add task_type_id to BERT to support ERNIE-2.0 and ERNIE-3.0 models (#18686)

* add_ernie * remove Tokenizer in ernie * polish code * format code style * polish code * fix style * update doc * make fix-copies * change model name * change model name * fix dependency * add more copied from * rename ErnieLMHeadModel to ErnieForCausalLM do not expose ErnieLayer update doc * fix * make style * polish code * polish code * fix * fix * fix * fix * fix * final fix Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>
2022-09-09 19:36:46 +08:00
parent 895c528886
commit 22f7218560
19 changed files with 2882 additions and 0 deletions
--- a/README.md
+++ b/README.md
@@ -295,6 +295,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[DPT](https://huggingface.co/docs/transformers/master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.
 1. **[ELECTRA](https://huggingface.co/docs/transformers/model_doc/electra)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
 1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (from Google Research) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+1. **[ERNIE](https://huggingface.co/docs/transformers/main/model_doc/ernie)** (from Baidu) released with the paper [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu.
 1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
 1. **[FLAVA](https://huggingface.co/docs/transformers/model_doc/flava)** (from Facebook AI) released with the paper [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) by Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela.
 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
--- a/README_ko.md
+++ b/README_ko.md
@@ -247,6 +247,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[DPT](https://huggingface.co/docs/transformers/master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.
 1. **[ELECTRA](https://huggingface.co/docs/transformers/model_doc/electra)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
 1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (from Google Research) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+1. **[ERNIE](https://huggingface.co/docs/transformers/main/model_doc/ernie)** (from Baidu) released with the paper [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu.
 1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
 1. **[FLAVA](https://huggingface.co/docs/transformers/model_doc/flava)** (from Facebook AI) released with the paper [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) by Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela.
 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
--- a/README_zh-hans.md
+++ b/README_zh-hans.md
@@ -271,6 +271,7 @@ conda install -c huggingface transformers
 1. **[DPT](https://huggingface.co/docs/transformers/master/model_doc/dpt)** (来自 Intel Labs) 伴随论文 [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) 由 René Ranftl, Alexey Bochkovskiy, Vladlen Koltun 发布。
 1. **[ELECTRA](https://huggingface.co/docs/transformers/model_doc/electra)** (来自 Google Research/Stanford University) 伴随论文 [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) 由 Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning 发布。
 1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (来自 Google Research) 伴随论文 [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) 由 Sascha Rothe, Shashi Narayan, Aliaksei Severyn 发布。
+1. **[ERNIE](https://huggingface.co/docs/transformers/main/model_doc/ernie)** (来自 Baidu) 伴随论文 [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu 发布。
 1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (来自 CNRS) 伴随论文 [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) 由 Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab 发布。
 1. **[FLAVA](https://huggingface.co/docs/transformers/model_doc/flava)** (来自 Facebook AI) 伴随论文 [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) 由 Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela 发布。
 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (来自 Google Research) 伴随论文 [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) 由 James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon 发布。
--- a/README_zh-hant.md
+++ b/README_zh-hant.md
@@ -283,6 +283,7 @@ conda install -c huggingface transformers
 1. **[DPT](https://huggingface.co/docs/transformers/master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.
 1. **[ELECTRA](https://huggingface.co/docs/transformers/model_doc/electra)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
 1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (from Google Research) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+1. **[ERNIE](https://huggingface.co/docs/transformers/main/model_doc/ernie)** (from Baidu) released with the paper [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu.
 1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
 1. **[FLAVA](https://huggingface.co/docs/transformers/model_doc/flava)** (from Facebook AI) released with the paper [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) by Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela.
 1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
--- a/docs/source/en/_toctree.yml
+++ b/docs/source/en/_toctree.yml
@@ -237,6 +237,8 @@
        title: ELECTRA
      - local: model_doc/encoder-decoder
        title: Encoder Decoder Models
+      - local: model_doc/ernie
+        title: ERNIE
      - local: model_doc/flaubert
        title: FlauBERT
      - local: model_doc/fnet
--- a/docs/source/en/index.mdx
+++ b/docs/source/en/index.mdx
@@ -87,6 +87,7 @@ The documentation is organized into five sections:
 1. **[DPT](master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.
 1. **[ELECTRA](model_doc/electra)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
 1. **[EncoderDecoder](model_doc/encoder-decoder)** (from Google Research) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+1. **[ERNIE](model_doc/ernie)** (from Baidu) released with the paper [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu.
 1. **[FlauBERT](model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
 1. **[FLAVA](model_doc/flava)** (from Facebook AI) released with the paper [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) by Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela.
 1. **[FNet](model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
@@ -230,6 +231,7 @@ Flax), PyTorch, and/or TensorFlow.
 |             DPT             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |           ELECTRA           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 |       Encoder decoder       |       ❌       |       ❌       |       ✅        |         ✅         |      ✅      |
+|            ERNIE            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 | FairSeq Machine-Translation |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
 |          FlauBERT           |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
 |            FLAVA            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
--- a/docs/source/en/model_doc/ernie.mdx
+++ b/docs/source/en/model_doc/ernie.mdx
@@ -0,0 +1,102 @@
+<!--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.
+-->
+
+# ERNIE
+
+## Overview
+ERNIE is a series of powerful models proposed by baidu, especially in Chinese tasks,
+including [ERNIE1.0](https://arxiv.org/abs/1904.09223), [ERNIE2.0](https://ojs.aaai.org/index.php/AAAI/article/view/6428),
+[ERNIE3.0](https://arxiv.org/abs/2107.02137), [ERNIE-Gram](https://arxiv.org/abs/2010.12148), [ERNIE-health](https://arxiv.org/abs/2110.07244), etc.
+
+These models are contributed by [nghuyong](https://huggingface.co/nghuyong) and the official code can be found in [PaddleNLP](https://github.com/PaddlePaddle/PaddleNLP) (in PaddlePaddle).
+
+### How to use
+Take `ernie-1.0-base-zh` as an example:
+
+```Python
+from transformers import AutoTokenizer, AutoModel
+tokenizer = AutoTokenizer.from_pretrained("nghuyong/ernie-1.0-base-zh")
+model = AutoModel.from_pretrained("nghuyong/ernie-1.0-base-zh")
+```
+
+### Supported Models
+
+|     Model Name      | Language |           Description           |
+|:-------------------:|:--------:|:-------------------------------:|
+|  ernie-1.0-base-zh  | Chinese  | Layer:12, Heads:12, Hidden:768  |
+|  ernie-2.0-base-en  | English  | Layer:12, Heads:12, Hidden:768  |
+| ernie-2.0-large-en  | English  | Layer:24, Heads:16, Hidden:1024 |
+|  ernie-3.0-base-zh  | Chinese  | Layer:12, Heads:12, Hidden:768  |
+| ernie-3.0-medium-zh | Chinese  |  Layer:6, Heads:12, Hidden:768  |
+|  ernie-3.0-mini-zh  | Chinese  |  Layer:6, Heads:12, Hidden:384  |
+| ernie-3.0-micro-zh  | Chinese  |  Layer:4, Heads:12, Hidden:384  |
+|  ernie-3.0-nano-zh  | Chinese  |  Layer:4, Heads:12, Hidden:312  |
+|   ernie-health-zh   | Chinese  | Layer:12, Heads:12, Hidden:768  |
+|    ernie-gram-zh    | Chinese  | Layer:12, Heads:12, Hidden:768  |
+
+You can find all the supported models from huggingface's model hub: [huggingface.co/nghuyong](https://huggingface.co/nghuyong), and model details from paddle's official
+repo: [PaddleNLP](https://paddlenlp.readthedocs.io/zh/latest/model_zoo/transformers/ERNIE/contents.html)
+and [ERNIE](https://github.com/PaddlePaddle/ERNIE/blob/repro).
+
+## ErnieConfig
+
+[[autodoc]] ErnieConfig
+    - all
+
+## Ernie specific outputs
+
+[[autodoc]] models.ernie.modeling_ernie.ErnieForPreTrainingOutput
+
+## ErnieModel
+
+[[autodoc]] ErnieModel
+    - forward
+
+## ErnieForPreTraining
+
+[[autodoc]] ErnieForPreTraining
+    - forward
+
+## ErnieForCausalLM
+
+[[autodoc]] ErnieForCausalLM
+    - forward
+
+## ErnieForMaskedLM
+
+[[autodoc]] ErnieForMaskedLM
+    - forward
+
+## ErnieForNextSentencePrediction
+
+[[autodoc]] ErnieForNextSentencePrediction
+    - forward
+
+## ErnieForSequenceClassification
+
+[[autodoc]] ErnieForSequenceClassification
+    - forward
+
+## ErnieForMultipleChoice
+
+[[autodoc]] ErnieForMultipleChoice
+    - forward
+
+## ErnieForTokenClassification
+
+[[autodoc]] ErnieForTokenClassification
+    - forward
+
+## ErnieForQuestionAnswering
+
+[[autodoc]] ErnieForQuestionAnswering
+    - forward
--- a/docs/source/en/serialization.mdx
+++ b/docs/source/en/serialization.mdx
@@ -67,6 +67,7 @@ Ready-made configurations include the following architectures:
 - DETR
 - DistilBERT
 - ELECTRA
+- ERNIE
 - FlauBERT
 - GPT Neo
 - GPT-J
--- a/src/transformers/init.py
+++ b/src/transformers/init.py
@@ -203,6 +203,10 @@ _import_structure = {
    "models.dpt": ["DPT_PRETRAINED_CONFIG_ARCHIVE_MAP", "DPTConfig"],
    "models.electra": ["ELECTRA_PRETRAINED_CONFIG_ARCHIVE_MAP", "ElectraConfig", "ElectraTokenizer"],
    "models.encoder_decoder": ["EncoderDecoderConfig"],
+    "models.ernie": [
+        "ERNIE_PRETRAINED_CONFIG_ARCHIVE_MAP",
+        "ErnieConfig",
+    ],
    "models.flaubert": ["FLAUBERT_PRETRAINED_CONFIG_ARCHIVE_MAP", "FlaubertConfig", "FlaubertTokenizer"],
    "models.flava": [
        "FLAVA_PRETRAINED_CONFIG_ARCHIVE_MAP",
@@ -1168,6 +1172,21 @@ else:
        ]
    )
    _import_structure["models.encoder_decoder"].append("EncoderDecoderModel")
+    _import_structure["models.ernie"].extend(
+        [
+            "ERNIE_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "ErnieForCausalLM",
+            "ErnieForMaskedLM",
+            "ErnieForMultipleChoice",
+            "ErnieForNextSentencePrediction",
+            "ErnieForPreTraining",
+            "ErnieForQuestionAnswering",
+            "ErnieForSequenceClassification",
+            "ErnieForTokenClassification",
+            "ErnieModel",
+            "ErniePreTrainedModel",
+        ]
+    )
    _import_structure["models.flaubert"].extend(
        [
            "FLAUBERT_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -3066,6 +3085,7 @@ if TYPE_CHECKING:
    from .models.dpt import DPT_PRETRAINED_CONFIG_ARCHIVE_MAP, DPTConfig
    from .models.electra import ELECTRA_PRETRAINED_CONFIG_ARCHIVE_MAP, ElectraConfig, ElectraTokenizer
    from .models.encoder_decoder import EncoderDecoderConfig
+    from .models.ernie import ERNIE_PRETRAINED_CONFIG_ARCHIVE_MAP, ErnieConfig
    from .models.flaubert import FLAUBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, FlaubertConfig, FlaubertTokenizer
    from .models.flava import (
        FLAVA_PRETRAINED_CONFIG_ARCHIVE_MAP,
@@ -3879,6 +3899,19 @@ if TYPE_CHECKING:
            load_tf_weights_in_electra,
        )
        from .models.encoder_decoder import EncoderDecoderModel
+        from .models.ernie import (
+            ERNIE_PRETRAINED_MODEL_ARCHIVE_LIST,
+            ErnieForCausalLM,
+            ErnieForMaskedLM,
+            ErnieForMultipleChoice,
+            ErnieForNextSentencePrediction,
+            ErnieForPreTraining,
+            ErnieForQuestionAnswering,
+            ErnieForSequenceClassification,
+            ErnieForTokenClassification,
+            ErnieModel,
+            ErniePreTrainedModel,
+        )
        from .models.flaubert import (
            FLAUBERT_PRETRAINED_MODEL_ARCHIVE_LIST,
            FlaubertForMultipleChoice,
--- a/src/transformers/models/init.py
+++ b/src/transformers/models/init.py
@@ -57,6 +57,7 @@ from . import (
    dpt,
    electra,
    encoder_decoder,
+    ernie,
    flaubert,
    flava,
    fnet,
--- a/src/transformers/models/auto/configuration_auto.py
+++ b/src/transformers/models/auto/configuration_auto.py
@@ -61,6 +61,7 @@ CONFIG_MAPPING_NAMES = OrderedDict(
        ("dpt", "DPTConfig"),
        ("electra", "ElectraConfig"),
        ("encoder-decoder", "EncoderDecoderConfig"),
+        ("ernie", "ErnieConfig"),
        ("flaubert", "FlaubertConfig"),
        ("flava", "FlavaConfig"),
        ("fnet", "FNetConfig"),
@@ -188,6 +189,7 @@ CONFIG_ARCHIVE_MAP_MAPPING_NAMES = OrderedDict(
        ("dpr", "DPR_PRETRAINED_CONFIG_ARCHIVE_MAP"),
        ("dpt", "DPT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
        ("electra", "ELECTRA_PRETRAINED_CONFIG_ARCHIVE_MAP"),
+        ("ernie", "ERNIE_PRETRAINED_CONFIG_ARCHIVE_MAP"),
        ("flaubert", "FLAUBERT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
        ("flava", "FLAVA_PRETRAINED_CONFIG_ARCHIVE_MAP"),
        ("fnet", "FNET_PRETRAINED_CONFIG_ARCHIVE_MAP"),
@@ -316,6 +318,7 @@ MODEL_NAMES_MAPPING = OrderedDict(
        ("dpt", "DPT"),
        ("electra", "ELECTRA"),
        ("encoder-decoder", "Encoder decoder"),
+        ("ernie", "ERNIE"),
        ("flaubert", "FlauBERT"),
        ("flava", "FLAVA"),
        ("fnet", "FNet"),
--- a/src/transformers/models/auto/modeling_auto.py
+++ b/src/transformers/models/auto/modeling_auto.py
@@ -60,6 +60,7 @@ MODEL_MAPPING_NAMES = OrderedDict(
        ("dpr", "DPRQuestionEncoder"),
        ("dpt", "DPTModel"),
        ("electra", "ElectraModel"),
+        ("ernie", "ErnieModel"),
        ("flaubert", "FlaubertModel"),
        ("flava", "FlavaModel"),
        ("fnet", "FNetModel"),
@@ -165,6 +166,7 @@ MODEL_FOR_PRETRAINING_MAPPING_NAMES = OrderedDict(
        ("deberta-v2", "DebertaV2ForMaskedLM"),
        ("distilbert", "DistilBertForMaskedLM"),
        ("electra", "ElectraForPreTraining"),
+        ("ernie", "ErnieForPreTraining"),
        ("flaubert", "FlaubertWithLMHeadModel"),
        ("flava", "FlavaForPreTraining"),
        ("fnet", "FNetForPreTraining"),
@@ -223,6 +225,7 @@ MODEL_WITH_LM_HEAD_MAPPING_NAMES = OrderedDict(
        ("distilbert", "DistilBertForMaskedLM"),
        ("electra", "ElectraForMaskedLM"),
        ("encoder-decoder", "EncoderDecoderModel"),
+        ("ernie", "ErnieForMaskedLM"),
        ("flaubert", "FlaubertWithLMHeadModel"),
        ("fnet", "FNetForMaskedLM"),
        ("fsmt", "FSMTForConditionalGeneration"),
@@ -284,6 +287,7 @@ MODEL_FOR_CAUSAL_LM_MAPPING_NAMES = OrderedDict(
        ("ctrl", "CTRLLMHeadModel"),
        ("data2vec-text", "Data2VecTextForCausalLM"),
        ("electra", "ElectraForCausalLM"),
+        ("ernie", "ErnieForCausalLM"),
        ("gpt2", "GPT2LMHeadModel"),
        ("gpt_neo", "GPTNeoForCausalLM"),
        ("gpt_neox", "GPTNeoXForCausalLM"),
@@ -413,6 +417,7 @@ MODEL_FOR_MASKED_LM_MAPPING_NAMES = OrderedDict(
        ("deberta-v2", "DebertaV2ForMaskedLM"),
        ("distilbert", "DistilBertForMaskedLM"),
        ("electra", "ElectraForMaskedLM"),
+        ("ernie", "ErnieForMaskedLM"),
        ("flaubert", "FlaubertWithLMHeadModel"),
        ("fnet", "FNetForMaskedLM"),
        ("funnel", "FunnelForMaskedLM"),
@@ -502,6 +507,7 @@ MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES = OrderedDict(
        ("deberta-v2", "DebertaV2ForSequenceClassification"),
        ("distilbert", "DistilBertForSequenceClassification"),
        ("electra", "ElectraForSequenceClassification"),
+        ("ernie", "ErnieForSequenceClassification"),
        ("flaubert", "FlaubertForSequenceClassification"),
        ("fnet", "FNetForSequenceClassification"),
        ("funnel", "FunnelForSequenceClassification"),
@@ -558,6 +564,7 @@ MODEL_FOR_QUESTION_ANSWERING_MAPPING_NAMES = OrderedDict(
        ("deberta-v2", "DebertaV2ForQuestionAnswering"),
        ("distilbert", "DistilBertForQuestionAnswering"),
        ("electra", "ElectraForQuestionAnswering"),
+        ("ernie", "ErnieForQuestionAnswering"),
        ("flaubert", "FlaubertForQuestionAnsweringSimple"),
        ("fnet", "FNetForQuestionAnswering"),
        ("funnel", "FunnelForQuestionAnswering"),
@@ -627,6 +634,7 @@ MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING_NAMES = OrderedDict(
        ("deberta-v2", "DebertaV2ForTokenClassification"),
        ("distilbert", "DistilBertForTokenClassification"),
        ("electra", "ElectraForTokenClassification"),
+        ("ernie", "ErnieForTokenClassification"),
        ("flaubert", "FlaubertForTokenClassification"),
        ("fnet", "FNetForTokenClassification"),
        ("funnel", "FunnelForTokenClassification"),
@@ -668,6 +676,7 @@ MODEL_FOR_MULTIPLE_CHOICE_MAPPING_NAMES = OrderedDict(
        ("deberta-v2", "DebertaV2ForMultipleChoice"),
        ("distilbert", "DistilBertForMultipleChoice"),
        ("electra", "ElectraForMultipleChoice"),
+        ("ernie", "ErnieForMultipleChoice"),
        ("flaubert", "FlaubertForMultipleChoice"),
        ("fnet", "FNetForMultipleChoice"),
        ("funnel", "FunnelForMultipleChoice"),
@@ -695,6 +704,7 @@ MODEL_FOR_MULTIPLE_CHOICE_MAPPING_NAMES = OrderedDict(
 MODEL_FOR_NEXT_SENTENCE_PREDICTION_MAPPING_NAMES = OrderedDict(
    [
        ("bert", "BertForNextSentencePrediction"),
+        ("ernie", "ErnieForNextSentencePrediction"),
        ("fnet", "FNetForNextSentencePrediction"),
        ("megatron-bert", "MegatronBertForNextSentencePrediction"),
        ("mobilebert", "MobileBertForNextSentencePrediction"),
--- a/src/transformers/models/auto/tokenization_auto.py
+++ b/src/transformers/models/auto/tokenization_auto.py
@@ -121,6 +121,7 @@ else:
                ),
            ),
            ("electra", ("ElectraTokenizer", "ElectraTokenizerFast" if is_tokenizers_available() else None)),
+            ("ernie", ("BertTokenizer", "BertTokenizerFast" if is_tokenizers_available() else None)),
            ("flaubert", ("FlaubertTokenizer", None)),
            ("fnet", ("FNetTokenizer", "FNetTokenizerFast" if is_tokenizers_available() else None)),
            ("fsmt", ("FSMTTokenizer", None)),
--- a/src/transformers/models/ernie/init.py
+++ b/src/transformers/models/ernie/init.py
@@ -0,0 +1,74 @@
+# flake8: noqa
+# There's no way to ignore "F401 '...' imported but unused" warnings in this
+# module, but to preserve other warnings. So, don't check this module at all.
+
+# Copyright 2022 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import TYPE_CHECKING
+
+from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_tensorflow_text_available, is_torch_available
+
+
+_import_structure = {
+    "configuration_ernie": ["ERNIE_PRETRAINED_CONFIG_ARCHIVE_MAP", "ErnieConfig", "ErnieOnnxConfig"],
+}
+
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["modeling_ernie"] = [
+        "ERNIE_PRETRAINED_MODEL_ARCHIVE_LIST",
+        "ErnieForCausalLM",
+        "ErnieForMaskedLM",
+        "ErnieForMultipleChoice",
+        "ErnieForNextSentencePrediction",
+        "ErnieForPreTraining",
+        "ErnieForQuestionAnswering",
+        "ErnieForSequenceClassification",
+        "ErnieForTokenClassification",
+        "ErnieModel",
+        "ErniePreTrainedModel",
+    ]
+
+if TYPE_CHECKING:
+    from .configuration_ernie import ERNIE_PRETRAINED_CONFIG_ARCHIVE_MAP, ErnieConfig, ErnieOnnxConfig
+
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .modeling_ernie import (
+            ERNIE_PRETRAINED_MODEL_ARCHIVE_LIST,
+            ErnieForCausalLM,
+            ErnieForMaskedLM,
+            ErnieForMultipleChoice,
+            ErnieForNextSentencePrediction,
+            ErnieForPreTraining,
+            ErnieForQuestionAnswering,
+            ErnieForSequenceClassification,
+            ErnieForTokenClassification,
+            ErnieModel,
+            ErniePreTrainedModel,
+        )
+
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)
--- a/src/transformers/models/ernie/configuration_ernie.py
+++ b/src/transformers/models/ernie/configuration_ernie.py
@@ -0,0 +1,169 @@
+# coding=utf-8
+# Copyright 2022 The Google AI Language Team Authors and The HuggingFace Inc. team.
+# Copyright (c) 2018, NVIDIA CORPORATION.  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.
+""" ERNIE model configuration"""
+from collections import OrderedDict
+from typing import Mapping
+
+from ...configuration_utils import PretrainedConfig
+from ...onnx import OnnxConfig
+from ...utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+ERNIE_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "nghuyong/ernie-1.0-base-zh": "https://huggingface.co/nghuyong/ernie-1.0-base-zh/resolve/main/config.json",
+    "nghuyong/ernie-2.0-base-en": "https://huggingface.co/nghuyong/ernie-2.0-base-en/resolve/main/config.json",
+    "nghuyong/ernie-2.0-large-en": "https://huggingface.co/nghuyong/ernie-2.0-large-en/resolve/main/config.json",
+    "nghuyong/ernie-3.0-base-zh": "https://huggingface.co/nghuyong/ernie-3.0-base-zh/resolve/main/config.json",
+    "nghuyong/ernie-3.0-medium-zh": "https://huggingface.co/nghuyong/ernie-3.0-medium-zh/resolve/main/config.json",
+    "nghuyong/ernie-3.0-mini-zh": "https://huggingface.co/nghuyong/ernie-3.0-mini-zh/resolve/main/config.json",
+    "nghuyong/ernie-3.0-micro-zh": "https://huggingface.co/nghuyong/ernie-3.0-micro-zh/resolve/main/config.json",
+    "nghuyong/ernie-3.0-nano-zh": "https://huggingface.co/nghuyong/ernie-3.0-nano-zh/resolve/main/config.json",
+    "nghuyong/ernie-gram-zh": "https://huggingface.co/nghuyong/ernie-gram-zh/resolve/main/config.json",
+    "nghuyong/ernie-health-zh": "https://huggingface.co/nghuyong/ernie-health-zh/resolve/main/config.json",
+}
+
+
+class ErnieConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`ErnieModel`] or a [`TFErnieModel`]. It is used to
+    instantiate a ERNIE 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 ERNIE
+    [nghuyong/ernie-3.0-base-zh](https://huggingface.co/nghuyong/ernie-3.0-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 ERNIE model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`ErnieModel`] or [`TFErnieModel`].
+        hidden_size (`int`, *optional*, defaults to 768):
+            Dimensionality of the encoder layers and the pooler layer.
+        num_hidden_layers (`int`, *optional*, defaults to 12):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 12):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        intermediate_size (`int`, *optional*, defaults to 3072):
+            Dimensionality of the "intermediate" (often named feed-forward) layer in the Transformer encoder.
+        hidden_act (`str` or `Callable`, *optional*, defaults to `"gelu"`):
+            The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
+            `"relu"`, `"silu"` and `"gelu_new"` are supported.
+        hidden_dropout_prob (`float`, *optional*, defaults to 0.1):
+            The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        attention_probs_dropout_prob (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for the attention probabilities.
+        max_position_embeddings (`int`, *optional*, defaults to 512):
+            The maximum sequence length that this model might ever be used with. Typically set this to something large
+            just in case (e.g., 512 or 1024 or 2048).
+        type_vocab_size (`int`, *optional*, defaults to 2):
+            The vocabulary size of the `token_type_ids` passed when calling [`ErnieModel`] or [`TFErnieModel`].
+        task_type_vocab_size (`int`, *optional*, defaults to 3):
+            The vocabulary size of the `task_type_ids` for ERNIE2.0/ERNIE3.0 model
+        use_task_id (`bool`, *optional*, defaults to `False`):
+            Whether or not the model support `task_type_ids`
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        layer_norm_eps (`float`, *optional*, defaults to 1e-12):
+            The epsilon used by the layer normalization layers.
+        position_embedding_type (`str`, *optional*, defaults to `"absolute"`):
+            Type of position embedding. Choose one of `"absolute"`, `"relative_key"`, `"relative_key_query"`. For
+            positional embeddings use `"absolute"`. For more information on `"relative_key"`, please refer to
+            [Self-Attention with Relative Position Representations (Shaw et al.)](https://arxiv.org/abs/1803.02155).
+            For more information on `"relative_key_query"`, please refer to *Method 4* in [Improve Transformer Models
+            with Better Relative Position Embeddings (Huang et al.)](https://arxiv.org/abs/2009.13658).
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`.
+        classifier_dropout (`float`, *optional*):
+            The dropout ratio for the classification head.
+
+    Examples:
+
+    ```python
+    >>> from transformers import ErnieModel, ErnieConfig
+
+    >>> # Initializing a ERNIE nghuyong/ernie-3.0-base-zh style configuration
+    >>> configuration = ErnieConfig()
+
+    >>> # Initializing a model from the nghuyong/ernie-3.0-base-zh style configuration
+    >>> model = ErnieModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+    model_type = "ernie"
+
+    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,
+        task_type_vocab_size=3,
+        use_task_id=False,
+        initializer_range=0.02,
+        layer_norm_eps=1e-12,
+        pad_token_id=0,
+        position_embedding_type="absolute",
+        use_cache=True,
+        classifier_dropout=None,
+        **kwargs
+    ):
+        super().__init__(pad_token_id=pad_token_id, **kwargs)
+
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.hidden_act = hidden_act
+        self.intermediate_size = intermediate_size
+        self.hidden_dropout_prob = hidden_dropout_prob
+        self.attention_probs_dropout_prob = attention_probs_dropout_prob
+        self.max_position_embeddings = max_position_embeddings
+        self.type_vocab_size = type_vocab_size
+        self.task_type_vocab_size = task_type_vocab_size
+        self.use_task_id = use_task_id
+        self.initializer_range = initializer_range
+        self.layer_norm_eps = layer_norm_eps
+        self.position_embedding_type = position_embedding_type
+        self.use_cache = use_cache
+        self.classifier_dropout = classifier_dropout
+
+
+class ErnieOnnxConfig(OnnxConfig):
+    @property
+    def inputs(self) -> Mapping[str, Mapping[int, str]]:
+        if self.task == "multiple-choice":
+            dynamic_axis = {0: "batch", 1: "choice", 2: "sequence"}
+        else:
+            dynamic_axis = {0: "batch", 1: "sequence"}
+        return OrderedDict(
+            [
+                ("input_ids", dynamic_axis),
+                ("attention_mask", dynamic_axis),
+                ("token_type_ids", dynamic_axis),
+                ("task_type_ids", dynamic_axis),
+            ]
+        )
--- a/src/transformers/models/ernie/modeling_ernie.py
+++ b/src/transformers/models/ernie/modeling_ernie.py
--- a/src/transformers/utils/dummy_pt_objects.py
+++ b/src/transformers/utils/dummy_pt_objects.py
@@ -1875,6 +1875,79 @@ class EncoderDecoderModel(metaclass=DummyObject):
        requires_backends(self, ["torch"])


+ERNIE_PRETRAINED_MODEL_ARCHIVE_LIST = None
+
+
+class ErnieForCausalLM(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class ErnieForMaskedLM(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class ErnieForMultipleChoice(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class ErnieForNextSentencePrediction(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class ErnieForPreTraining(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class ErnieForQuestionAnswering(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class ErnieForSequenceClassification(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class ErnieForTokenClassification(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class ErnieModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class ErniePreTrainedModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
 FLAUBERT_PRETRAINED_MODEL_ARCHIVE_LIST = None


--- a/tests/models/ernie/init.py
+++ b/tests/models/ernie/init.py
--- a/tests/models/ernie/test_modeling_ernie.py
+++ b/tests/models/ernie/test_modeling_ernie.py
@@ -0,0 +1,577 @@
+# 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 os
+import tempfile
+import unittest
+
+from transformers import ErnieConfig, is_torch_available
+from transformers.models.auto import get_values
+from transformers.testing_utils import require_torch, require_torch_gpu, slow, torch_device
+
+from ...generation.test_generation_utils import GenerationTesterMixin
+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,
+        ErnieForCausalLM,
+        ErnieForMaskedLM,
+        ErnieForMultipleChoice,
+        ErnieForNextSentencePrediction,
+        ErnieForPreTraining,
+        ErnieForQuestionAnswering,
+        ErnieForSequenceClassification,
+        ErnieForTokenClassification,
+        ErnieModel,
+    )
+    from transformers.models.ernie.modeling_ernie import ERNIE_PRETRAINED_MODEL_ARCHIVE_LIST
+
+
+class ErnieModelTester:
+    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,
+        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.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_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, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+
+    def get_config(self):
+        """
+        Returns a tiny configuration by default.
+        """
+        return ErnieConfig(
+            vocab_size=self.vocab_size,
+            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,
+            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,
+            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, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = ErnieModel(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids)
+        result = model(input_ids, token_type_ids=token_type_ids)
+        result = model(input_ids)
+        self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
+        self.parent.assertEqual(result.pooler_output.shape, (self.batch_size, self.hidden_size))
+
+    def create_and_check_model_as_decoder(
+        self,
+        config,
+        input_ids,
+        token_type_ids,
+        input_mask,
+        sequence_labels,
+        token_labels,
+        choice_labels,
+        encoder_hidden_states,
+        encoder_attention_mask,
+    ):
+        config.add_cross_attention = True
+        model = ErnieModel(config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_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,
+            attention_mask=input_mask,
+            token_type_ids=token_type_ids,
+            encoder_hidden_states=encoder_hidden_states,
+        )
+        result = model(input_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))
+        self.parent.assertEqual(result.pooler_output.shape, (self.batch_size, self.hidden_size))
+
+    def create_and_check_for_causal_lm(
+        self,
+        config,
+        input_ids,
+        token_type_ids,
+        input_mask,
+        sequence_labels,
+        token_labels,
+        choice_labels,
+        encoder_hidden_states,
+        encoder_attention_mask,
+    ):
+        model = ErnieForCausalLM(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=token_labels)
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
+
+    def create_and_check_for_masked_lm(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = ErnieForMaskedLM(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=token_labels)
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
+
+    def create_and_check_model_for_causal_lm_as_decoder(
+        self,
+        config,
+        input_ids,
+        token_type_ids,
+        input_mask,
+        sequence_labels,
+        token_labels,
+        choice_labels,
+        encoder_hidden_states,
+        encoder_attention_mask,
+    ):
+        config.add_cross_attention = True
+        model = ErnieForCausalLM(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_ids,
+            attention_mask=input_mask,
+            token_type_ids=token_type_ids,
+            labels=token_labels,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+        )
+        result = model(
+            input_ids,
+            attention_mask=input_mask,
+            token_type_ids=token_type_ids,
+            labels=token_labels,
+            encoder_hidden_states=encoder_hidden_states,
+        )
+        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,
+        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 = ErnieForCausalLM(config=config).to(torch_device).eval()
+
+        # first forward pass
+        outputs = model(
+            input_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_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_attention_mask = torch.cat([input_mask, next_mask], dim=-1)
+
+        output_from_no_past = model(
+            next_input_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,
+            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_next_sequence_prediction(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = ErnieForNextSentencePrediction(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_ids,
+            attention_mask=input_mask,
+            token_type_ids=token_type_ids,
+            labels=sequence_labels,
+        )
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, 2))
+
+    def create_and_check_for_pretraining(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = ErnieForPreTraining(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_ids,
+            attention_mask=input_mask,
+            token_type_ids=token_type_ids,
+            labels=token_labels,
+            next_sentence_label=sequence_labels,
+        )
+        self.parent.assertEqual(result.prediction_logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
+        self.parent.assertEqual(result.seq_relationship_logits.shape, (self.batch_size, 2))
+
+    def create_and_check_for_question_answering(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = ErnieForQuestionAnswering(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_ids,
+            attention_mask=input_mask,
+            token_type_ids=token_type_ids,
+            start_positions=sequence_labels,
+            end_positions=sequence_labels,
+        )
+        self.parent.assertEqual(result.start_logits.shape, (self.batch_size, self.seq_length))
+        self.parent.assertEqual(result.end_logits.shape, (self.batch_size, self.seq_length))
+
+    def create_and_check_for_sequence_classification(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        config.num_labels = self.num_labels
+        model = ErnieForSequenceClassification(config)
+        model.to(torch_device)
+        model.eval()
+        result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=sequence_labels)
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels))
+
+    def create_and_check_for_token_classification(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        config.num_labels = self.num_labels
+        model = ErnieForTokenClassification(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=token_labels)
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.num_labels))
+
+    def create_and_check_for_multiple_choice(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        config.num_choices = self.num_choices
+        model = ErnieForMultipleChoice(config=config)
+        model.to(torch_device)
+        model.eval()
+        multiple_choice_inputs_ids = input_ids.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
+        multiple_choice_token_type_ids = token_type_ids.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
+        multiple_choice_input_mask = input_mask.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
+        result = model(
+            multiple_choice_inputs_ids,
+            attention_mask=multiple_choice_input_mask,
+            token_type_ids=multiple_choice_token_type_ids,
+            labels=choice_labels,
+        )
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_choices))
+
+    def prepare_config_and_inputs_for_common(self):
+        config_and_inputs = self.prepare_config_and_inputs()
+        (
+            config,
+            input_ids,
+            token_type_ids,
+            input_mask,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+        ) = config_and_inputs
+        inputs_dict = {"input_ids": input_ids, "token_type_ids": token_type_ids, "attention_mask": input_mask}
+        return config, inputs_dict
+
+
+@require_torch
+class ErnieModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase):
+    all_model_classes = (
+        (
+            ErnieModel,
+            ErnieForCausalLM,
+            ErnieForMaskedLM,
+            ErnieForMultipleChoice,
+            ErnieForNextSentencePrediction,
+            ErnieForPreTraining,
+            ErnieForQuestionAnswering,
+            ErnieForSequenceClassification,
+            ErnieForTokenClassification,
+        )
+        if is_torch_available()
+        else ()
+    )
+    all_generative_model_classes = (ErnieForCausalLM,) if is_torch_available() else ()
+    fx_compatible = False
+
+    # 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"] = torch.zeros(
+                    (self.model_tester.batch_size, self.model_tester.seq_length), dtype=torch.long, device=torch_device
+                )
+                inputs_dict["next_sentence_label"] = torch.zeros(
+                    self.model_tester.batch_size, dtype=torch.long, device=torch_device
+                )
+        return inputs_dict
+
+    def setUp(self):
+        self.model_tester = ErnieModelTester(self)
+        self.config_tester = ConfigTester(self, config_class=ErnieConfig, 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_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,
+            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,
+            token_type_ids,
+            input_mask,
+            sequence_labels,
+            token_labels,
+            choice_labels,
+            encoder_hidden_states,
+            encoder_attention_mask,
+        )
+
+    def test_for_causal_lm(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs_for_decoder()
+        self.model_tester.create_and_check_for_causal_lm(*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_causal_lm_decoder(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs_for_decoder()
+        self.model_tester.create_and_check_model_for_causal_lm_as_decoder(*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_multiple_choice(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_for_multiple_choice(*config_and_inputs)
+
+    def test_for_next_sequence_prediction(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_for_next_sequence_prediction(*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_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)
+
+    @slow
+    def test_model_from_pretrained(self):
+        for model_name in ERNIE_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
+            model = ErnieModel.from_pretrained(model_name)
+            self.assertIsNotNone(model)
+
+    @slow
+    @require_torch_gpu
+    def test_torchscript_device_change(self):
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+        for model_class in self.all_model_classes:
+            # ErnieForMultipleChoice behaves incorrectly in JIT environments.
+            if model_class == ErnieForMultipleChoice:
+                return
+
+            config.torchscript = True
+            model = model_class(config=config)
+
+            inputs_dict = self._prepare_for_class(inputs_dict, model_class)
+            traced_model = torch.jit.trace(
+                model, (inputs_dict["input_ids"].to("cpu"), inputs_dict["attention_mask"].to("cpu"))
+            )
+
+            with tempfile.TemporaryDirectory() as tmp:
+                torch.jit.save(traced_model, os.path.join(tmp, "ernie.pt"))
+                loaded = torch.jit.load(os.path.join(tmp, "ernie.pt"), map_location=torch_device)
+                loaded(inputs_dict["input_ids"].to(torch_device), inputs_dict["attention_mask"].to(torch_device))