
+)
+from .configuration_roc_bert import RoCBertConfig
+
+
+logger = logging.get_logger(__name__)
+
+_CHECKPOINT_FOR_DOC = "weiweishi/roc-bert-base-zh"
+_CONFIG_FOR_DOC = "RoCBertConfig"
+_TOKENIZER_FOR_DOC = "RoCBertTokenizer"
+
+ROC_BERT_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "weiweishi/roc-bert-base-zh",
+    # See all RoCBert models at https://huggingface.co/models?filter=roc_bert
+]
+
+
+# Copied from transformers.models.bert.modeling_bert.load_tf_weights_in_bert with bert->roc_bert
+def load_tf_weights_in_roc_bert(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
+        if any(
+            n in ["adam_v", "adam_m", "AdamWeightDecayOptimizer", "AdamWeightDecayOptimizer_1", "global_step"]
+            for n in name
+        ):
+            logger.info(f"Skipping {'/'.join(name)}")
+            continue
+        pointer = model
+        for m_name in name:
+            if re.fullmatch(r"[A-Za-z]+_\d+", 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")
+            elif scope_names[0] == "squad":
+                pointer = getattr(pointer, "classifier")
+            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 == "kernel":
+            array = np.transpose(array)
+        try:
+            if pointer.shape != array.shape:
+                raise ValueError(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 RoCBertEmbeddings(nn.Module):
+    """Construct the embeddings from word, position, shape, pronunciation and token_type embeddings."""
+
+    def __init__(self, config):
+        super().__init__()
+        self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id)
+        self.pronunciation_embed = nn.Embedding(
+            config.pronunciation_vocab_size, config.pronunciation_embed_dim, padding_idx=config.pad_token_id
+        )
+        self.shape_embed = nn.Embedding(
+            config.shape_vocab_size, config.shape_embed_dim, padding_idx=config.pad_token_id
+        )
+        self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size)
+        self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.hidden_size)
+
+        self.enable_pronunciation = config.enable_pronunciation
+        self.enable_shape = config.enable_shape
+
+        if config.concat_input:
+            input_dim = config.hidden_size
+            if self.enable_pronunciation:
+                pronunciation_dim = config.pronunciation_embed_dim
+                input_dim += pronunciation_dim
+            if self.enable_shape:
+                shape_dim = config.shape_embed_dim
+                input_dim += shape_dim
+            self.map_inputs_layer = torch.nn.Linear(input_dim, config.hidden_size)
+        else:
+            self.map_inputs_layer = None
+
+        # 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")
+        self.register_buffer(
+            "token_type_ids",
+            torch.zeros(self.position_ids.size(), dtype=torch.long, device=self.position_ids.device),
+            persistent=False,
+        )
+
+    def forward(
+        self,
+        input_ids=None,
+        input_shape_ids=None,
+        input_pronunciation_ids=None,
+        token_type_ids=None,
+        position_ids=None,
+        inputs_embeds=None,
+        past_key_values_length=0,
+    ):
+        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[:, past_key_values_length : seq_length + past_key_values_length]
+
+        # Setting the token_type_ids to the registered buffer in constructor where it is all zeros, which usually occurs
+        # when its auto-generated, registered buffer helps users when tracing the model without passing token_type_ids, solves
+        # issue #5664
+        if token_type_ids is None:
+            if hasattr(self, "token_type_ids"):
+                buffered_token_type_ids = self.token_type_ids[:, :seq_length]
+                buffered_token_type_ids_expanded = buffered_token_type_ids.expand(input_shape[0], seq_length)
+                token_type_ids = buffered_token_type_ids_expanded
+            else:
+                token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=self.position_ids.device)
+
+        if self.map_inputs_layer is None:
+            if inputs_embeds is None:
+                inputs_embeds = self.word_embeddings(input_ids)
+            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.position_embeddings(position_ids)
+                embeddings += position_embeddings
+            embeddings = self.LayerNorm(embeddings)
+            embeddings = self.dropout(embeddings)
+
+            denominator = 1
+            embedding_in = torch.clone(embeddings)
+            if self.enable_shape and input_shape_ids is not None:
+                embedding_shape = self.shape_embed(input_shape_ids)
+                embedding_in += embedding_shape
+                denominator += 1
+            if self.enable_pronunciation and input_pronunciation_ids is not None:
+                embedding_pronunciation = self.pronunciation_embed(input_pronunciation_ids)
+                embedding_in += embedding_pronunciation
+                denominator += 1
+
+            embedding_in /= denominator
+            return embedding_in
+        else:
+            if inputs_embeds is None:
+                inputs_embeds = self.word_embeddings(input_ids)  # embedding_word
+            device = inputs_embeds.device
+
+            embedding_in = torch.clone(inputs_embeds)
+            if self.enable_shape:
+                if input_shape_ids is None:
+                    input_shape_ids = torch.zeros(input_shape, dtype=torch.long, device=device)
+                embedding_shape = self.shape_embed(input_shape_ids)
+                embedding_in = torch.cat((embedding_in, embedding_shape), -1)
+            if self.enable_pronunciation:
+                if input_pronunciation_ids is None:
+                    input_pronunciation_ids = torch.zeros(input_shape, dtype=torch.long, device=device)
+                embedding_pronunciation = self.pronunciation_embed(input_pronunciation_ids)
+                embedding_in = torch.cat((embedding_in, embedding_pronunciation), -1)
+
+            embedding_in = self.map_inputs_layer(embedding_in)  # batch_size * seq_len * hidden_dim
+
+            token_type_embeddings = self.token_type_embeddings(token_type_ids)
+            embedding_in += token_type_embeddings
+            if self.position_embedding_type == "absolute":
+                position_embeddings = self.position_embeddings(position_ids)
+                embedding_in += position_embeddings
+
+            embedding_in = self.LayerNorm(embedding_in)
+            embedding_in = self.dropout(embedding_in)
+            return embedding_in
+
+
+# Copied from transformers.models.bert.modeling_bert.BertSelfAttention with Bert->RoCBert
+class RoCBertSelfAttention(nn.Module):
+    def __init__(self, config, position_embedding_type=None):
+        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 = position_embedding_type or 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)
+
+        self.is_decoder = config.is_decoder
+
+    def transpose_for_scores(self, x: torch.Tensor) -> torch.Tensor:
+        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,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        head_mask: Optional[torch.FloatTensor] = None,
+        encoder_hidden_states: Optional[torch.FloatTensor] = None,
+        encoder_attention_mask: Optional[torch.FloatTensor] = None,
+        past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
+        output_attentions: Optional[bool] = False,
+    ) -> Tuple[torch.Tensor]:
+        mixed_query_layer = self.query(hidden_states)
+
+        # 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.
+        is_cross_attention = encoder_hidden_states is not None
+
+        if is_cross_attention and past_key_value is not None:
+            # reuse k,v, cross_attentions
+            key_layer = past_key_value[0]
+            value_layer = past_key_value[1]
+            attention_mask = encoder_attention_mask
+        elif is_cross_attention:
+            key_layer = self.transpose_for_scores(self.key(encoder_hidden_states))
+            value_layer = self.transpose_for_scores(self.value(encoder_hidden_states))
+            attention_mask = encoder_attention_mask
+        elif past_key_value is not None:
+            key_layer = self.transpose_for_scores(self.key(hidden_states))
+            value_layer = self.transpose_for_scores(self.value(hidden_states))
+            key_layer = torch.cat([past_key_value[0], key_layer], dim=2)
+            value_layer = torch.cat([past_key_value[1], value_layer], dim=2)
+        else:
+            key_layer = self.transpose_for_scores(self.key(hidden_states))
+            value_layer = self.transpose_for_scores(self.value(hidden_states))
+
+        query_layer = self.transpose_for_scores(mixed_query_layer)
+
+        if self.is_decoder:
+            # if cross_attention save Tuple(torch.Tensor, torch.Tensor) of all cross attention key/value_states.
+            # Further calls to cross_attention layer can then reuse all cross-attention
+            # key/value_states (first "if" case)
+            # if uni-directional self-attention (decoder) save Tuple(torch.Tensor, torch.Tensor) of
+            # all previous decoder key/value_states. Further calls to uni-directional self-attention
+            # can concat previous decoder key/value_states to current projected key/value_states (third "elif" case)
+            # if encoder bi-directional self-attention `past_key_value` is always `None`
+            past_key_value = (key_layer, value_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 = hidden_states.size()[1]
+            position_ids_l = torch.arange(seq_length, dtype=torch.long, device=hidden_states.device).view(-1, 1)
+            position_ids_r = torch.arange(seq_length, dtype=torch.long, device=hidden_states.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:
+            # Apply the attention mask is (precomputed for all layers in RoCBertModel forward() function)
+            attention_scores = attention_scores + attention_mask
+
+        # Normalize the attention scores to probabilities.
+        attention_probs = nn.functional.softmax(attention_scores, dim=-1)
+
+        # 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,)
+
+        if self.is_decoder:
+            outputs = outputs + (past_key_value,)
+        return outputs
+
+
+# Copied from transformers.models.bert.modeling_bert.BertSelfOutput with Bert->RoCBert
+class RoCBertSelfOutput(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: torch.Tensor, input_tensor: torch.Tensor) -> torch.Tensor:
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states + input_tensor)
+        return hidden_states
+
+
+# Copied from transformers.models.bert.modeling_bert.BertAttention with Bert->RoCBert
+class RoCBertAttention(nn.Module):
+    def __init__(self, config, position_embedding_type=None):
+        super().__init__()
+        self.self = RoCBertSelfAttention(config, position_embedding_type=position_embedding_type)
+        self.output = RoCBertSelfOutput(config)
+        self.pruned_heads = set()
+
+    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: torch.Tensor,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        head_mask: Optional[torch.FloatTensor] = None,
+        encoder_hidden_states: Optional[torch.FloatTensor] = None,
+        encoder_attention_mask: Optional[torch.FloatTensor] = None,
+        past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
+        output_attentions: Optional[bool] = False,
+    ) -> Tuple[torch.Tensor]:
+        self_outputs = self.self(
+            hidden_states,
+            attention_mask,
+            head_mask,
+            encoder_hidden_states,
+            encoder_attention_mask,
+            past_key_value,
+            output_attentions,
+        )
+        attention_output = self.output(self_outputs[0], hidden_states)
+        outputs = (attention_output,) + self_outputs[1:]  # add attentions if we output them
+        return outputs
+
+
+# Copied from transformers.models.bert.modeling_bert.BertIntermediate with Bert->RoCBert
+class RoCBertIntermediate(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: torch.Tensor) -> torch.Tensor:
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.intermediate_act_fn(hidden_states)
+        return hidden_states
+
+
+# Copied from transformers.models.bert.modeling_bert.BertOutput with Bert->RoCBert
+class RoCBertOutput(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: torch.Tensor, input_tensor: torch.Tensor) -> torch.Tensor:
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states + input_tensor)
+        return hidden_states
+
+
+# Copied from transformers.models.bert.modeling_bert.BertLayer with Bert->RoCBert
+class RoCBertLayer(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.chunk_size_feed_forward = config.chunk_size_feed_forward
+        self.seq_len_dim = 1
+        self.attention = RoCBertAttention(config)
+        self.is_decoder = config.is_decoder
+        self.add_cross_attention = config.add_cross_attention
+        if self.add_cross_attention:
+            if not self.is_decoder:
+                raise ValueError(f"{self} should be used as a decoder model if cross attention is added")
+            self.crossattention = RoCBertAttention(config, position_embedding_type="absolute")
+        self.intermediate = RoCBertIntermediate(config)
+        self.output = RoCBertOutput(config)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        head_mask: Optional[torch.FloatTensor] = None,
+        encoder_hidden_states: Optional[torch.FloatTensor] = None,
+        encoder_attention_mask: Optional[torch.FloatTensor] = None,
+        past_key_value: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
+        output_attentions: Optional[bool] = False,
+    ) -> Tuple[torch.Tensor]:
+        # decoder uni-directional self-attention cached key/values tuple is at positions 1,2
+        self_attn_past_key_value = past_key_value[:2] if past_key_value is not None else None
+        self_attention_outputs = self.attention(
+            hidden_states,
+            attention_mask,
+            head_mask,
+            output_attentions=output_attentions,
+            past_key_value=self_attn_past_key_value,
+        )
+        attention_output = self_attention_outputs[0]
+
+        # if decoder, the last output is tuple of self-attn cache
+        if self.is_decoder:
+            outputs = self_attention_outputs[1:-1]
+            present_key_value = self_attention_outputs[-1]
+        else:
+            outputs = self_attention_outputs[1:]  # add self attentions if we output attention weights
+
+        cross_attn_present_key_value = None
+        if self.is_decoder and encoder_hidden_states is not None:
+            if not hasattr(self, "crossattention"):
+                raise ValueError(
+                    f"If `encoder_hidden_states` are passed, {self} has to be instantiated with cross-attention layers"
+                    " by setting `config.add_cross_attention=True`"
+                )
+
+            # cross_attn cached key/values tuple is at positions 3,4 of past_key_value tuple
+            cross_attn_past_key_value = past_key_value[-2:] if past_key_value is not None else None
+            cross_attention_outputs = self.crossattention(
+                attention_output,
+                attention_mask,
+                head_mask,
+                encoder_hidden_states,
+                encoder_attention_mask,
+                cross_attn_past_key_value,
+                output_attentions,
+            )
+            attention_output = cross_attention_outputs[0]
+            outputs = outputs + cross_attention_outputs[1:-1]  # add cross attentions if we output attention weights
+
+            # add cross-attn cache to positions 3,4 of present_key_value tuple
+            cross_attn_present_key_value = cross_attention_outputs[-1]
+            present_key_value = present_key_value + cross_attn_present_key_value
+
+        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
+
+        # if decoder, return the attn key/values as the last output
+        if self.is_decoder:
+            outputs = outputs + (present_key_value,)
+
+        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
+
+
+# Copied from transformers.models.bert.modeling_bert.BertEncoder with Bert->RoCBert
+class RoCBertEncoder(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.layer = nn.ModuleList([RoCBertLayer(config) for _ in range(config.num_hidden_layers)])
+        self.gradient_checkpointing = False
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        head_mask: Optional[torch.FloatTensor] = None,
+        encoder_hidden_states: Optional[torch.FloatTensor] = None,
+        encoder_attention_mask: Optional[torch.FloatTensor] = None,
+        past_key_values: Optional[Tuple[Tuple[torch.FloatTensor]]] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = False,
+        output_hidden_states: Optional[bool] = False,
+        return_dict: Optional[bool] = True,
+    ) -> Union[Tuple[torch.Tensor], BaseModelOutputWithPastAndCrossAttentions]:
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attentions = () if output_attentions else None
+        all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
+
+        next_decoder_cache = () if use_cache 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
+            past_key_value = past_key_values[i] if past_key_values is not None else None
+
+            if self.gradient_checkpointing and self.training:
+
+                if use_cache:
+                    logger.warning(
+                        "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                    )
+                    use_cache = False
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        return module(*inputs, past_key_value, output_attentions)
+
+                    return custom_forward
+
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(layer_module),
+                    hidden_states,
+                    attention_mask,
+                    layer_head_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                )
+            else:
+                layer_outputs = layer_module(
+                    hidden_states,
+                    attention_mask,
+                    layer_head_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                    past_key_value,
+                    output_attentions,
+                )
+
+            hidden_states = layer_outputs[0]
+            if use_cache:
+                next_decoder_cache += (layer_outputs[-1],)
+            if output_attentions:
+                all_self_attentions = all_self_attentions + (layer_outputs[1],)
+                if self.config.add_cross_attention:
+                    all_cross_attentions = all_cross_attentions + (layer_outputs[2],)
+
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(
+                v
+                for v in [
+                    hidden_states,
+                    next_decoder_cache,
+                    all_hidden_states,
+                    all_self_attentions,
+                    all_cross_attentions,
+                ]
+                if v is not None
+            )
+        return BaseModelOutputWithPastAndCrossAttentions(
+            last_hidden_state=hidden_states,
+            past_key_values=next_decoder_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+            cross_attentions=all_cross_attentions,
+        )
+
+
+# Copied from transformers.models.bert.modeling_bert.BertPooler with Bert->RoCBert
+class RoCBertPooler(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: torch.Tensor) -> torch.Tensor:
+        # 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
+
+
+# Copied from transformers.models.bert.modeling_bert.BertPredictionHeadTransform with Bert->RoCBert
+class RoCBertPredictionHeadTransform(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: torch.Tensor) -> torch.Tensor:
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.transform_act_fn(hidden_states)
+        hidden_states = self.LayerNorm(hidden_states)
+        return hidden_states
+
+
+# Copied from transformers.models.bert.modeling_bert.BertLMPredictionHead with Bert->RoCBert
+class RoCBertLMPredictionHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.transform = RoCBertPredictionHeadTransform(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
+
+
+# Copied from transformers.models.bert.modeling_bert.BertOnlyMLMHead with Bert->RoCBert
+class RoCBertOnlyMLMHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.predictions = RoCBertLMPredictionHead(config)
+
+    def forward(self, sequence_output: torch.Tensor) -> torch.Tensor:
+        prediction_scores = self.predictions(sequence_output)
+        return prediction_scores
+
+
+# Copied from transformers.models.bert.modeling_bert.BertPreTrainedModel with Bert->RoCBert,bert->roc_bert
+class RoCBertPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = RoCBertConfig
+    load_tf_weights = load_tf_weights_in_roc_bert
+    base_model_prefix = "roc_bert"
+    supports_gradient_checkpointing = True
+    _keys_to_ignore_on_load_missing = [r"position_ids"]
+
+    def _init_weights(self, module):
+        """Initialize the weights"""
+        if isinstance(module, nn.Linear):
+            # 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)
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, RoCBertEncoder):
+            module.gradient_checkpointing = value
+
+
+ROC_BERT_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 ([`RoCBertConfig`]): 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 [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+ROC_BERT_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `({0})`):
+            Indices of input sequence tokens in the vocabulary.
+
+            Indices can be obtained using [`RoCBertTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        input_shape_ids (`torch.LongTensor` of shape `({0})`):
+            Indices of input sequence tokens in the shape vocabulary.
+
+            Indices can be obtained using [`RoCBertTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input_shape_ids)
+        input_pronunciation_ids (`torch.LongTensor` of shape `({0})`):
+            Indices of input sequence tokens in the pronunciation vocabulary.
+
+            Indices can be obtained using [`RoCBertTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input_pronunciation_ids)
+        attention_mask (`torch.FloatTensor` of shape `({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#attention-mask)
+        token_type_ids (`torch.LongTensor` of shape `({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#token-type-ids)
+        position_ids (`torch.LongTensor` of shape `({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#position-ids)
+        head_mask (`torch.FloatTensor` of shape `(num_heads,)` or `(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 (`torch.FloatTensor` of shape `({0}, hidden_size)`, *optional*):
+            Optionally, instead of passing `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 (`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 (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+@add_start_docstrings(
+    "The bare RoCBert Model transformer outputting raw hidden-states without any specific head on top.",
+    ROC_BERT_START_DOCSTRING,
+)
+class RoCBertModel(RoCBertPreTrainedModel):
+    """
+
+    The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of
+    cross-attention is added between the self-attention layers, following the architecture described in [Attention is
+    all you need](https://arxiv.org/abs/1706.03762) by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit,
+    Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.
+
+    To behave as an decoder the model needs to be initialized with the `is_decoder` argument of the configuration set
+    to `True`. To be used in a Seq2Seq model, the model needs to be initialized with both `is_decoder` argument and
+    `add_cross_attention` set to `True`; an `encoder_hidden_states` is then expected as an input to the forward pass.
+    """
+
+    # Copied from transformers.models.bert.modeling_bert.BertModel.__init__ with Bert->RoCBert
+    def __init__(self, config, add_pooling_layer=True):
+        super().__init__(config)
+        self.config = config
+
+        self.embeddings = RoCBertEmbeddings(config)
+        self.encoder = RoCBertEncoder(config)
+
+        self.pooler = RoCBertPooler(config) if add_pooling_layer else None
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    # Copied from transformers.models.bert.modeling_bert.BertModel.get_input_embeddings
+    def get_input_embeddings(self):
+        return self.embeddings.word_embeddings
+
+    # Copied from transformers.models.bert.modeling_bert.BertModel.set_input_embeddings
+    def set_input_embeddings(self, value):
+        self.embeddings.word_embeddings = value
+
+    def get_pronunciation_embeddings(self):
+        return self.embeddings.pronunciation_embed
+
+    def set_pronunciation_embeddings(self, value):
+        self.embeddings.pronunciation_embed = value
+
+    def get_shape_embeddings(self):
+        return self.embeddings.shape_embed
+
+    def set_shape_embeddings(self, value):
+        self.embeddings.shape_embed = value
+
+    # Copied from transformers.models.bert.modeling_bert.BertModel._prune_heads
+    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)
+
+    @add_start_docstrings_to_model_forward(ROC_BERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_code_sample_docstrings(
+        processor_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=BaseModelOutputWithPoolingAndCrossAttentions,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        input_shape_ids: Optional[torch.Tensor] = None,
+        input_pronunciation_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        encoder_attention_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple[torch.Tensor], BaseModelOutputWithPoolingAndCrossAttentions]:
+        r"""
+        encoder_hidden_states  (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
+            the model is configured as a decoder.
+        encoder_attention_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
+            the cross-attention if the model is configured as a decoder. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+        past_key_values (`tuple(tuple(torch.FloatTensor))` of length `config.n_layers` with each tuple having 4 tensors of shape `(batch_size, num_heads, sequence_length - 1, embed_size_per_head)`):
+            Contains precomputed key and value hidden states of the attention blocks. Can be used to speed up decoding.
+            If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those that
+            don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of all
+            `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+        use_cache (`bool`, *optional*):
+            If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
+            `past_key_values`).
+        """
+        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
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if self.config.is_decoder:
+            use_cache = use_cache if use_cache is not None else self.config.use_cache
+        else:
+            use_cache = False
+
+        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()
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.size()[:-1]
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        batch_size, seq_length = input_shape
+        device = input_ids.device if input_ids is not None else inputs_embeds.device
+
+        # past_key_values_length
+        past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0
+
+        if attention_mask is None:
+            attention_mask = torch.ones(((batch_size, seq_length + past_key_values_length)), device=device)
+
+        if token_type_ids is None:
+            if hasattr(self.embeddings, "token_type_ids"):
+                buffered_token_type_ids = self.embeddings.token_type_ids[:, :seq_length]
+                buffered_token_type_ids_expanded = buffered_token_type_ids.expand(batch_size, seq_length)
+                token_type_ids = buffered_token_type_ids_expanded
+            else:
+                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)
+
+        # If a 2D or 3D attention mask is provided for the cross-attention
+        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
+        if self.config.is_decoder and encoder_hidden_states is not None:
+            encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
+            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
+            if encoder_attention_mask is None:
+                encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
+            encoder_extended_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+        else:
+            encoder_extended_attention_mask = None
+
+        # 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)
+
+        embedding_output = self.embeddings(
+            input_ids=input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_ids,
+            position_ids=position_ids,
+            token_type_ids=token_type_ids,
+            inputs_embeds=inputs_embeds,
+            past_key_values_length=past_key_values_length,
+        )
+        encoder_outputs = self.encoder(
+            embedding_output,
+            attention_mask=extended_attention_mask,
+            head_mask=head_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_extended_attention_mask,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        sequence_output = encoder_outputs[0]
+        pooled_output = self.pooler(sequence_output) if self.pooler is not None else None
+
+        if not return_dict:
+            return (sequence_output, pooled_output) + encoder_outputs[1:]
+
+        return BaseModelOutputWithPoolingAndCrossAttentions(
+            last_hidden_state=sequence_output,
+            pooler_output=pooled_output,
+            past_key_values=encoder_outputs.past_key_values,
+            hidden_states=encoder_outputs.hidden_states,
+            attentions=encoder_outputs.attentions,
+            cross_attentions=encoder_outputs.cross_attentions,
+        )
+
+
+@add_start_docstrings(
+    """
+    RoCBert Model with contrastive loss and masked_lm_loss during the pretraining.
+    """,
+    ROC_BERT_START_DOCSTRING,
+)
+class RoCBertForPreTraining(RoCBertPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.roc_bert = RoCBertModel(config)
+        self.cls = RoCBertOnlyMLMHead(config)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    # Copied from transformers.models.bert.modeling_bert.BertForPreTraining.get_output_embeddings
+    def get_output_embeddings(self):
+        return self.cls.predictions.decoder
+
+    # Copied from transformers.models.bert.modeling_bert.BertForPreTraining.set_output_embeddings
+    def set_output_embeddings(self, new_embeddings):
+        self.cls.predictions.decoder = new_embeddings
+
+    @add_start_docstrings_to_model_forward(ROC_BERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @replace_return_docstrings(output_type=MaskedLMOutput, config_class=_CONFIG_FOR_DOC)
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        input_shape_ids: Optional[torch.Tensor] = None,
+        input_pronunciation_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        attack_input_ids: Optional[torch.Tensor] = None,
+        attack_input_shape_ids: Optional[torch.Tensor] = None,
+        attack_input_pronunciation_ids: Optional[torch.Tensor] = None,
+        attack_attention_mask: Optional[torch.Tensor] = None,
+        attack_token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        labels_input_ids: Optional[torch.Tensor] = None,
+        labels_input_shape_ids: Optional[torch.Tensor] = None,
+        labels_input_pronunciation_ids: Optional[torch.Tensor] = None,
+        labels_attention_mask: Optional[torch.Tensor] = None,
+        labels_token_type_ids: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        **kwargs,
+    ) -> Union[Tuple[torch.Tensor], MaskedLMOutput]:
+        r"""
+            attack_input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                attack sample ids for computing the contrastive loss. Indices should be in `[-100, 0, ...,
+                config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored (masked),
+                the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`
+            attack_input_shape_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                attack sample shape ids for computing the contrastive loss. Indices should be in `[-100, 0, ...,
+                config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored (masked),
+                the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`
+            attack_input_pronunciation_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                attack sample pronunciation ids for computing the contrastive loss. Indices should be in `[-100, 0,
+                ..., config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored
+                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`
+            labels_input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                target ids for computing the contrastive loss and masked_lm_loss . Indices should be in `[-100, 0, ...,
+                config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored (masked),
+                the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`
+            labels_input_shape_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                target shape ids for computing the contrastive loss and masked_lm_loss . Indices should be in `[-100,
+                0, ..., config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored
+                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`
+            labels_input_pronunciation_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                target pronunciation ids for computing the contrastive loss and masked_lm_loss . Indices should be in
+                `[-100, 0, ..., config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are
+                 ignored (masked), the loss is only computed for the tokens with labels in `[0, ...,
+                 config.vocab_size]`
+
+            kwargs (`Dict[str, any]`, optional, defaults to *{}*):
+                Used to hide legacy arguments that have been deprecated.
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from transformers import RoCBertTokenizer, RoCBertForPreTraining
+        >>> import torch
+
+        >>> tokenizer = RoCBertTokenizer.from_pretrained("weiweishi/roc-bert-base-zh")
+        >>> model = RoCBertForPreTraining.from_pretrained("weiweishi/roc-bert-base-zh")
+
+        >>> inputs = tokenizer("你好，很高兴认识你", return_tensors="pt")
+        >>> attack_inputs = tokenizer("你号，很高兴认识你", return_tensors="pt")
+        >>> attack_keys = list(attack_inputs.keys())
+        >>> for key in attack_keys:
+        ...     attack_inputs[f"attack_{key}"] = attack_inputs.pop(key)
+        >>> label_inputs = tokenizer("你好，很高兴认识你", return_tensors="pt")
+        >>> label_keys = list(attack_inputs.keys())
+        >>> for key in label_keys:
+        ...     label_inputs[f"labels_{key}"] = label_inputs.pop(key)
+
+        >>> inputs.update(label_inputs)
+        >>> inputs.update(attack_inputs)
+        >>> outputs = model(**inputs)
+
+        >>> logits = outputs.logits
+        ```
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.roc_bert(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_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, pooled_output = outputs[:2]
+        prediction_scores = self.cls(sequence_output)
+
+        loss = None
+        if labels_input_ids is not None:
+            loss_fct = CrossEntropyLoss()  # -100 index = padding token
+            masked_lm_loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), labels_input_ids.view(-1))
+
+            if attack_input_ids is not None:
+                batch_size, _ = labels_input_ids.shape
+                device = labels_input_ids.device
+
+                target_inputs = torch.clone(labels_input_ids)
+                target_inputs[target_inputs == -100] = self.config.pad_token_id
+
+                labels_output = self.roc_bert(
+                    target_inputs,
+                    input_shape_ids=labels_input_shape_ids,
+                    input_pronunciation_ids=labels_input_pronunciation_ids,
+                    attention_mask=labels_attention_mask,
+                    token_type_ids=labels_token_type_ids,
+                    return_dict=return_dict,
+                )
+                attack_output = self.roc_bert(
+                    attack_input_ids,
+                    input_shape_ids=attack_input_shape_ids,
+                    input_pronunciation_ids=attack_input_pronunciation_ids,
+                    attention_mask=attack_attention_mask,
+                    token_type_ids=attack_token_type_ids,
+                    return_dict=return_dict,
+                )
+
+                labels_pooled_output = labels_output[1]
+                attack_pooled_output = attack_output[1]
+
+                pooled_output_norm = torch.nn.functional.normalize(pooled_output, dim=-1)
+                labels_pooled_output_norm = torch.nn.functional.normalize(labels_pooled_output, dim=-1)
+                attack_pooled_output_norm = torch.nn.functional.normalize(attack_pooled_output, dim=-1)
+
+                sim_matrix = torch.matmul(pooled_output_norm, attack_pooled_output_norm.T)  # batch_size * hidden_dim
+                sim_matrix_target = torch.matmul(labels_pooled_output_norm, attack_pooled_output_norm.T)
+                batch_labels = torch.tensor([i for i in range(batch_size)], device=device)
+                contrastive_loss = (
+                    loss_fct(100 * sim_matrix.view(batch_size, -1), batch_labels.view(-1))
+                    + loss_fct(100 * sim_matrix_target.view(batch_size, -1), batch_labels.view(-1))
+                ) / 2
+
+                loss = contrastive_loss + masked_lm_loss
+            else:
+                loss = masked_lm_loss
+
+        if not return_dict:
+            output = (prediction_scores,) + outputs[2:]
+            return ((loss,) + output) if loss is not None else output
+
+        return MaskedLMOutput(
+            loss=loss,
+            logits=prediction_scores,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+@add_start_docstrings("""RoCBert Model with a `language modeling` head on top.""", ROC_BERT_START_DOCSTRING)
+class RoCBertForMaskedLM(RoCBertPreTrainedModel):
+    _keys_to_ignore_on_load_unexpected = [r"pooler"]
+    _keys_to_ignore_on_load_missing = [r"position_ids", r"predictions.decoder.bias"]
+
+    # Copied from transformers.models.bert.modeling_bert.BertForMaskedLM.__init__ with Bert->RoCBert,bert->roc_bert
+    def __init__(self, config):
+        super().__init__(config)
+
+        if config.is_decoder:
+            logger.warning(
+                "If you want to use `RoCBertForMaskedLM` make sure `config.is_decoder=False` for "
+                "bi-directional self-attention."
+            )
+
+        self.roc_bert = RoCBertModel(config, add_pooling_layer=False)
+        self.cls = RoCBertOnlyMLMHead(config)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    # Copied from transformers.models.bert.modeling_bert.BertForMaskedLM.get_output_embeddings
+    def get_output_embeddings(self):
+        return self.cls.predictions.decoder
+
+    # Copied from transformers.models.bert.modeling_bert.BertForMaskedLM.set_output_embeddings
+    def set_output_embeddings(self, new_embeddings):
+        self.cls.predictions.decoder = new_embeddings
+
+    @add_start_docstrings_to_model_forward(ROC_BERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_code_sample_docstrings(
+        processor_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=MaskedLMOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        input_shape_ids: Optional[torch.Tensor] = None,
+        input_pronunciation_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        encoder_attention_mask: Optional[torch.Tensor] = None,
+        labels: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple[torch.Tensor], MaskedLMOutput]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should be in `[-100, 0, ...,
+            config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored (masked), the
+            loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.roc_bert(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        sequence_output = outputs[0]
+        prediction_scores = self.cls(sequence_output)
+
+        masked_lm_loss = None
+        if labels is not None:
+            loss_fct = CrossEntropyLoss()  # -100 index = padding token
+            masked_lm_loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), labels.view(-1))
+
+        if not return_dict:
+            output = (prediction_scores,) + outputs[2:]
+            return ((masked_lm_loss,) + output) if masked_lm_loss is not None else output
+
+        return MaskedLMOutput(
+            loss=masked_lm_loss,
+            logits=prediction_scores,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+    def prepare_inputs_for_generation(
+        self, input_ids, input_shape_ids=None, input_pronunciation_ids=None, attention_mask=None, **model_kwargs
+    ):
+        input_shape = input_ids.shape
+        effective_batch_size = input_shape[0]
+
+        #  add a dummy token
+        if self.config.pad_token_id is None:
+            raise ValueError("The PAD token should be defined for generation")
+
+        attention_mask = torch.cat([attention_mask, attention_mask.new_zeros((attention_mask.shape[0], 1))], dim=-1)
+        dummy_token = torch.full(
+            (effective_batch_size, 1), self.config.pad_token_id, dtype=torch.long, device=input_ids.device
+        )
+        input_ids = torch.cat([input_ids, dummy_token], dim=1)
+        if input_shape_ids is not None:
+            input_shape_ids = torch.cat([input_shape_ids, dummy_token], dim=1)
+        if input_pronunciation_ids is not None:
+            input_pronunciation_ids = torch.cat([input_pronunciation_ids, dummy_token], dim=1)
+
+        return {
+            "input_ids": input_ids,
+            "input_shape_ids": input_shape_ids,
+            "input_pronunciation_ids": input_pronunciation_ids,
+            "attention_mask": attention_mask,
+        }
+
+
+@add_start_docstrings(
+    """RoCBert Model with a `language modeling` head on top for CLM fine-tuning.""", ROC_BERT_START_DOCSTRING
+)
+class RoCBertForCausalLM(RoCBertPreTrainedModel):
+    _keys_to_ignore_on_load_unexpected = [r"pooler"]
+    _keys_to_ignore_on_load_missing = [r"position_ids", r"predictions.decoder.bias"]
+
+    # Copied from transformers.models.bert.modeling_bert.BertLMHeadModel.__init__ with BertLMHeadModel->RoCBertForCausalLM,Bert->RoCBert,bert->roc_bert
+    def __init__(self, config):
+        super().__init__(config)
+
+        if not config.is_decoder:
+            logger.warning("If you want to use `RoCRoCBertForCausalLM` as a standalone, add `is_decoder=True.`")
+
+        self.roc_bert = RoCBertModel(config, add_pooling_layer=False)
+        self.cls = RoCBertOnlyMLMHead(config)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    # Copied from transformers.models.bert.modeling_bert.BertLMHeadModel.get_output_embeddings
+    def get_output_embeddings(self):
+        return self.cls.predictions.decoder
+
+    # Copied from transformers.models.bert.modeling_bert.BertLMHeadModel.set_output_embeddings
+    def set_output_embeddings(self, new_embeddings):
+        self.cls.predictions.decoder = new_embeddings
+
+    @add_start_docstrings_to_model_forward(ROC_BERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @replace_return_docstrings(output_type=CausalLMOutputWithCrossAttentions, config_class=_CONFIG_FOR_DOC)
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        input_shape_ids: Optional[torch.Tensor] = None,
+        input_pronunciation_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        encoder_attention_mask: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[List[torch.Tensor]] = None,
+        labels: Optional[torch.Tensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple[torch.Tensor], CausalLMOutputWithCrossAttentions]:
+        r"""
+        encoder_hidden_states  (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if
+            the model is configured as a decoder.
+        encoder_attention_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on the padding token indices of the encoder input. This mask is used in
+            the cross-attention if the model is configured as a decoder. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+        past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
+            `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and 2 additional tensors of shape
+            `(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)`. The two additional tensors are
+            only required when the model is used as a decoder in a Sequence to Sequence model.
+
+            Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
+            blocks) that can be used (see `past_key_values` input) to speed up sequential decoding.
+
+            If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those that
+            don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of all
+            `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the left-to-right language modeling loss (next word prediction). Indices should be in
+            `[-100, 0, ..., config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are
+            ignored (masked), the loss is only computed for the tokens with labels n `[0, ..., config.vocab_size]`.
+        use_cache (`bool`, *optional*):
+            If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
+            `past_key_values`).
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from transformers import RoCBertTokenizer, RoCBertForCausalLM, RoCBertConfig
+        >>> import torch
+
+        >>> tokenizer = RoCBertTokenizer.from_pretrained("weiweishi/roc-bert-base-zh")
+        >>> config = RoCBertConfig.from_pretrained("weiweishi/roc-bert-base-zh")
+        >>> config.is_decoder = True
+        >>> model = RoCBertForCausalLM.from_pretrained("weiweishi/roc-bert-base-zh", config=config)
+
+        >>> inputs = tokenizer("你好，很高兴认识你", return_tensors="pt")
+        >>> outputs = model(**inputs)
+
+        >>> prediction_logits = outputs.logits
+        ```"""
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.roc_bert(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        sequence_output = outputs[0]
+        prediction_scores = self.cls(sequence_output)
+
+        lm_loss = None
+        if labels is not None:
+            # we are doing next-token prediction; shift prediction scores and input ids by one
+            shifted_prediction_scores = prediction_scores[:, :-1, :].contiguous()
+            labels = labels[:, 1:].contiguous()
+            loss_fct = CrossEntropyLoss()
+            lm_loss = loss_fct(shifted_prediction_scores.view(-1, self.config.vocab_size), labels.view(-1))
+
+        if not return_dict:
+            output = (prediction_scores,) + outputs[2:]
+            return ((lm_loss,) + output) if lm_loss is not None else output
+
+        return CausalLMOutputWithCrossAttentions(
+            loss=lm_loss,
+            logits=prediction_scores,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+            cross_attentions=outputs.cross_attentions,
+        )
+
+    def prepare_inputs_for_generation(
+        self,
+        input_ids,
+        input_shape_ids=None,
+        input_pronunciation_ids=None,
+        past=None,
+        attention_mask=None,
+        **model_kwargs
+    ):
+        input_shape = input_ids.shape
+
+        # if model is used as a decoder in encoder-decoder model, the decoder attention mask is created on the fly
+        if attention_mask is None:
+            attention_mask = input_ids.new_ones(input_shape)
+
+        # cut decoder_input_ids if past is used
+        if past is not None:
+            input_ids = input_ids[:, -1:]
+            if input_shape_ids is not None:
+                input_shape_ids = input_shape_ids[:, -1:]
+            if input_pronunciation_ids is not None:
+                input_pronunciation_ids = input_pronunciation_ids[:, -1:]
+
+        return {
+            "input_ids": input_ids,
+            "input_shape_ids": input_shape_ids,
+            "input_pronunciation_ids": input_pronunciation_ids,
+            "attention_mask": attention_mask,
+            "past_key_values": past,
+        }
+
+    # Copied from transformers.models.bert.modeling_bert.BertLMHeadModel._reorder_cache
+    def _reorder_cache(self, past, beam_idx):
+        reordered_past = ()
+        for layer_past in past:
+            reordered_past += (tuple(past_state.index_select(0, beam_idx) for past_state in layer_past),)
+        return reordered_past
+
+
+@add_start_docstrings(
+    """RoCBert Model transformer with a sequence classification/regression head on top (a linear layer on top of
+    the pooled output) e.g. for GLUE tasks.""",
+    ROC_BERT_START_DOCSTRING,
+)
+class RoCBertForSequenceClassification(RoCBertPreTrainedModel):
+    # Copied from transformers.models.bert.modeling_bert.BertForSequenceClassification.__init__ with Bert->RoCBert,bert->roc_bert
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+        self.config = config
+
+        self.roc_bert = RoCBertModel(config)
+        classifier_dropout = (
+            config.classifier_dropout if config.classifier_dropout is not None else config.hidden_dropout_prob
+        )
+        self.dropout = nn.Dropout(classifier_dropout)
+        self.classifier = nn.Linear(config.hidden_size, config.num_labels)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @add_start_docstrings_to_model_forward(ROC_BERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_code_sample_docstrings(
+        processor_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=SequenceClassifierOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        input_shape_ids: Optional[torch.Tensor] = None,
+        input_pronunciation_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        labels: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple[torch.Tensor], SequenceClassifierOutput]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
+            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
+            `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.roc_bert(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_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(
+    """RoCBert 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.""",
+    ROC_BERT_START_DOCSTRING,
+)
+class RoCBertForMultipleChoice(RoCBertPreTrainedModel):
+    # Copied from transformers.models.bert.modeling_bert.BertForMultipleChoice.__init__ with Bert->RoCBert,bert->roc_bert
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.roc_bert = RoCBertModel(config)
+        classifier_dropout = (
+            config.classifier_dropout if config.classifier_dropout is not None else config.hidden_dropout_prob
+        )
+        self.dropout = nn.Dropout(classifier_dropout)
+        self.classifier = nn.Linear(config.hidden_size, 1)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @add_start_docstrings_to_model_forward(
+        ROC_BERT_INPUTS_DOCSTRING.format("batch_size, num_choices, sequence_length")
+    )
+    @add_code_sample_docstrings(
+        processor_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=MultipleChoiceModelOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        input_shape_ids: Optional[torch.Tensor] = None,
+        input_pronunciation_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        labels: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple[torch.Tensor], MultipleChoiceModelOutput]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the multiple choice classification loss. Indices should be in `[0, ...,
+            num_choices-1]` where `num_choices` is the size of the second dimension of the input tensors. (See
+            `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
+        input_shape_ids = input_shape_ids.view(-1, input_shape_ids.size(-1)) if input_shape_ids is not None else None
+        input_pronunciation_ids = (
+            input_pronunciation_ids.view(-1, input_pronunciation_ids.size(-1))
+            if input_pronunciation_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.roc_bert(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_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(
+    """RoCBert 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.""",
+    ROC_BERT_START_DOCSTRING,
+)
+class RoCBertForTokenClassification(RoCBertPreTrainedModel):
+    _keys_to_ignore_on_load_unexpected = [r"pooler"]
+
+    # Copied from transformers.models.bert.modeling_bert.BertForTokenClassification.__init__ with Bert->RoCBert,bert->roc_bert
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+
+        self.roc_bert = RoCBertModel(config, add_pooling_layer=False)
+        classifier_dropout = (
+            config.classifier_dropout if config.classifier_dropout is not None else config.hidden_dropout_prob
+        )
+        self.dropout = nn.Dropout(classifier_dropout)
+        self.classifier = nn.Linear(config.hidden_size, config.num_labels)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @add_start_docstrings_to_model_forward(ROC_BERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_code_sample_docstrings(
+        processor_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=TokenClassifierOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        input_shape_ids: Optional[torch.Tensor] = None,
+        input_pronunciation_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        labels: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ):
+        r"""
+        labels (`torch.LongTensor` of shape `(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.roc_bert(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_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()
+            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(
+    """RoCBert 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`).""",
+    ROC_BERT_START_DOCSTRING,
+)
+class RoCBertForQuestionAnswering(RoCBertPreTrainedModel):
+    _keys_to_ignore_on_load_unexpected = [r"pooler"]
+
+    # Copied from transformers.models.bert.modeling_bert.BertForQuestionAnswering.__init__ with Bert->RoCBert,bert->roc_bert
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+
+        self.roc_bert = RoCBertModel(config, add_pooling_layer=False)
+        self.qa_outputs = nn.Linear(config.hidden_size, config.num_labels)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @add_start_docstrings_to_model_forward(ROC_BERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_code_sample_docstrings(
+        processor_class=_TOKENIZER_FOR_DOC,
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=QuestionAnsweringModelOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        input_shape_ids: Optional[torch.Tensor] = None,
+        input_pronunciation_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        start_positions: Optional[torch.Tensor] = None,
+        end_positions: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple[torch.Tensor], QuestionAnsweringModelOutput]:
+        r"""
+        start_positions (`torch.LongTensor` of shape `(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 (`sequence_length`). Position outside of the sequence
+            are not taken into account for computing the loss.
+        end_positions (`torch.LongTensor` of shape `(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 (`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.roc_bert(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_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 = start_positions.clamp(0, ignored_index)
+            end_positions = 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/roc_bert/tokenization_roc_bert.py

+# coding=utf-8
+# Copyright 2022 WeChatAI 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 RoCBert."""
+
+import collections
+import itertools
+import json
+import os
+import unicodedata
+from typing import Dict, List, Optional, Tuple, Union
+
+from ...tokenization_utils import PreTrainedTokenizer, _is_control, _is_punctuation, _is_whitespace
+from ...tokenization_utils_base import (
+    ENCODE_KWARGS_DOCSTRING,
+    ENCODE_PLUS_ADDITIONAL_KWARGS_DOCSTRING,
+    BatchEncoding,
+    EncodedInput,
+    EncodedInputPair,
+    PaddingStrategy,
+    PreTokenizedInput,
+    PreTokenizedInputPair,
+    TensorType,
+    TextInput,
+    TextInputPair,
+    TruncationStrategy,
+)
+from ...utils import add_end_docstrings, logging
+
+
+logger = logging.get_logger(__name__)
+
+VOCAB_FILES_NAMES = {
+    "vocab_file": "vocab.txt",
+    "word_shape_file": "word_shape.json",
+    "word_pronunciation_file": "word_pronunciation.json",
+}
+
+PRETRAINED_VOCAB_FILES_MAP = {
+    "vocab_file": {
+        "weiweishi/roc-bert-base-zh": "https://huggingface.co/weiweishi/roc-bert-base-zh/resolve/main/vocab.txt"
+    },
+    "word_shape_file": {
+        "weiweishi/roc-bert-base-zh": "https://huggingface.co/weiweishi/roc-bert-base-zh/resolve/main/word_shape.json"
+    },
+    "word_pronunciation_file": {
+        "weiweishi/roc-bert-base-zh": (
+            "https://huggingface.co/weiweishi/roc-bert-base-zh/resolve/main/word_pronunciation.json"
+        )
+    },
+}
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    "weiweishi/roc-bert-base-zh": 512,
+}
+
+PRETRAINED_INIT_CONFIGURATION = {
+    "weiweishi/roc-bert-base-zh": {"do_lower_case": True},
+}
+
+
+# Copied from transformers.models.bert.tokenization_bert.load_vocab
+def load_vocab(vocab_file):
+    """Loads a vocabulary file into a dictionary."""
+    vocab = collections.OrderedDict()
+    with open(vocab_file, "r", encoding="utf-8") as reader:
+        tokens = reader.readlines()
+    for index, token in enumerate(tokens):
+        token = token.rstrip("\n")
+        vocab[token] = index
+    return vocab
+
+
+# Copied from transformers.models.bert.tokenization_bert.whitespace_tokenize
+def whitespace_tokenize(text):
+    """Runs basic whitespace cleaning and splitting on a piece of text."""
+    text = text.strip()
+    if not text:
+        return []
+    tokens = text.split()
+    return tokens
+
+
+class RoCBertTokenizer(PreTrainedTokenizer):
+    r"""
+    Args:
+    Construct a RoCBert tokenizer. Based on WordPiece. This tokenizer inherits from [`PreTrainedTokenizer`] which
+    contains most of the main methods. Users should refer to this superclass for more information regarding those
+    methods.
+        vocab_file (`str`):
+            File containing the vocabulary.
+        word_shape_file (`str`):
+            File containing the word => shape info.
+        word_pronunciation_file (`str`):
+            File containing the word => pronunciation info.
+        do_lower_case (`bool`, *optional*, defaults to `True`):
+            Whether or not to lowercase the input when tokenizing.
+        do_basic_tokenize (`bool`, *optional*, defaults to `True`):
+            Whether or not to do basic tokenization before WordPiece.
+        never_split (`Iterable`, *optional*):
+            Collection of tokens which will never be split during tokenization. Only has an effect when
+            `do_basic_tokenize=True`
+        unk_token (`str`, *optional*, defaults to `"[UNK]"`):
+            The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this
+            token instead.
+        sep_token (`str`, *optional*, defaults to `"[SEP]"`):
+            The separator token, which is used when building a sequence from multiple sequences, e.g. two sequences for
+            sequence classification or for a text and a question for question answering. It is also used as the last
+            token of a sequence built with special tokens.
+        pad_token (`str`, *optional*, defaults to `"[PAD]"`):
+            The token used for padding, for example when batching sequences of different lengths.
+        cls_token (`str`, *optional*, defaults to `"[CLS]"`):
+            The classifier token which is used when doing sequence classification (classification of the whole sequence
+            instead of per-token classification). It is the first token of the sequence when built with special tokens.
+        mask_token (`str`, *optional*, defaults to `"[MASK]"`):
+            The token used for masking values. This is the token used when training this model with masked language
+            modeling. This is the token which the model will try to predict.
+        tokenize_chinese_chars (`bool`, *optional*, defaults to `True`):
+            Whether or not to tokenize Chinese characters. This should likely be deactivated for Japanese (see this
+            [issue](https://github.com/huggingface/transformers/issues/328)).
+        strip_accents (`bool`, *optional*):
+            Whether or not to strip all accents. If this option is not specified, then it will be determined by the
+            value for `lowercase` (as in the original BERT).
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    pretrained_init_configuration = PRETRAINED_INIT_CONFIGURATION
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+
+    def __init__(
+        self,
+        vocab_file,
+        word_shape_file,
+        word_pronunciation_file,
+        do_lower_case=True,
+        do_basic_tokenize=True,
+        never_split=None,
+        unk_token="[UNK]",
+        sep_token="[SEP]",
+        pad_token="[PAD]",
+        cls_token="[CLS]",
+        mask_token="[MASK]",
+        tokenize_chinese_chars=True,
+        strip_accents=None,
+        **kwargs
+    ):
+        super().__init__(
+            do_lower_case=do_lower_case,
+            do_basic_tokenize=do_basic_tokenize,
+            never_split=never_split,
+            unk_token=unk_token,
+            sep_token=sep_token,
+            pad_token=pad_token,
+            cls_token=cls_token,
+            mask_token=mask_token,
+            tokenize_chinese_chars=tokenize_chinese_chars,
+            strip_accents=strip_accents,
+            **kwargs,
+        )
+
+        for cur_file in [vocab_file, word_shape_file, word_pronunciation_file]:
+            if cur_file is None or not os.path.isfile(cur_file):
+                raise ValueError(
+                    f"Can't find a vocabulary file at path '{vocab_file}'. To load the vocabulary from a Google "
+                    "pretrained model use `tokenizer = RoCBertTokenizer.from_pretrained(PRETRAINED_MODEL_NAME)`"
+                )
+
+        self.vocab = load_vocab(vocab_file)
+
+        with open(word_shape_file, "r", encoding="utf8") as in_file:
+            self.word_shape = json.load(in_file)
+
+        with open(word_pronunciation_file, "r", encoding="utf8") as in_file:
+            self.word_pronunciation = json.load(in_file)
+
+        self.ids_to_tokens = collections.OrderedDict([(ids, tok) for tok, ids in self.vocab.items()])
+
+        self.do_basic_tokenize = do_basic_tokenize
+        if do_basic_tokenize:
+            self.basic_tokenizer = RoCBertBasicTokenizer(
+                do_lower_case=do_lower_case,
+                never_split=never_split,
+                tokenize_chinese_chars=tokenize_chinese_chars,
+                strip_accents=strip_accents,
+            )
+        self.wordpiece_tokenizer = RoCBertWordpieceTokenizer(vocab=self.vocab, unk_token=self.unk_token)
+
+    @property
+    def do_lower_case(self):
+        return self.basic_tokenizer.do_lower_case
+
+    @property
+    def vocab_size(self):
+        return len(self.vocab)
+
+    # Copied from transformers.models.bert.tokenization_bert.BertTokenizer.get_vocab
+    def get_vocab(self):
+        return dict(self.vocab, **self.added_tokens_encoder)
+
+    # Copied from transformers.models.bert.tokenization_bert.BertTokenizer._tokenize
+    def _tokenize(self, text):
+        split_tokens = []
+        if self.do_basic_tokenize:
+            for token in self.basic_tokenizer.tokenize(text, never_split=self.all_special_tokens):
+
+                # If the token is part of the never_split set
+                if token in self.basic_tokenizer.never_split:
+                    split_tokens.append(token)
+                else:
+                    split_tokens += self.wordpiece_tokenizer.tokenize(token)
+        else:
+            split_tokens = self.wordpiece_tokenizer.tokenize(text)
+        return split_tokens
+
+    def _encode_plus(
+        self,
+        text: Union[TextInput, PreTokenizedInput, EncodedInput],
+        text_pair: Optional[Union[TextInput, PreTokenizedInput, EncodedInput]] = None,
+        add_special_tokens: bool = True,
+        padding_strategy: PaddingStrategy = PaddingStrategy.DO_NOT_PAD,
+        truncation_strategy: TruncationStrategy = TruncationStrategy.DO_NOT_TRUNCATE,
+        max_length: Optional[int] = None,
+        stride: int = 0,
+        is_split_into_words: bool = False,
+        pad_to_multiple_of: Optional[int] = None,
+        return_tensors: Optional[Union[str, TensorType]] = None,
+        return_token_type_ids: Optional[bool] = None,
+        return_attention_mask: Optional[bool] = None,
+        return_overflowing_tokens: bool = False,
+        return_special_tokens_mask: bool = False,
+        return_offsets_mapping: bool = False,
+        return_length: bool = False,
+        verbose: bool = True,
+        **kwargs
+    ) -> BatchEncoding:
+        def get_input_ids(text):
+            if isinstance(text, str):
+                tokens = self.tokenize(text, **kwargs)
+                tokens_ids = self.convert_tokens_to_ids(tokens)
+                tokens_shape_ids = self.convert_tokens_to_shape_ids(tokens)
+                tokens_proun_ids = self.convert_tokens_to_pronunciation_ids(tokens)
+                return tokens_ids, tokens_shape_ids, tokens_proun_ids
+            elif isinstance(text, (list, tuple)) and len(text) > 0 and isinstance(text[0], str):
+                if is_split_into_words:
+                    tokens = list(
+                        itertools.chain(*(self.tokenize(t, is_split_into_words=True, **kwargs) for t in text))
+                    )
+                    tokens_ids = self.convert_tokens_to_ids(tokens)
+                    tokens_shape_ids = self.convert_tokens_to_shape_ids(tokens)
+                    tokens_proun_ids = self.convert_tokens_to_pronunciation_ids(tokens)
+                    return tokens_ids, tokens_shape_ids, tokens_proun_ids
+                else:
+                    tokens_ids = self.convert_tokens_to_ids(text)
+                    tokens_shape_ids = self.convert_tokens_to_shape_ids(text)
+                    tokens_proun_ids = self.convert_tokens_to_pronunciation_ids(text)
+                    return tokens_ids, tokens_shape_ids, tokens_proun_ids
+            elif isinstance(text, (list, tuple)) and len(text) > 0 and isinstance(text[0], int):
+                return text, [0] * len(text), [0] * len(text)  # shape and proun id is pad_value
+            else:
+                if is_split_into_words:
+                    raise ValueError(
+                        f"Input {text} is not valid. Should be a string or a list/tuple of strings when"
+                        " `is_split_into_words=True`."
+                    )
+                else:
+                    raise ValueError(
+                        f"Input {text} is not valid. Should be a string, a list/tuple of strings or a list/tuple of"
+                        " integers."
+                    )
+
+        if return_offsets_mapping:
+            raise NotImplementedError(
+                "return_offset_mapping is not available when using Python tokenizers. "
+                "To use this feature, change your tokenizer to one deriving from "
+                "transformers.PreTrainedTokenizerFast. "
+                "More information on available tokenizers at "
+                "https://github.com/huggingface/transformers/pull/2674"
+            )
+
+        first_ids, first_shape_ids, first_proun_ids = get_input_ids(text)
+        if text_pair is not None:
+            second_ids, second_shape_ids, second_proun_ids = get_input_ids(text_pair)
+        else:
+            second_ids, second_shape_ids, second_proun_ids = None, None, None
+
+        return self.prepare_for_model(
+            first_ids,
+            first_shape_ids,
+            first_proun_ids,
+            pair_ids=second_ids,
+            pair_shape_ids=second_shape_ids,
+            pair_pronunciation_ids=second_proun_ids,
+            add_special_tokens=add_special_tokens,
+            padding=padding_strategy.value,
+            truncation=truncation_strategy.value,
+            max_length=max_length,
+            stride=stride,
+            pad_to_multiple_of=pad_to_multiple_of,
+            return_tensors=return_tensors,
+            prepend_batch_axis=True,
+            return_attention_mask=return_attention_mask,
+            return_token_type_ids=return_token_type_ids,
+            return_overflowing_tokens=return_overflowing_tokens,
+            return_special_tokens_mask=return_special_tokens_mask,
+            return_length=return_length,
+            verbose=verbose,
+        )
+
+    @add_end_docstrings(ENCODE_KWARGS_DOCSTRING, ENCODE_PLUS_ADDITIONAL_KWARGS_DOCSTRING)
+    def prepare_for_model(
+        self,
+        ids: List[int],
+        shape_ids: List[int],
+        pronunciation_ids: List[int],
+        pair_ids: Optional[List[int]] = None,
+        pair_shape_ids: Optional[List[int]] = None,
+        pair_pronunciation_ids: Optional[List[int]] = None,
+        add_special_tokens: bool = True,
+        padding: Union[bool, str, PaddingStrategy] = False,
+        truncation: Union[bool, str, TruncationStrategy] = None,
+        max_length: Optional[int] = None,
+        stride: int = 0,
+        pad_to_multiple_of: Optional[int] = None,
+        return_tensors: Optional[Union[str, TensorType]] = None,
+        return_token_type_ids: Optional[bool] = None,
+        return_attention_mask: Optional[bool] = None,
+        return_overflowing_tokens: bool = False,
+        return_special_tokens_mask: bool = False,
+        return_offsets_mapping: bool = False,
+        return_length: bool = False,
+        verbose: bool = True,
+        prepend_batch_axis: bool = False,
+        **kwargs
+    ) -> BatchEncoding:
+        """
+        Prepares a sequence of input id, or a pair of sequences of inputs ids so that it can be used by the model. It
+        adds special tokens, truncates sequences if overflowing while taking into account the special tokens and
+        manages a moving window (with user defined stride) for overflowing tokens. Please Note, for *pair_ids*
+        different than `None` and *truncation_strategy = longest_first* or `True`, it is not possible to return
+        overflowing tokens. Such a combination of arguments will raise an error.
+
+        Args:
+            ids (`List[int]`):
+                Tokenized input ids of the first sequence. Can be obtained from a string by chaining the `tokenize` and
+                `convert_tokens_to_id` methods.
+            shape_ids (`List[int]`):
+                Tokenized input ids of the first sequence. Can be obtained from a string by chaining the `tokenize` and
+                `convert_token_to_shape_id` methods.
+            pronunciation_ids (`List[int]`):
+                Tokenized input ids of the first sequence. Can be obtained from a string by chaining the `tokenize` and
+                `convert_token_to_pronunciation_id` methods.
+            pair_ids (`List[int]`, *optional*):
+                Tokenized input ids of the second sequence. Can be obtained from a string by chaining the `tokenize`
+                and `convert_tokens_to_id` methods.
+            pair_shape_ids (`List[int]`, *optional*):
+                Tokenized input ids of the second sequence. Can be obtained from a string by chaining the `tokenize`
+                and `convert_token_to_shape_id` methods.
+            pair_pronunciation_ids (`List[int]`, *optional*):
+                Tokenized input ids of the second sequence. Can be obtained from a string by chaining the `tokenize`
+                and `convert_token_to_pronunciation_id` methods.
+        """
+
+        # Backward compatibility for 'truncation_strategy', 'pad_to_max_length'
+        padding_strategy, truncation_strategy, max_length, kwargs = self._get_padding_truncation_strategies(
+            padding=padding,
+            truncation=truncation,
+            max_length=max_length,
+            pad_to_multiple_of=pad_to_multiple_of,
+            verbose=verbose,
+            **kwargs,
+        )
+
+        pair = bool(pair_ids is not None)
+        len_ids = len(ids)
+        len_pair_ids = len(pair_ids) if pair else 0
+
+        if return_token_type_ids and not add_special_tokens:
+            raise ValueError(
+                "Asking to return token_type_ids while setting add_special_tokens to False "
+                "results in an undefined behavior. Please set add_special_tokens to True or "
+                "set return_token_type_ids to None."
+            )
+
+        if (
+            return_overflowing_tokens
+            and truncation_strategy == TruncationStrategy.LONGEST_FIRST
+            and pair_ids is not None
+        ):
+            raise ValueError(
+                "Not possible to return overflowing tokens for pair of sequences with the "
+                "`longest_first`. Please select another truncation strategy than `longest_first`, "
+                "for instance `only_second` or `only_first`."
+            )
+
+        # Load from model defaults
+        if return_token_type_ids is None:
+            return_token_type_ids = "token_type_ids" in self.model_input_names
+        if return_attention_mask is None:
+            return_attention_mask = "attention_mask" in self.model_input_names
+
+        encoded_inputs = {}
+
+        # Compute the total size of the returned encodings
+        total_len = len_ids + len_pair_ids + (self.num_special_tokens_to_add(pair=pair) if add_special_tokens else 0)
+
+        # Truncation: Handle max sequence length
+        overflowing_tokens = []
+        if truncation_strategy != TruncationStrategy.DO_NOT_TRUNCATE and max_length and total_len > max_length:
+            ids, pair_ids, overflowing_tokens = self.truncate_sequences(
+                ids,
+                pair_ids=pair_ids,
+                num_tokens_to_remove=total_len - max_length,
+                truncation_strategy=truncation_strategy,
+                stride=stride,
+            )
+            shape_ids, pair_shape_ids, _ = self.truncate_sequences(
+                shape_ids,
+                pair_ids=pair_shape_ids,
+                num_tokens_to_remove=total_len - max_length,
+                truncation_strategy=truncation_strategy,
+                stride=stride,
+            )
+            pronunciation_ids, pair_pronunciation_ids, _ = self.truncate_sequences(
+                pronunciation_ids,
+                pair_ids=pair_pronunciation_ids,
+                num_tokens_to_remove=total_len - max_length,
+                truncation_strategy=truncation_strategy,
+                stride=stride,
+            )
+
+        if return_overflowing_tokens:
+            encoded_inputs["overflowing_tokens"] = overflowing_tokens
+            encoded_inputs["num_truncated_tokens"] = total_len - max_length
+
+        # Add special tokens
+        if add_special_tokens:
+            sequence = self.build_inputs_with_special_tokens(ids, pair_ids)
+            token_type_ids = self.create_token_type_ids_from_sequences(ids, pair_ids)
+            input_shape_ids = self.build_inputs_with_special_tokens(
+                shape_ids, pair_shape_ids, self.word_shape["[UNK]"], self.word_shape["[UNK]"]
+            )
+            input_pronunciation_ids = self.build_inputs_with_special_tokens(
+                pronunciation_ids,
+                pair_pronunciation_ids,
+                self.word_pronunciation["[UNK]"],
+                self.word_pronunciation["[UNK]"],
+            )
+        else:
+            sequence = ids + pair_ids if pair_ids else ids
+            token_type_ids = [0] * len(ids) + ([0] * len(pair_ids) if pair_ids else [])
+            input_shape_ids = shape_ids + pair_shape_ids if pair_shape_ids else shape_ids
+            input_pronunciation_ids = (
+                pronunciation_ids + pair_pronunciation_ids if pair_pronunciation_ids else pronunciation_ids
+            )
+
+        # Build output dictionary
+        encoded_inputs["input_ids"] = sequence
+        encoded_inputs["input_shape_ids"] = input_shape_ids
+        encoded_inputs["input_pronunciation_ids"] = input_pronunciation_ids
+        if return_token_type_ids:
+            encoded_inputs["token_type_ids"] = token_type_ids
+        if return_special_tokens_mask:
+            if add_special_tokens:
+                encoded_inputs["special_tokens_mask"] = self.get_special_tokens_mask(ids, pair_ids)
+            else:
+                encoded_inputs["special_tokens_mask"] = [0] * len(sequence)
+
+        # Check lengths
+        self._eventual_warn_about_too_long_sequence(encoded_inputs["input_ids"], max_length, verbose)
+
+        # Padding
+        if padding_strategy != PaddingStrategy.DO_NOT_PAD or return_attention_mask:
+            encoded_inputs = self.pad(
+                encoded_inputs,
+                max_length=max_length,
+                padding=padding_strategy.value,
+                pad_to_multiple_of=pad_to_multiple_of,
+                return_attention_mask=return_attention_mask,
+            )
+
+        if return_length:
+            encoded_inputs["length"] = len(encoded_inputs["input_ids"])
+
+        batch_outputs = BatchEncoding(
+            encoded_inputs, tensor_type=return_tensors, prepend_batch_axis=prepend_batch_axis
+        )
+
+        return batch_outputs
+
+    def _pad(
+        self,
+        encoded_inputs: Union[Dict[str, EncodedInput], BatchEncoding],
+        max_length: Optional[int] = None,
+        padding_strategy: PaddingStrategy = PaddingStrategy.DO_NOT_PAD,
+        pad_to_multiple_of: Optional[int] = None,
+        return_attention_mask: Optional[bool] = None,
+    ) -> dict:
+        # Load from model defaults
+        if return_attention_mask is None:
+            return_attention_mask = "attention_mask" in self.model_input_names
+
+        required_input = encoded_inputs[self.model_input_names[0]]
+
+        if padding_strategy == PaddingStrategy.LONGEST:
+            max_length = len(required_input)
+
+        if max_length is not None and pad_to_multiple_of is not None and (max_length % pad_to_multiple_of != 0):
+            max_length = ((max_length // pad_to_multiple_of) + 1) * pad_to_multiple_of
+
+        needs_to_be_padded = padding_strategy != PaddingStrategy.DO_NOT_PAD and len(required_input) != max_length
+
+        # Initialize attention mask if not present.
+        if return_attention_mask and "attention_mask" not in encoded_inputs:
+            encoded_inputs["attention_mask"] = [1] * len(required_input)
+
+        if needs_to_be_padded:
+            difference = max_length - len(required_input)
+
+            if self.padding_side == "right":
+                if return_attention_mask:
+                    encoded_inputs["attention_mask"] = encoded_inputs["attention_mask"] + [0] * difference
+                if "token_type_ids" in encoded_inputs:
+                    encoded_inputs["token_type_ids"] = (
+                        encoded_inputs["token_type_ids"] + [self.pad_token_type_id] * difference
+                    )
+                if "special_tokens_mask" in encoded_inputs:
+                    encoded_inputs["special_tokens_mask"] = encoded_inputs["special_tokens_mask"] + [1] * difference
+                for key in ["input_shape_ids", "input_pronunciation_ids"]:
+                    if key in encoded_inputs:
+                        encoded_inputs[key] = encoded_inputs[key] + [self.pad_token_id] * difference
+                encoded_inputs[self.model_input_names[0]] = required_input + [self.pad_token_id] * difference
+            elif self.padding_side == "left":
+                if return_attention_mask:
+                    encoded_inputs["attention_mask"] = [0] * difference + encoded_inputs["attention_mask"]
+                if "token_type_ids" in encoded_inputs:
+                    encoded_inputs["token_type_ids"] = [self.pad_token_type_id] * difference + encoded_inputs[
+                        "token_type_ids"
+                    ]
+                if "special_tokens_mask" in encoded_inputs:
+                    encoded_inputs["special_tokens_mask"] = [1] * difference + encoded_inputs["special_tokens_mask"]
+                for key in ["input_shape_ids", "input_pronunciation_ids"]:
+                    if key in encoded_inputs:
+                        encoded_inputs[key] = [self.pad_token_id] * difference + encoded_inputs[key]
+                encoded_inputs[self.model_input_names[0]] = [self.pad_token_id] * difference + required_input
+            else:
+                raise ValueError("Invalid padding strategy:" + str(self.padding_side))
+
+        return encoded_inputs
+
+    def _batch_encode_plus(
+        self,
+        batch_text_or_text_pairs: Union[
+            List[TextInput],
+            List[TextInputPair],
+            List[PreTokenizedInput],
+            List[PreTokenizedInputPair],
+            List[EncodedInput],
+            List[EncodedInputPair],
+        ],
+        add_special_tokens: bool = True,
+        padding_strategy: PaddingStrategy = PaddingStrategy.DO_NOT_PAD,
+        truncation_strategy: TruncationStrategy = TruncationStrategy.DO_NOT_TRUNCATE,
+        max_length: Optional[int] = None,
+        stride: int = 0,
+        is_split_into_words: bool = False,
+        pad_to_multiple_of: Optional[int] = None,
+        return_tensors: Optional[Union[str, TensorType]] = None,
+        return_token_type_ids: Optional[bool] = None,
+        return_attention_mask: Optional[bool] = None,
+        return_overflowing_tokens: bool = False,
+        return_special_tokens_mask: bool = False,
+        return_offsets_mapping: bool = False,
+        return_length: bool = False,
+        verbose: bool = True,
+        **kwargs
+    ) -> BatchEncoding:
+        def get_input_ids(text):
+            if isinstance(text, str):
+                tokens = self.tokenize(text, **kwargs)
+                tokens_ids = self.convert_tokens_to_ids(tokens)
+                tokens_shape_ids = self.convert_tokens_to_shape_ids(tokens)
+                tokens_proun_ids = self.convert_tokens_to_pronunciation_ids(tokens)
+                return tokens_ids, tokens_shape_ids, tokens_proun_ids
+            elif isinstance(text, (list, tuple)) and len(text) > 0 and isinstance(text[0], str):
+                if is_split_into_words:
+                    tokens = list(
+                        itertools.chain(*(self.tokenize(t, is_split_into_words=True, **kwargs) for t in text))
+                    )
+                    tokens_ids = self.convert_tokens_to_ids(tokens)
+                    tokens_shape_ids = self.convert_tokens_to_shape_ids(tokens)
+                    tokens_proun_ids = self.convert_tokens_to_pronunciation_ids(tokens)
+                    return tokens_ids, tokens_shape_ids, tokens_proun_ids
+                else:
+                    tokens_ids = self.convert_tokens_to_ids(text)
+                    tokens_shape_ids = self.convert_tokens_to_shape_ids(text)
+                    tokens_proun_ids = self.convert_tokens_to_pronunciation_ids(text)
+                    return tokens_ids, tokens_shape_ids, tokens_proun_ids
+            elif isinstance(text, (list, tuple)) and len(text) > 0 and isinstance(text[0], int):
+                return text, [0] * len(text), [0] * len(text)  # shape and proun id is pad_value
+            else:
+                raise ValueError(
+                    "Input is not valid. Should be a string, a list/tuple of strings or a list/tuple of integers."
+                )
+
+        if return_offsets_mapping:
+            raise NotImplementedError(
+                "return_offset_mapping is not available when using Python tokenizers. "
+                "To use this feature, change your tokenizer to one deriving from "
+                "transformers.PreTrainedTokenizerFast."
+            )
+
+        input_ids = []
+        input_shape_ids = []
+        input_pronunciation_ids = []
+        for ids_or_pair_ids in batch_text_or_text_pairs:
+            if not isinstance(ids_or_pair_ids, (list, tuple)):
+                ids, pair_ids = ids_or_pair_ids, None
+            elif is_split_into_words and not isinstance(ids_or_pair_ids[0], (list, tuple)):
+                ids, pair_ids = ids_or_pair_ids, None
+            else:
+                ids, pair_ids = ids_or_pair_ids
+
+            first_ids, first_shape_ids, first_proun_ids = get_input_ids(ids)
+            if pair_ids is not None:
+                second_ids, second_shape_ids, second_proun_ids = get_input_ids(pair_ids)
+            else:
+                second_ids, second_shape_ids, second_proun_ids = None, None, None
+
+            input_ids.append((first_ids, second_ids))
+            input_shape_ids.append((first_shape_ids, second_shape_ids))
+            input_pronunciation_ids.append((first_proun_ids, second_proun_ids))
+
+        batch_outputs = self._batch_prepare_for_model(
+            input_ids,
+            batch_shape_ids_pairs=input_shape_ids,
+            batch_pronunciation_ids_pairs=input_pronunciation_ids,
+            add_special_tokens=add_special_tokens,
+            padding_strategy=padding_strategy,
+            truncation_strategy=truncation_strategy,
+            max_length=max_length,
+            stride=stride,
+            pad_to_multiple_of=pad_to_multiple_of,
+            return_attention_mask=return_attention_mask,
+            return_token_type_ids=return_token_type_ids,
+            return_overflowing_tokens=return_overflowing_tokens,
+            return_special_tokens_mask=return_special_tokens_mask,
+            return_length=return_length,
+            return_tensors=return_tensors,
+            verbose=verbose,
+        )
+
+        return BatchEncoding(batch_outputs)
+
+    @add_end_docstrings(ENCODE_KWARGS_DOCSTRING, ENCODE_PLUS_ADDITIONAL_KWARGS_DOCSTRING)
+    def _batch_prepare_for_model(
+        self,
+        batch_ids_pairs: List[Union[PreTokenizedInputPair, Tuple[List[int], None]]],
+        batch_shape_ids_pairs: List[Union[PreTokenizedInputPair, Tuple[List[int], None]]],
+        batch_pronunciation_ids_pairs: List[Union[PreTokenizedInputPair, Tuple[List[int], None]]],
+        add_special_tokens: bool = True,
+        padding_strategy: PaddingStrategy = PaddingStrategy.DO_NOT_PAD,
+        truncation_strategy: TruncationStrategy = TruncationStrategy.DO_NOT_TRUNCATE,
+        max_length: Optional[int] = None,
+        stride: int = 0,
+        pad_to_multiple_of: Optional[int] = None,
+        return_tensors: Optional[str] = None,
+        return_token_type_ids: Optional[bool] = None,
+        return_attention_mask: Optional[bool] = None,
+        return_overflowing_tokens: bool = False,
+        return_special_tokens_mask: bool = False,
+        return_length: bool = False,
+        verbose: bool = True,
+    ) -> BatchEncoding:
+        """
+        Prepares a sequence of input id, or a pair of sequences of inputs ids so that it can be used by the model. It
+        adds special tokens, truncates sequences if overflowing while taking into account the special tokens and
+        manages a moving window (with user defined stride) for overflowing tokens
+
+        Args:
+            batch_ids_pairs: list of tokenized input ids or input ids pairs
+            batch_shape_ids_pairs: list of tokenized input shape ids or input shape ids pairs
+            batch_pronunciation_ids_pairs: list of tokenized input pronunciation ids or input pronunciation ids pairs
+        """
+
+        batch_outputs = {}
+        for i, (first_ids, second_ids) in enumerate(batch_ids_pairs):
+            first_shape_ids, second_shape_ids = batch_shape_ids_pairs[i]
+            first_pronunciation_ids, second_pronunciation_ids = batch_pronunciation_ids_pairs[i]
+            outputs = self.prepare_for_model(
+                first_ids,
+                first_shape_ids,
+                first_pronunciation_ids,
+                pair_ids=second_ids,
+                pair_shape_ids=second_shape_ids,
+                pair_pronunciation_ids=second_pronunciation_ids,
+                add_special_tokens=add_special_tokens,
+                padding=PaddingStrategy.DO_NOT_PAD.value,  # we pad in batch afterward
+                truncation=truncation_strategy.value,
+                max_length=max_length,
+                stride=stride,
+                pad_to_multiple_of=None,  # we pad in batch afterward
+                return_attention_mask=False,  # we pad in batch afterward
+                return_token_type_ids=return_token_type_ids,
+                return_overflowing_tokens=return_overflowing_tokens,
+                return_special_tokens_mask=return_special_tokens_mask,
+                return_length=return_length,
+                return_tensors=None,  # We convert the whole batch to tensors at the end
+                prepend_batch_axis=False,
+                verbose=verbose,
+            )
+
+            for key, value in outputs.items():
+                if key not in batch_outputs:
+                    batch_outputs[key] = []
+                batch_outputs[key].append(value)
+
+        batch_outputs = self.pad(
+            batch_outputs,
+            padding=padding_strategy.value,
+            max_length=max_length,
+            pad_to_multiple_of=pad_to_multiple_of,
+            return_attention_mask=return_attention_mask,
+        )
+
+        batch_outputs = BatchEncoding(batch_outputs, tensor_type=return_tensors)
+
+        return batch_outputs
+
+    # Copied from transformers.models.bert.tokenization_bert.BertTokenizer._convert_token_to_id
+    def _convert_token_to_id(self, token):
+        """Converts a token (str) in an id using the vocab."""
+        return self.vocab.get(token, self.vocab.get(self.unk_token))
+
+    def _convert_token_to_shape_id(self, token):
+        """Converts a token (str) in an shape_id using the shape vocab."""
+        return self.word_shape.get(token, self.word_shape.get(self.unk_token))
+
+    def convert_tokens_to_shape_ids(self, tokens: Union[str, List[str]]) -> Union[int, List[int]]:
+        if tokens is None:
+            return None
+
+        ids = []
+        for token in tokens:
+            ids.append(self._convert_token_to_shape_id(token))
+        return ids
+
+    def _convert_token_to_pronunciation_id(self, token):
+        """Converts a token (str) in an shape_id using the shape vocab."""
+        return self.word_pronunciation.get(token, self.word_pronunciation.get(self.unk_token))
+
+    def convert_tokens_to_pronunciation_ids(self, tokens: Union[str, List[str]]) -> Union[int, List[int]]:
+        if tokens is None:
+            return None
+
+        ids = []
+        for token in tokens:
+            ids.append(self._convert_token_to_pronunciation_id(token))
+        return ids
+
+    # Copied from transformers.models.bert.tokenization_bert.BertTokenizer._convert_id_to_token
+    def _convert_id_to_token(self, index):
+        """Converts an index (integer) in a token (str) using the vocab."""
+        return self.ids_to_tokens.get(index, self.unk_token)
+
+    # Copied from transformers.models.bert.tokenization_bert.BertTokenizer.convert_tokens_to_string
+    def convert_tokens_to_string(self, tokens):
+        """Converts a sequence of tokens (string) in a single string."""
+        out_string = " ".join(tokens).replace(" ##", "").strip()
+        return out_string
+
+    def build_inputs_with_special_tokens(
+        self,
+        token_ids_0: List[int],
+        token_ids_1: Optional[List[int]] = None,
+        cls_token_id: int = None,
+        sep_token_id: int = None,
+    ) -> List[int]:
+        """
+        Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
+        adding special tokens. A BERT sequence has the following format:
+
+        - single sequence: `[CLS] X [SEP]`
+        - pair of sequences: `[CLS] A [SEP] B [SEP]`
+
+        Args:
+            token_ids_0 (`List[int]`):
+                List of IDs to which the special tokens will be added.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            `List[int]`: List of [input IDs](../glossary#input-ids) with the appropriate special tokens.
+        """
+        cls = [self.cls_token_id] if cls_token_id is None else [cls_token_id]
+        sep = [self.sep_token_id] if sep_token_id is None else [sep_token_id]
+        if token_ids_1 is None:
+            return cls + token_ids_0 + sep
+        return cls + token_ids_0 + sep + token_ids_1 + sep
+
+    # Copied from transformers.models.bert.tokenization_bert.BertTokenizer.get_special_tokens_mask
+    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 (`List[int]`):
+                List of IDs.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+            already_has_special_tokens (`bool`, *optional*, defaults to `False`):
+                Whether or not the token list is already formatted with special tokens for the model.
+
+        Returns:
+            `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
+            )
+
+        if token_ids_1 is not None:
+            return [1] + ([0] * len(token_ids_0)) + [1] + ([0] * len(token_ids_1)) + [1]
+        return [1] + ([0] * len(token_ids_0)) + [1]
+
+    # Copied from transformers.models.bert.tokenization_bert.BertTokenizer.create_token_type_ids_from_sequences
+    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 BERT 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 `token_ids_1` is `None`, this method only returns the first portion of the mask (0s).
+
+        Args:
+            token_ids_0 (`List[int]`):
+                List of IDs.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            `List[int]`: List of [token type IDs](../glossary#token-type-ids) according to the given sequence(s).
+        """
+        sep = [self.sep_token_id]
+        cls = [self.cls_token_id]
+        if token_ids_1 is None:
+            return len(cls + token_ids_0 + sep) * [0]
+        return len(cls + token_ids_0 + sep) * [0] + len(token_ids_1 + sep) * [1]
+
+    def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> Tuple[str, str, str]:
+        index = 0
+        if os.path.isdir(save_directory):
+            vocab_file = os.path.join(
+                save_directory,
+                (filename_prefix + "-" if filename_prefix else "") + self.vocab_files_names["vocab_file"],
+            )
+            word_shape_file = os.path.join(
+                save_directory,
+                (filename_prefix + "-" if filename_prefix else "") + self.vocab_files_names["word_shape_file"],
+            )
+            word_pronunciation_file = os.path.join(
+                save_directory,
+                (filename_prefix + "-" if filename_prefix else "") + self.vocab_files_names["word_pronunciation_file"],
+            )
+        else:
+            raise ValueError(
+                f"Can't find a directory at path '{save_directory}'. To load the vocabulary from a Google "
+                "pretrained model use `tokenizer = RoCBertTokenizer.from_pretrained(PRETRAINED_MODEL_NAME)`"
+            )
+
+        with open(vocab_file, "w", encoding="utf-8") as writer:
+            for token, token_index in sorted(self.vocab.items(), key=lambda kv: kv[1]):
+                if index != token_index:
+                    logger.warning(
+                        f"Saving vocabulary to {vocab_file}: vocabulary indices are not consecutive."
+                        " Please check that the vocabulary is not corrupted!"
+                    )
+                    index = token_index
+                writer.write(token + "\n")
+                index += 1
+
+        with open(word_shape_file, "w", encoding="utf8") as writer:
+            json.dump(self.word_shape, writer, ensure_ascii=False, indent=4, separators=(", ", ": "))
+
+        with open(word_pronunciation_file, "w", encoding="utf8") as writer:
+            json.dump(self.word_pronunciation, writer, ensure_ascii=False, indent=4, separators=(", ", ": "))
+
+        return (
+            vocab_file,
+            word_shape_file,
+            word_pronunciation_file,
+        )
+
+
+# Copied from  transformers.models.bert.tokenization_bert.BasicTokenizer with BasicTokenizer->RoCBertBasicTokenizer
+class RoCBertBasicTokenizer(object):
+    """
+    Constructs a RoCBertBasicTokenizer that will run basic tokenization (punctuation splitting, lower casing, etc.).
+
+    Args:
+        do_lower_case (`bool`, *optional*, defaults to `True`):
+            Whether or not to lowercase the input when tokenizing.
+        never_split (`Iterable`, *optional*):
+            Collection of tokens which will never be split during tokenization. Only has an effect when
+            `do_basic_tokenize=True`
+        tokenize_chinese_chars (`bool`, *optional*, defaults to `True`):
+            Whether or not to tokenize Chinese characters.
+
+            This should likely be deactivated for Japanese (see this
+            [issue](https://github.com/huggingface/transformers/issues/328)).
+        strip_accents (`bool`, *optional*):
+            Whether or not to strip all accents. If this option is not specified, then it will be determined by the
+            value for `lowercase` (as in the original BERT).
+    """
+
+    def __init__(self, do_lower_case=True, never_split=None, tokenize_chinese_chars=True, strip_accents=None):
+        if never_split is None:
+            never_split = []
+        self.do_lower_case = do_lower_case
+        self.never_split = set(never_split)
+        self.tokenize_chinese_chars = tokenize_chinese_chars
+        self.strip_accents = strip_accents
+
+    def tokenize(self, text, never_split=None):
+        """
+        Basic Tokenization of a piece of text. Split on "white spaces" only, for sub-word tokenization, see
+        WordPieceTokenizer.
+
+        Args:
+            never_split (`List[str]`, *optional*)
+                Kept for backward compatibility purposes. Now implemented directly at the base class level (see
+                [`PreTrainedTokenizer.tokenize`]) List of token not to split.
+        """
+        # union() returns a new set by concatenating the two sets.
+        never_split = self.never_split.union(set(never_split)) if never_split else self.never_split
+        text = self._clean_text(text)
+
+        # This was added on November 1st, 2018 for the multilingual and Chinese
+        # models. This is also applied to the English models now, but it doesn't
+        # matter since the English models were not trained on any Chinese data
+        # and generally don't have any Chinese data in them (there are Chinese
+        # characters in the vocabulary because Wikipedia does have some Chinese
+        # words in the English Wikipedia.).
+        if self.tokenize_chinese_chars:
+            text = self._tokenize_chinese_chars(text)
+        orig_tokens = whitespace_tokenize(text)
+        split_tokens = []
+        for token in orig_tokens:
+            if token not in never_split:
+                if self.do_lower_case:
+                    token = token.lower()
+                    if self.strip_accents is not False:
+                        token = self._run_strip_accents(token)
+                elif self.strip_accents:
+                    token = self._run_strip_accents(token)
+            split_tokens.extend(self._run_split_on_punc(token, never_split))
+
+        output_tokens = whitespace_tokenize(" ".join(split_tokens))
+        return output_tokens
+
+    def _run_strip_accents(self, text):
+        """Strips accents from a piece of text."""
+        text = unicodedata.normalize("NFD", text)
+        output = []
+        for char in text:
+            cat = unicodedata.category(char)
+            if cat == "Mn":
+                continue
+            output.append(char)
+        return "".join(output)
+
+    def _run_split_on_punc(self, text, never_split=None):
+        """Splits punctuation on a piece of text."""
+        if never_split is not None and text in never_split:
+            return [text]
+        chars = list(text)
+        i = 0
+        start_new_word = True
+        output = []
+        while i < len(chars):
+            char = chars[i]
+            if _is_punctuation(char):
+                output.append([char])
+                start_new_word = True
+            else:
+                if start_new_word:
+                    output.append([])
+                start_new_word = False
+                output[-1].append(char)
+            i += 1
+
+        return ["".join(x) for x in output]
+
+    def _tokenize_chinese_chars(self, text):
+        """Adds whitespace around any CJK character."""
+        output = []
+        for char in text:
+            cp = ord(char)
+            if self._is_chinese_char(cp):
+                output.append(" ")
+                output.append(char)
+                output.append(" ")
+            else:
+                output.append(char)
+        return "".join(output)
+
+    def _is_chinese_char(self, cp):
+        """Checks whether CP is the codepoint of a CJK character."""
+        # This defines a "chinese character" as anything in the CJK Unicode block:
+        #   https://en.wikipedia.org/wiki/CJK_Unified_Ideographs_(Unicode_block)
+        #
+        # Note that the CJK Unicode block is NOT all Japanese and Korean characters,
+        # despite its name. The modern Korean Hangul alphabet is a different block,
+        # as is Japanese Hiragana and Katakana. Those alphabets are used to write
+        # space-separated words, so they are not treated specially and handled
+        # like the all of the other languages.
+        if (
+            (cp >= 0x4E00 and cp <= 0x9FFF)
+            or (cp >= 0x3400 and cp <= 0x4DBF)  #
+            or (cp >= 0x20000 and cp <= 0x2A6DF)  #
+            or (cp >= 0x2A700 and cp <= 0x2B73F)  #
+            or (cp >= 0x2B740 and cp <= 0x2B81F)  #
+            or (cp >= 0x2B820 and cp <= 0x2CEAF)  #
+            or (cp >= 0xF900 and cp <= 0xFAFF)
+            or (cp >= 0x2F800 and cp <= 0x2FA1F)  #
+        ):  #
+            return True
+
+        return False
+
+    def _clean_text(self, text):
+        """Performs invalid character removal and whitespace cleanup on text."""
+        output = []
+        for char in text:
+            cp = ord(char)
+            if cp == 0 or cp == 0xFFFD or _is_control(char):
+                continue
+            if _is_whitespace(char):
+                output.append(" ")
+            else:
+                output.append(char)
+        return "".join(output)
+
+
+# Copied from  transformers.models.bert.tokenization_bert.WordpieceTokenizer with WordpieceTokenizer->RoCBertWordpieceTokenizer
+class RoCBertWordpieceTokenizer(object):
+    """Runs WordPiece tokenization."""
+
+    def __init__(self, vocab, unk_token, max_input_chars_per_word=100):
+        self.vocab = vocab
+        self.unk_token = unk_token
+        self.max_input_chars_per_word = max_input_chars_per_word
+
+    def tokenize(self, text):
+        """
+        Tokenizes a piece of text into its word pieces. This uses a greedy longest-match-first algorithm to perform
+        tokenization using the given vocabulary.
+
+        For example, `input = "unaffable"` wil return as output `["un", "##aff", "##able"]`.
+
+        Args:
+            text: A single token or whitespace separated tokens. This should have
+                already been passed through *BasicTokenizer*.
+
+        Returns:
+            A list of wordpiece tokens.
+        """
+
+        output_tokens = []
+        for token in whitespace_tokenize(text):
+            chars = list(token)
+            if len(chars) > self.max_input_chars_per_word:
+                output_tokens.append(self.unk_token)
+                continue
+
+            is_bad = False
+            start = 0
+            sub_tokens = []
+            while start < len(chars):
+                end = len(chars)
+                cur_substr = None
+                while start < end:
+                    substr = "".join(chars[start:end])
+                    if start > 0:
+                        substr = "##" + substr
+                    if substr in self.vocab:
+                        cur_substr = substr
+                        break
+                    end -= 1
+                if cur_substr is None:
+                    is_bad = True
+                    break
+                sub_tokens.append(cur_substr)
+                start = end
+
+            if is_bad:
+                output_tokens.append(self.unk_token)
+            else:
+                output_tokens.extend(sub_tokens)
+        return output_tokens

src/transformers/utils/dummy_pt_objects.py

@@ -4533,6 +4533,83 @@ class RobertaPreTrainedModel(metaclass=DummyObject):
         requires_backends(self, ["torch"])
 
 
+ROC_BERT_PRETRAINED_MODEL_ARCHIVE_LIST = None
+
+
+class RoCBertForCausalLM(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class RoCBertForMaskedLM(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class RoCBertForMultipleChoice(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class RoCBertForPreTraining(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class RoCBertForQuestionAnswering(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class RoCBertForSequenceClassification(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class RoCBertForTokenClassification(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class RoCBertLayer(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class RoCBertModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class RoCBertPreTrainedModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+def load_tf_weights_in_roc_bert(*args, **kwargs):
+    requires_backends(load_tf_weights_in_roc_bert, ["torch"])
+
+
 ROFORMER_PRETRAINED_MODEL_ARCHIVE_LIST = None
 
 

tests/models/roc_bert/test_modeling_roc_bert.py

+# coding=utf-8
+# Copyright 2022 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 RoCBert model. """
+
+import unittest
+
+from transformers import RoCBertConfig, is_torch_available
+from transformers.models.auto import get_values
+from transformers.testing_utils import require_torch, slow, torch_device
+
+from ...test_configuration_common import ConfigTester
+from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor, random_attention_mask
+
+
+if is_torch_available():
+    import torch
+
+    from transformers import (
+        MODEL_FOR_PRETRAINING_MAPPING,
+        RoCBertForCausalLM,
+        RoCBertForMaskedLM,
+        RoCBertForMultipleChoice,
+        RoCBertForPreTraining,
+        RoCBertForQuestionAnswering,
+        RoCBertForSequenceClassification,
+        RoCBertForTokenClassification,
+        RoCBertModel,
+    )
+    from transformers.models.roc_bert.modeling_roc_bert import ROC_BERT_PRETRAINED_MODEL_ARCHIVE_LIST
+
+
+class RoCBertModelTester:
+    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,
+        vocab_size=99,
+        pronunciation_vocab_size=99,
+        shape_vocab_size=99,
+        pronunciation_embed_dim=32,
+        shape_embed_dim=32,
+        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.vocab_size = vocab_size
+        self.pronunciation_vocab_size = pronunciation_vocab_size
+        self.shape_vocab_size = shape_vocab_size
+        self.pronunciation_embed_dim = pronunciation_embed_dim
+        self.shape_embed_dim = shape_embed_dim
+        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):
+        input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
+        input_shape_ids = ids_tensor([self.batch_size, self.seq_length], self.shape_vocab_size)
+        input_pronunciation_ids = ids_tensor([self.batch_size, self.seq_length], self.pronunciation_vocab_size)
+
+        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([self.batch_size, self.seq_length], 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 = self.get_config()
+
+        return (
+            config,
+            input_ids,
+            input_shape_ids,
+            input_pronunciation_ids,
+            token_type_ids,
+            input_mask,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+        )
+
+    def get_config(self):
+        return RoCBertConfig(
+            vocab_size=self.vocab_size,
+            shape_vocab_size=self.shape_vocab_size,
+            pronunciation_vocab_size=self.pronunciation_vocab_size,
+            shape_embed_dim=self.shape_embed_dim,
+            pronunciation_embed_dim=self.pronunciation_embed_dim,
+            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,
+        )
+
+    def prepare_config_and_inputs_for_decoder(self):
+        (
+            config,
+            input_ids,
+            input_shape_ids,
+            input_pronunciation_ids,
+            token_type_ids,
+            input_mask,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+        ) = self.prepare_config_and_inputs()
+
+        config.is_decoder = True
+        encoder_hidden_states = floats_tensor([self.batch_size, self.seq_length, self.hidden_size])
+        encoder_attention_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
+
+        return (
+            config,
+            input_ids,
+            input_shape_ids,
+            input_pronunciation_ids,
+            token_type_ids,
+            input_mask,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+            encoder_hidden_states,
+            encoder_attention_mask,
+        )
+
+    def create_and_check_model(
+        self,
+        config,
+        input_ids,
+        input_shape_ids,
+        input_pronunciation_ids,
+        token_type_ids,
+        input_mask,
+        sequence_labels,
+        token_labels,
+        choice_labels,
+    ):
+        model = RoCBertModel(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_ids,
+            attention_mask=input_mask,
+            token_type_ids=token_type_ids,
+        )
+        result = model(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_ids,
+            token_type_ids=token_type_ids,
+        )
+        result = model(input_ids, input_shape_ids=input_shape_ids, input_pronunciation_ids=input_pronunciation_ids)
+        self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
+
+    def create_and_check_model_as_decoder(
+        self,
+        config,
+        input_ids,
+        input_shape_ids,
+        input_pronunciation_ids,
+        token_type_ids,
+        input_mask,
+        sequence_labels,
+        token_labels,
+        choice_labels,
+        encoder_hidden_states,
+        encoder_attention_mask,
+    ):
+        config.add_cross_attention = True
+        model = RoCBertModel(config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_ids,
+            attention_mask=input_mask,
+            token_type_ids=token_type_ids,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+        )
+        result = model(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_ids,
+            attention_mask=input_mask,
+            token_type_ids=token_type_ids,
+            encoder_hidden_states=encoder_hidden_states,
+        )
+        result = model(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_ids,
+            attention_mask=input_mask,
+            token_type_ids=token_type_ids,
+        )
+        self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
+
+    def create_and_check_for_causal_lm(
+        self,
+        config,
+        input_ids,
+        input_shape_ids,
+        input_pronunciation_ids,
+        token_type_ids,
+        input_mask,
+        sequence_labels,
+        token_labels,
+        choice_labels,
+        encoder_hidden_states,
+        encoder_attention_mask,
+    ):
+        model = RoCBertForCausalLM(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_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.vocab_size))
+
+    def create_and_check_for_masked_lm(
+        self,
+        config,
+        input_ids,
+        input_shape_ids,
+        input_pronunciation_ids,
+        token_type_ids,
+        input_mask,
+        sequence_labels,
+        token_labels,
+        choice_labels,
+    ):
+        model = RoCBertForMaskedLM(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_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.vocab_size))
+
+    def create_and_check_decoder_model_past_large_inputs(
+        self,
+        config,
+        input_ids,
+        input_shape_ids,
+        input_pronunciation_ids,
+        token_type_ids,
+        input_mask,
+        sequence_labels,
+        token_labels,
+        choice_labels,
+        encoder_hidden_states,
+        encoder_attention_mask,
+    ):
+        config.is_decoder = True
+        config.add_cross_attention = True
+        model = RoCBertForCausalLM(config=config)
+        model.to(torch_device)
+        model.eval()
+
+        # first forward pass
+        outputs = model(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_ids,
+            attention_mask=input_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+            use_cache=True,
+        )
+        past_key_values = outputs.past_key_values
+
+        # create hypothetical multiple next token and extent to next_input_ids
+        next_tokens = ids_tensor((self.batch_size, 3), config.vocab_size)
+        next_shape_tokens = ids_tensor((self.batch_size, 3), config.shape_vocab_size)
+        next_pronunciation_tokens = ids_tensor((self.batch_size, 3), config.pronunciation_vocab_size)
+        next_mask = ids_tensor((self.batch_size, 3), vocab_size=2)
+
+        # append to next input_ids and
+        next_input_ids = torch.cat([input_ids, next_tokens], dim=-1)
+        next_input_shape_ids = torch.cat([input_shape_ids, next_shape_tokens], dim=-1)
+        next_input_pronunciation_ids = torch.cat([input_pronunciation_ids, next_pronunciation_tokens], dim=-1)
+        next_attention_mask = torch.cat([input_mask, next_mask], dim=-1)
+
+        output_from_no_past = model(
+            next_input_ids,
+            input_shape_ids=next_input_shape_ids,
+            input_pronunciation_ids=next_input_pronunciation_ids,
+            attention_mask=next_attention_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+            output_hidden_states=True,
+        )["hidden_states"][0]
+        output_from_past = model(
+            next_tokens,
+            input_shape_ids=next_shape_tokens,
+            input_pronunciation_ids=next_pronunciation_tokens,
+            attention_mask=next_attention_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+            past_key_values=past_key_values,
+            output_hidden_states=True,
+        )["hidden_states"][0]
+
+        # select random slice
+        random_slice_idx = ids_tensor((1,), output_from_past.shape[-1]).item()
+        output_from_no_past_slice = output_from_no_past[:, -3:, random_slice_idx].detach()
+        output_from_past_slice = output_from_past[:, :, random_slice_idx].detach()
+
+        self.parent.assertTrue(output_from_past_slice.shape[1] == next_tokens.shape[1])
+
+        # test that outputs are equal for slice
+        self.parent.assertTrue(torch.allclose(output_from_past_slice, output_from_no_past_slice, atol=1e-3))
+
+    def create_and_check_for_question_answering(
+        self,
+        config,
+        input_ids,
+        input_shape_ids,
+        input_pronunciation_ids,
+        token_type_ids,
+        input_mask,
+        sequence_labels,
+        token_labels,
+        choice_labels,
+    ):
+        model = RoCBertForQuestionAnswering(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_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,
+        input_shape_ids,
+        input_pronunciation_ids,
+        token_type_ids,
+        input_mask,
+        sequence_labels,
+        token_labels,
+        choice_labels,
+    ):
+        config.num_labels = self.num_labels
+        model = RoCBertForSequenceClassification(config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_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,
+        input_shape_ids,
+        input_pronunciation_ids,
+        token_type_ids,
+        input_mask,
+        sequence_labels,
+        token_labels,
+        choice_labels,
+    ):
+        config.num_labels = self.num_labels
+        model = RoCBertForTokenClassification(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_ids,
+            input_shape_ids=input_shape_ids,
+            input_pronunciation_ids=input_pronunciation_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,
+        input_shape_ids,
+        input_pronunciation_ids,
+        token_type_ids,
+        input_mask,
+        sequence_labels,
+        token_labels,
+        choice_labels,
+    ):
+        config.num_choices = self.num_choices
+        model = RoCBertForMultipleChoice(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_inputs_shape_ids = input_shape_ids.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
+        multiple_choice_inputs_pronunciation_ids = (
+            input_pronunciation_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,
+            input_shape_ids=multiple_choice_inputs_shape_ids,
+            input_pronunciation_ids=multiple_choice_inputs_pronunciation_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,
+            input_shape_ids,
+            input_pronunciation_ids,
+            token_type_ids,
+            input_mask,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+        ) = config_and_inputs
+        inputs_dict = {
+            "input_ids": input_ids,
+            "input_shape_ids": input_shape_ids,
+            "input_pronunciation_ids": input_pronunciation_ids,
+            "token_type_ids": token_type_ids,
+            "attention_mask": input_mask,
+        }
+        return config, inputs_dict
+
+    def create_and_check_for_pretraining(
+        self,
+        config,
+        input_ids,
+        input_shape_ids,
+        input_pronunciation_ids,
+        token_type_ids,
+        input_mask,
+        sequence_labels,
+        token_labels,
+        choice_labels,
+    ):
+        model = RoCBertForPreTraining(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_ids,
+            input_shape_ids,
+            input_pronunciation_ids,
+            attention_mask=input_mask,
+            token_type_ids=token_type_ids,
+            attack_input_ids=input_ids,
+            attack_input_shape_ids=input_shape_ids,
+            attack_input_pronunciation_ids=input_pronunciation_ids,
+            attack_attention_mask=input_mask,
+            attack_token_type_ids=token_type_ids,
+            labels_input_ids=token_labels,
+            labels_input_shape_ids=input_shape_ids,
+            labels_input_pronunciation_ids=input_pronunciation_ids,
+            labels_attention_mask=input_mask,
+            labels_token_type_ids=token_type_ids,
+        )
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
+
+
+@require_torch
+class RoCBertModelTest(ModelTesterMixin, unittest.TestCase):
+    all_model_classes = (
+        (
+            RoCBertModel,
+            RoCBertForMaskedLM,
+            RoCBertForCausalLM,
+            RoCBertForMultipleChoice,
+            RoCBertForQuestionAnswering,
+            RoCBertForSequenceClassification,
+            RoCBertForTokenClassification,
+            RoCBertForPreTraining,
+        )
+        if is_torch_available()
+        else ()
+    )
+    all_generative_model_classes = (RoCBertForCausalLM,) if is_torch_available() else ()
+
+    # special case for ForPreTraining model
+    def _prepare_for_class(self, inputs_dict, model_class, return_labels=False):
+        inputs_dict = super()._prepare_for_class(inputs_dict, model_class, return_labels=return_labels)
+
+        if return_labels:
+            if model_class in get_values(MODEL_FOR_PRETRAINING_MAPPING):
+                inputs_dict["labels_input_ids"] = torch.zeros(
+                    (self.model_tester.batch_size, self.model_tester.seq_length), dtype=torch.long, device=torch_device
+                )
+                inputs_dict["labels_input_shape_ids"] = torch.zeros(
+                    (self.model_tester.batch_size, self.model_tester.seq_length), dtype=torch.long, device=torch_device
+                )
+                inputs_dict["labels_input_pronunciation_ids"] = torch.zeros(
+                    (self.model_tester.batch_size, self.model_tester.seq_length), dtype=torch.long, device=torch_device
+                )
+                inputs_dict["attack_input_ids"] = torch.zeros(
+                    (self.model_tester.batch_size, self.model_tester.seq_length), dtype=torch.long, device=torch_device
+                )
+                inputs_dict["attack_input_shape_ids"] = torch.zeros(
+                    (self.model_tester.batch_size, self.model_tester.seq_length), dtype=torch.long, device=torch_device
+                )
+                inputs_dict["attack_input_pronunciation_ids"] = torch.zeros(
+                    (self.model_tester.batch_size, self.model_tester.seq_length), dtype=torch.long, device=torch_device
+                )
+        return inputs_dict
+
+    def setUp(self):
+        self.model_tester = RoCBertModelTester(self)
+        self.config_tester = ConfigTester(self, config_class=RoCBertConfig, 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_model_various_embeddings(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        for type in ["absolute", "relative_key", "relative_key_query"]:
+            config_and_inputs[0].position_embedding_type = type
+            self.model_tester.create_and_check_model(*config_and_inputs)
+
+    def test_for_masked_lm(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_for_masked_lm(*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_decoder_model_past_with_large_inputs(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs_for_decoder()
+        self.model_tester.create_and_check_decoder_model_past_large_inputs(*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_for_pretraining(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_for_pretraining(*config_and_inputs)
+
+    def test_model_as_decoder(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs_for_decoder()
+        self.model_tester.create_and_check_model_as_decoder(*config_and_inputs)
+
+    def test_model_as_decoder_with_default_input_mask(self):
+        # This regression test was failing with PyTorch < 1.3
+        (
+            config,
+            input_ids,
+            input_shape_ids,
+            input_pronunciation_ids,
+            token_type_ids,
+            input_mask,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+            encoder_hidden_states,
+            encoder_attention_mask,
+        ) = self.model_tester.prepare_config_and_inputs_for_decoder()
+
+        input_mask = None
+
+        self.model_tester.create_and_check_model_as_decoder(
+            config,
+            input_ids,
+            input_shape_ids,
+            input_pronunciation_ids,
+            token_type_ids,
+            input_mask,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+            encoder_hidden_states,
+            encoder_attention_mask,
+        )
+
+    @slow
+    def test_model_from_pretrained(self):
+        for model_name in ROC_BERT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
+            model = RoCBertModel.from_pretrained(model_name)
+            self.assertIsNotNone(model)
+
+
+@require_torch
+class RoCBertModelIntegrationTest(unittest.TestCase):
+    @slow
+    def test_inference_masked_lm(self):
+        model = RoCBertForMaskedLM.from_pretrained("weiweishi/roc-bert-base-zh")
+
+        # input_text: ['[CLS]', 'b', 'a', '里', '系', '[MASK]', '国', '的', '首', '都', '[SEP]'] is the adversarial text
+        # of ['[CLS]', '巴', '黎', '是', '[MASK]', '国', '的', '首', '都', '[SEP]'], means
+        # "Paris is the [MASK] of France" in English
+        input_ids = torch.tensor([[101, 144, 143, 7027, 5143, 103, 1744, 4638, 7674, 6963, 102]])
+        input_shape_ids = torch.tensor([[2, 20324, 23690, 8740, 706, 1, 10900, 23343, 20205, 5850, 2]])
+        input_pronunciation_ids = torch.tensor([[2, 718, 397, 52, 61, 1, 168, 273, 180, 243, 2]])
+
+        output = model(input_ids, input_shape_ids, input_pronunciation_ids)
+        output_ids = torch.argmax(output.logits, dim=2)
+
+        # convert to tokens is: ['[CLS]', '巴', '*', '黎', '是', '法', '国', '的', '首', '都', '[SEP]']
+        expected_output = torch.tensor([[101, 2349, 115, 7944, 3221, 3791, 1744, 4638, 7674, 6963, 102]])
+
+        self.assertTrue(output_ids, expected_output)

tests/models/roc_bert/test_tokenization_roc_bert.py

+# coding=utf-8
+# 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.
+
+
+import json
+import os
+import unittest
+
+from transformers.models.roc_bert.tokenization_roc_bert import (
+    VOCAB_FILES_NAMES,
+    RoCBertBasicTokenizer,
+    RoCBertTokenizer,
+    RoCBertWordpieceTokenizer,
+    _is_control,
+    _is_punctuation,
+    _is_whitespace,
+)
+from transformers.testing_utils import require_tokenizers, slow
+
+from ...test_tokenization_common import TokenizerTesterMixin, filter_non_english
+
+
+@require_tokenizers
+class BertTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
+    tokenizer_class = RoCBertTokenizer
+    rust_tokenizer_class = None
+    test_rust_tokenizer = False
+    space_between_special_tokens = True
+    from_pretrained_filter = filter_non_english
+
+    def setUp(self):
+        super().setUp()
+
+        vocab_tokens = ["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]", "你", "好", "是", "谁", "a", "b", "c", "d"]
+        word_shape = dict()
+        word_pronunciation = dict()
+        for i, value in enumerate(vocab_tokens):
+            word_shape[value] = i
+            word_pronunciation[value] = i
+        self.vocab_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES["vocab_file"])
+        self.word_shape_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES["word_shape_file"])
+        self.word_pronunciation_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES["word_pronunciation_file"])
+        with open(self.vocab_file, "w", encoding="utf-8") as vocab_writer:
+            vocab_writer.write("".join([x + "\n" for x in vocab_tokens]))
+        with open(self.word_shape_file, "w", encoding="utf-8") as word_shape_writer:
+            json.dump(word_shape, word_shape_writer, ensure_ascii=False)
+        with open(self.word_pronunciation_file, "w", encoding="utf-8") as word_pronunciation_writer:
+            json.dump(word_pronunciation, word_pronunciation_writer, ensure_ascii=False)
+
+    def test_full_tokenizer(self):
+        tokenizer = self.tokenizer_class(self.vocab_file, self.word_shape_file, self.word_pronunciation_file)
+
+        tokens = tokenizer.tokenize("你好[SEP]你是谁")
+        self.assertListEqual(tokens, ["你", "好", "[SEP]", "你", "是", "谁"])
+        self.assertListEqual(tokenizer.convert_tokens_to_ids(tokens), [5, 6, 2, 5, 7, 8])
+        self.assertListEqual(tokenizer.convert_tokens_to_shape_ids(tokens), [5, 6, 2, 5, 7, 8])
+        self.assertListEqual(tokenizer.convert_tokens_to_pronunciation_ids(tokens), [5, 6, 2, 5, 7, 8])
+
+    # Copied from tests.models.bert.test_tokenization_bert.test_chinese with BasicTokenizer->RoCBertBertBasicTokenizer
+    def test_chinese(self):
+        tokenizer = RoCBertBasicTokenizer()
+
+        self.assertListEqual(tokenizer.tokenize("ah\u535A\u63A8zz"), ["ah", "\u535A", "\u63A8", "zz"])
+
+    # Copied from tests.models.bert.test_tokenization_bert.test_basic_tokenizer_lower with BasicTokenizer->RoCBertBertBasicTokenizer
+    def test_basic_tokenizer_lower(self):
+        tokenizer = RoCBertBasicTokenizer(do_lower_case=True)
+
+        self.assertListEqual(
+            tokenizer.tokenize(" \tHeLLo!how  \n Are yoU?  "), ["hello", "!", "how", "are", "you", "?"]
+        )
+        self.assertListEqual(tokenizer.tokenize("H\u00E9llo"), ["hello"])
+
+    # Copied from tests.models.bert.test_tokenization_bert.test_basic_tokenizer_lower_strip_accents_false with BasicTokenizer->RoCBertBertBasicTokenizer
+    def test_basic_tokenizer_lower_strip_accents_false(self):
+        tokenizer = RoCBertBasicTokenizer(do_lower_case=True, strip_accents=False)
+
+        self.assertListEqual(
+            tokenizer.tokenize(" \tHäLLo!how  \n Are yoU?  "), ["hällo", "!", "how", "are", "you", "?"]
+        )
+        self.assertListEqual(tokenizer.tokenize("H\u00E9llo"), ["h\u00E9llo"])
+
+    # Copied from tests.models.bert.test_tokenization_bert.test_basic_tokenizer_lower_strip_accents_true with BertBasicTokenizer->RoCBertBertBasicTokenizer
+    def test_basic_tokenizer_lower_strip_accents_true(self):
+        tokenizer = RoCBertBasicTokenizer(do_lower_case=True, strip_accents=True)
+
+        self.assertListEqual(
+            tokenizer.tokenize(" \tHäLLo!how  \n Are yoU?  "), ["hallo", "!", "how", "are", "you", "?"]
+        )
+        self.assertListEqual(tokenizer.tokenize("H\u00E9llo"), ["hello"])
+
+    # Copied from tests.models.bert.test_tokenization_bert.test_basic_tokenizer_lower_strip_accents_default with BasicTokenizer->RoCBertBertBasicTokenizer
+    def test_basic_tokenizer_lower_strip_accents_default(self):
+        tokenizer = RoCBertBasicTokenizer(do_lower_case=True)
+
+        self.assertListEqual(
+            tokenizer.tokenize(" \tHäLLo!how  \n Are yoU?  "), ["hallo", "!", "how", "are", "you", "?"]
+        )
+        self.assertListEqual(tokenizer.tokenize("H\u00E9llo"), ["hello"])
+
+    # Copied from tests.models.bert.test_tokenization_bert.test_basic_tokenizer_no_lower with BasicTokenizer->RoCBertBertBasicTokenizer
+    def test_basic_tokenizer_no_lower(self):
+        tokenizer = RoCBertBasicTokenizer(do_lower_case=False)
+
+        self.assertListEqual(
+            tokenizer.tokenize(" \tHeLLo!how  \n Are yoU?  "), ["HeLLo", "!", "how", "Are", "yoU", "?"]
+        )
+
+    # Copied from tests.models.bert.test_tokenization_bert.test_basic_tokenizer_no_lower_strip_accents_false with BertBasicTokenizer->RoCBertBertBasicTokenizer
+    def test_basic_tokenizer_no_lower_strip_accents_false(self):
+        tokenizer = RoCBertBasicTokenizer(do_lower_case=False, strip_accents=False)
+
+        self.assertListEqual(
+            tokenizer.tokenize(" \tHäLLo!how  \n Are yoU?  "), ["HäLLo", "!", "how", "Are", "yoU", "?"]
+        )
+
+    # Copied from tests.models.bert.test_tokenization_bert.test_basic_tokenizer_no_lower_strip_accents_true with BasicTokenizer->RoCBertBertBasicTokenizer
+    def test_basic_tokenizer_no_lower_strip_accents_true(self):
+        tokenizer = RoCBertBasicTokenizer(do_lower_case=False, strip_accents=True)
+
+        self.assertListEqual(
+            tokenizer.tokenize(" \tHäLLo!how  \n Are yoU?  "), ["HaLLo", "!", "how", "Are", "yoU", "?"]
+        )
+
+    # Copied from tests.models.bert.test_tokenization_bert.test_basic_tokenizer_respects_never_split_tokens with BasicTokenizer->RoCBertBertBasicTokenizer
+    def test_basic_tokenizer_respects_never_split_tokens(self):
+        tokenizer = RoCBertBasicTokenizer(do_lower_case=False, never_split=["[UNK]"])
+
+        self.assertListEqual(
+            tokenizer.tokenize(" \tHeLLo!how  \n Are yoU? [UNK]"), ["HeLLo", "!", "how", "Are", "yoU", "?", "[UNK]"]
+        )
+
+    # Copied from tests.models.bert.test_tokenization_bert.test_wordpiece_tokenizer with WordpieceTokenizer->RoCBertWordpieceTokenizer
+    def test_wordpiece_tokenizer(self):
+        vocab_tokens = ["[UNK]", "[CLS]", "[SEP]", "want", "##want", "##ed", "wa", "un", "runn", "##ing"]
+
+        vocab = {}
+        for i, token in enumerate(vocab_tokens):
+            vocab[token] = i
+        tokenizer = RoCBertWordpieceTokenizer(vocab=vocab, unk_token="[UNK]")
+
+        self.assertListEqual(tokenizer.tokenize(""), [])
+
+        self.assertListEqual(tokenizer.tokenize("unwanted running"), ["un", "##want", "##ed", "runn", "##ing"])
+
+        self.assertListEqual(tokenizer.tokenize("unwantedX running"), ["[UNK]", "runn", "##ing"])
+
+    # Copied from tests.models.bert.test_tokenization_bert.test_is_whitespace
+    def test_is_whitespace(self):
+        self.assertTrue(_is_whitespace(" "))
+        self.assertTrue(_is_whitespace("\t"))
+        self.assertTrue(_is_whitespace("\r"))
+        self.assertTrue(_is_whitespace("\n"))
+        self.assertTrue(_is_whitespace("\u00A0"))
+
+        self.assertFalse(_is_whitespace("A"))
+        self.assertFalse(_is_whitespace("-"))
+
+    # Copied from tests.models.bert.test_tokenization_bert.test_is_control
+    def test_is_control(self):
+        self.assertTrue(_is_control("\u0005"))
+
+        self.assertFalse(_is_control("A"))
+        self.assertFalse(_is_control(" "))
+        self.assertFalse(_is_control("\t"))
+        self.assertFalse(_is_control("\r"))
+
+    # Copied from tests.models.bert.test_tokenization_bert.test_is_punctuation
+    def test_is_punctuation(self):
+        self.assertTrue(_is_punctuation("-"))
+        self.assertTrue(_is_punctuation("$"))
+        self.assertTrue(_is_punctuation("`"))
+        self.assertTrue(_is_punctuation("."))
+
+        self.assertFalse(_is_punctuation("A"))
+        self.assertFalse(_is_punctuation(" "))
+
+    def test_clean_text(self):
+        tokenizer = self.get_tokenizer()
+
+        # Example taken from the issue https://github.com/huggingface/tokenizers/issues/340
+        self.assertListEqual([tokenizer.tokenize(t) for t in ["Test", "\xad", "test"]], [["[UNK]"], [], ["[UNK]"]])
+
+        if self.test_rust_tokenizer:
+            rust_tokenizer = self.get_rust_tokenizer()
+            self.assertListEqual(
+                [rust_tokenizer.tokenize(t) for t in ["Test", "\xad", "test"]], [["[UNK]"], [], ["[UNK]"]]
+            )
+
+    # Copied from tests.models.bert.test_tokenization_bert. test_offsets_with_special_characters
+    def test_offsets_with_special_characters(self):
+        for tokenizer, pretrained_name, kwargs in self.tokenizers_list:
+            with self.subTest(f"{tokenizer.__class__.__name__} ({pretrained_name})"):
+                tokenizer_r = self.rust_tokenizer_class.from_pretrained(pretrained_name, **kwargs)
+
+                sentence = f"A, naïve {tokenizer_r.mask_token} AllenNLP sentence."
+                tokens = tokenizer_r.encode_plus(
+                    sentence,
+                    return_attention_mask=False,
+                    return_token_type_ids=False,
+                    return_offsets_mapping=True,
+                    add_special_tokens=True,
+                )
+
+                do_lower_case = tokenizer_r.do_lower_case if hasattr(tokenizer_r, "do_lower_case") else False
+                expected_results = (
+                    [
+                        ((0, 0), tokenizer_r.cls_token),
+                        ((0, 1), "A"),
+                        ((1, 2), ","),
+                        ((3, 5), "na"),
+                        ((5, 6), "##ï"),
+                        ((6, 8), "##ve"),
+                        ((9, 15), tokenizer_r.mask_token),
+                        ((16, 21), "Allen"),
+                        ((21, 23), "##NL"),
+                        ((23, 24), "##P"),
+                        ((25, 33), "sentence"),
+                        ((33, 34), "."),
+                        ((0, 0), tokenizer_r.sep_token),
+                    ]
+                    if not do_lower_case
+                    else [
+                        ((0, 0), tokenizer_r.cls_token),
+                        ((0, 1), "a"),
+                        ((1, 2), ","),
+                        ((3, 8), "naive"),
+                        ((9, 15), tokenizer_r.mask_token),
+                        ((16, 21), "allen"),
+                        ((21, 23), "##nl"),
+                        ((23, 24), "##p"),
+                        ((25, 33), "sentence"),
+                        ((33, 34), "."),
+                        ((0, 0), tokenizer_r.sep_token),
+                    ]
+                )
+
+                self.assertEqual(
+                    [e[1] for e in expected_results], tokenizer_r.convert_ids_to_tokens(tokens["input_ids"])
+                )
+                self.assertEqual([e[0] for e in expected_results], tokens["offset_mapping"])
+
+    # Copied from tests.models.bert.test_tokenization_bert. test_change_tokenize_chinese_chars
+    def test_change_tokenize_chinese_chars(self):
+        list_of_commun_chinese_char = ["的", "人", "有"]
+        text_with_chinese_char = "".join(list_of_commun_chinese_char)
+        for tokenizer, pretrained_name, kwargs in self.tokenizers_list:
+            with self.subTest(f"{tokenizer.__class__.__name__} ({pretrained_name})"):
+                kwargs["tokenize_chinese_chars"] = True
+                tokenizer_p = self.tokenizer_class.from_pretrained(pretrained_name, **kwargs)
+                tokenizer_r = self.rust_tokenizer_class.from_pretrained(pretrained_name, **kwargs)
+
+                ids_without_spe_char_p = tokenizer_p.encode(text_with_chinese_char, add_special_tokens=False)
+                ids_without_spe_char_r = tokenizer_r.encode(text_with_chinese_char, add_special_tokens=False)
+
+                tokens_without_spe_char_r = tokenizer_r.convert_ids_to_tokens(ids_without_spe_char_r)
+                tokens_without_spe_char_p = tokenizer_p.convert_ids_to_tokens(ids_without_spe_char_p)
+
+                # it is expected that each Chinese character is not preceded by "##"
+                self.assertListEqual(tokens_without_spe_char_p, list_of_commun_chinese_char)
+                self.assertListEqual(tokens_without_spe_char_r, list_of_commun_chinese_char)
+
+                kwargs["tokenize_chinese_chars"] = False
+                tokenizer_r = self.rust_tokenizer_class.from_pretrained(pretrained_name, **kwargs)
+                tokenizer_p = self.tokenizer_class.from_pretrained(pretrained_name, **kwargs)
+
+                ids_without_spe_char_r = tokenizer_r.encode(text_with_chinese_char, add_special_tokens=False)
+                ids_without_spe_char_p = tokenizer_p.encode(text_with_chinese_char, add_special_tokens=False)
+
+                tokens_without_spe_char_r = tokenizer_r.convert_ids_to_tokens(ids_without_spe_char_r)
+                tokens_without_spe_char_p = tokenizer_p.convert_ids_to_tokens(ids_without_spe_char_p)
+
+                # it is expected that only the first Chinese character is not preceded by "##".
+                expected_tokens = [
+                    f"##{token}" if idx != 0 else token for idx, token in enumerate(list_of_commun_chinese_char)
+                ]
+                self.assertListEqual(tokens_without_spe_char_p, expected_tokens)
+                self.assertListEqual(tokens_without_spe_char_r, expected_tokens)
+
+    @slow
+    def test_sequence_builders(self):
+        tokenizer = self.tokenizer_class(self.vocab_file, self.word_shape_file, self.word_pronunciation_file)
+
+        text = tokenizer.encode("你好", add_special_tokens=False)
+        text_2 = tokenizer.encode("你是谁", add_special_tokens=False)
+
+        encoded_sentence = tokenizer.build_inputs_with_special_tokens(text)
+        encoded_pair = tokenizer.build_inputs_with_special_tokens(text, text_2)
+
+        assert encoded_sentence == [101] + text + [102]
+        assert encoded_pair == [101] + text + [102] + text_2 + [102]
+
+    def test_prepare_for_model(self):
+        tokenizers = self.get_tokenizers(do_lower_case=False)
+        for tokenizer in tokenizers:
+            with self.subTest(f"{tokenizer.__class__.__name__}"):
+                string_sequence = "你好，你是谁"
+                tokens = tokenizer.tokenize(string_sequence)
+                tokens_ids = tokenizer.convert_tokens_to_ids(tokens)
+                tokens_shape_ids = tokenizer.convert_tokens_to_shape_ids(tokens)
+                tokens_proun_ids = tokenizer.convert_tokens_to_pronunciation_ids(tokens)
+                prepared_input_dict = tokenizer.prepare_for_model(
+                    tokens_ids, tokens_shape_ids, tokens_proun_ids, add_special_tokens=True
+                )
+
+                input_dict = tokenizer.encode_plus(string_sequence, add_special_tokens=True)
+
+                self.assertEqual(input_dict, prepared_input_dict)