Handle rebasing

.github/workflows/self-push.yml

@@ -4,10 +4,12 @@ on:
   push:
     branches:
       - master
+      - model-templates
     paths:
       - "src/**"
       - "tests/**"
       - ".github/**"
+      - "templates/**"
   # pull_request:
   repository_dispatch:
 
@@ -55,6 +57,15 @@ jobs:
           python -c "import torch; print('Cuda available:', torch.cuda.is_available())"
           python -c "import torch; print('Number of GPUs available:', torch.cuda.device_count())"
 
+      - name: Create model files
+        run: |
+          source .env/bin/activate
+          transformers-cli add-new-model --testing --testing_file=templates/cookiecutter/tests/encoder-bert-tokenizer.json --path=templates/cookiecutter
+          transformers-cli add-new-model --testing --testing_file=templates/cookiecutter/tests/pt-encoder-bert-tokenizer.json --path=templates/cookiecutter
+          transformers-cli add-new-model --testing --testing_file=templates/cookiecutter/tests/seq-to-seq-roberta-tokenizer.json --path=templates/cookiecutter
+          transformers-cli add-new-model --testing --testing_file=templates/cookiecutter/tests/standalone.json --path=templates/cookiecutter
+          transformers-cli add-new-model --testing --testing_file=templates/cookiecutter/tests/tf-encoder-bert-tokenizer.json --path=templates/cookiecutter
+
       - name: Run all non-slow tests on GPU
         env:
           OMP_NUM_THREADS: 1
@@ -104,6 +115,15 @@ jobs:
           TF_CPP_MIN_LOG_LEVEL=3 python -c "import tensorflow as tf; print('TF GPUs available:', bool(tf.config.list_physical_devices('GPU')))"
           TF_CPP_MIN_LOG_LEVEL=3 python -c "import tensorflow as tf; print('Number of TF GPUs available:', len(tf.config.list_physical_devices('GPU')))"
 
+      - name: Create model files
+        run: |
+          source .env/bin/activate
+          transformers-cli add-new-model --testing --testing_file=templates/cookiecutter/tests/encoder-bert-tokenizer.json --path=templates/cookiecutter
+          transformers-cli add-new-model --testing --testing_file=templates/cookiecutter/tests/pt-encoder-bert-tokenizer.json --path=templates/cookiecutter
+          transformers-cli add-new-model --testing --testing_file=templates/cookiecutter/tests/seq-to-seq-roberta-tokenizer.json --path=templates/cookiecutter
+          transformers-cli add-new-model --testing --testing_file=templates/cookiecutter/tests/standalone.json --path=templates/cookiecutter
+          transformers-cli add-new-model --testing --testing_file=templates/cookiecutter/tests/tf-encoder-bert-tokenizer.json --path=templates/cookiecutter
+
       - name: Run all non-slow tests on GPU
         env:
           OMP_NUM_THREADS: 1

setup.py

@@ -98,12 +98,13 @@ else:
 
 extras["tokenizers"] = ["tokenizers==0.9.2"]
 extras["onnxruntime"] = ["onnxruntime>=1.4.0", "onnxruntime-tools>=1.4.2"]
+extras["modelcreation"] = ["cookiecutter==1.7.2"]
 
 extras["serving"] = ["pydantic", "uvicorn", "fastapi", "starlette"]
 
 extras["sentencepiece"] = ["sentencepiece==0.1.91"]
 extras["retrieval"] = ["faiss-cpu", "datasets"]
-extras["testing"] = ["pytest", "pytest-xdist", "timeout-decorator", "parameterized", "psutil"] + extras["retrieval"]
+extras["testing"] = ["pytest", "pytest-xdist", "timeout-decorator", "parameterized", "psutil"] + extras["retrieval"] + extras["modelcreation"]
 # sphinx-rtd-theme==0.5.0 introduced big changes in the style.
 extras["docs"] = ["recommonmark", "sphinx", "sphinx-markdown-tables", "sphinx-rtd-theme==0.4.3", "sphinx-copybutton"]
 extras["quality"] = ["black >= 20.8b1", "isort >= 5.5.4", "flake8 >= 3.8.3"]
@@ -111,7 +112,7 @@ extras["quality"] = ["black >= 20.8b1", "isort >= 5.5.4", "flake8 >= 3.8.3"]
 
 extras["all"] = extras["tf"] + extras["torch"] + extras["flax"] + extras["sentencepiece"] + extras["tokenizers"]
 
-extras["dev"] = extras["all"] + extras["testing"] + extras["quality"] + extras["ja"] + extras["docs"] + extras["sklearn"]
+extras["dev"] = extras["all"] + extras["testing"] + extras["quality"] + extras["ja"] + extras["docs"] + extras["sklearn"] + extras["modelcreation"]
 
 
 setup(

src/transformers/commands/add_new_model.py

@@ -0,0 +1,192 @@
+import json
+import os
+import shutil
+from argparse import ArgumentParser, Namespace
+from pathlib import Path
+from typing import List
+
+from cookiecutter.main import cookiecutter
+from transformers.commands import BaseTransformersCLICommand
+
+from ..utils import logging
+
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+def add_new_model_command_factory(args: Namespace):
+    return AddNewModelCommand(args.testing, args.testing_file, path=args.path)
+
+
+class AddNewModelCommand(BaseTransformersCLICommand):
+    @staticmethod
+    def register_subcommand(parser: ArgumentParser):
+        add_new_model_parser = parser.add_parser("add-new-model")
+        add_new_model_parser.add_argument("--testing", action="store_true", help="If in testing mode.")
+        add_new_model_parser.add_argument("--testing_file", type=str, help="Configuration file on which to run.")
+        add_new_model_parser.add_argument(
+            "--path", type=str, help="Path to cookiecutter. Should only be used for testing purposes."
+        )
+        add_new_model_parser.set_defaults(func=add_new_model_command_factory)
+
+    def __init__(self, testing: bool, testing_file: str, path=None, *args):
+        self._testing = testing
+        self._testing_file = testing_file
+        self._path = path
+
+    def run(self):
+        # Ensure that there is no other `cookiecutter-template-xxx` directory in the current working directory
+        directories = [directory for directory in os.listdir() if "cookiecutter-template-" in directory[:22]]
+        if len(directories) > 0:
+            raise ValueError(
+                "Several directories starting with `cookiecutter-template-` in current working directory. "
+                "Please clean your directory by removing all folders startign with `cookiecutter-template-` or "
+                "change your working directory."
+            )
+
+        path_to_transformer_root = (
+            Path(__file__).parent.parent.parent.parent if self._path is None else Path(self._path).parent.parent
+        )
+        path_to_cookiecutter = path_to_transformer_root / "templates" / "cookiecutter"
+
+        # Execute cookiecutter
+        if not self._testing:
+            cookiecutter(str(path_to_cookiecutter))
+        else:
+            with open(self._testing_file, "r") as configuration_file:
+                testing_configuration = json.load(configuration_file)
+
+            cookiecutter(
+                str(path_to_cookiecutter if self._path is None else self._path),
+                no_input=True,
+                extra_context=testing_configuration,
+            )
+
+        directory = [directory for directory in os.listdir() if "cookiecutter-template-" in directory[:22]][0]
+
+        # Retrieve configuration
+        with open(directory + "/configuration.json", "r") as configuration_file:
+            configuration = json.load(configuration_file)
+
+        lowercase_model_name = configuration["lowercase_modelname"]
+        pytorch_or_tensorflow = configuration["generate_tensorflow_and_pytorch"]
+        os.remove(f"{directory}/configuration.json")
+
+        output_pytorch = "PyTorch" in pytorch_or_tensorflow
+        output_tensorflow = "TensorFlow" in pytorch_or_tensorflow
+
+        shutil.move(
+            f"{directory}/configuration_{lowercase_model_name}.py",
+            f"{path_to_transformer_root}/src/transformers/configuration_{lowercase_model_name}.py",
+        )
+
+        def remove_copy_lines(path):
+            with open(path, "r") as f:
+                lines = f.readlines()
+            with open(path, "w") as f:
+                for line in lines:
+                    if "# Copied from transformers." not in line:
+                        f.write(line)
+
+        if output_pytorch:
+            if not self._testing:
+                remove_copy_lines(f"{directory}/modeling_{lowercase_model_name}.py")
+
+            shutil.move(
+                f"{directory}/modeling_{lowercase_model_name}.py",
+                f"{path_to_transformer_root}/src/transformers/modeling_{lowercase_model_name}.py",
+            )
+
+            shutil.move(
+                f"{directory}/test_modeling_{lowercase_model_name}.py",
+                f"{path_to_transformer_root}/tests/test_modeling_{lowercase_model_name}.py",
+            )
+        else:
+            os.remove(f"{directory}/modeling_{lowercase_model_name}.py")
+            os.remove(f"{directory}/test_modeling_{lowercase_model_name}.py")
+
+        if output_tensorflow:
+            if not self._testing:
+                remove_copy_lines(f"{directory}/modeling_tf_{lowercase_model_name}.py")
+
+            shutil.move(
+                f"{directory}/modeling_tf_{lowercase_model_name}.py",
+                f"{path_to_transformer_root}/src/transformers/modeling_tf_{lowercase_model_name}.py",
+            )
+
+            shutil.move(
+                f"{directory}/test_modeling_tf_{lowercase_model_name}.py",
+                f"{path_to_transformer_root}/tests/test_modeling_tf_{lowercase_model_name}.py",
+            )
+        else:
+            os.remove(f"{directory}/modeling_tf_{lowercase_model_name}.py")
+            os.remove(f"{directory}/test_modeling_tf_{lowercase_model_name}.py")
+
+        shutil.move(
+            f"{directory}/{lowercase_model_name}.rst",
+            f"{path_to_transformer_root}/docs/source/model_doc/{lowercase_model_name}.rst",
+        )
+
+        shutil.move(
+            f"{directory}/tokenization_{lowercase_model_name}.py",
+            f"{path_to_transformer_root}/src/transformers/tokenization_{lowercase_model_name}.py",
+        )
+
+        from os import fdopen, remove
+        from shutil import copymode, move
+        from tempfile import mkstemp
+
+        def replace(original_file: str, line_to_copy_below: str, lines_to_copy: List[str]):
+            # Create temp file
+            fh, abs_path = mkstemp()
+            line_found = False
+            with fdopen(fh, "w") as new_file:
+                with open(original_file) as old_file:
+                    for line in old_file:
+                        new_file.write(line)
+                        if line_to_copy_below in line:
+                            line_found = True
+                            for line_to_copy in lines_to_copy:
+                                new_file.write(line_to_copy)
+
+            if not line_found:
+                raise ValueError(f"Line {line_to_copy_below} was not found in file.")
+
+            # Copy the file permissions from the old file to the new file
+            copymode(original_file, abs_path)
+            # Remove original file
+            remove(original_file)
+            # Move new file
+            move(abs_path, original_file)
+
+        def skip_units(line):
+            return ("generating PyTorch" in line and not output_pytorch) or (
+                "generating TensorFlow" in line and not output_tensorflow
+            )
+
+        def replace_in_files(path_to_datafile):
+            with open(path_to_datafile) as datafile:
+                lines_to_copy = []
+                skip_file = False
+                skip_snippet = False
+                for line in datafile:
+                    if "# To replace in: " in line and "##" not in line:
+                        file_to_replace_in = line.split('"')[1]
+                        skip_file = skip_units(line)
+                    elif "# Below: " in line and "##" not in line:
+                        line_to_copy_below = line.split('"')[1]
+                        skip_snippet = skip_units(line)
+                    elif "# End." in line and "##" not in line:
+                        if not skip_file and not skip_snippet:
+                            replace(file_to_replace_in, line_to_copy_below, lines_to_copy)
+
+                        lines_to_copy = []
+                    elif "# Replace with" in line and "##" not in line:
+                        lines_to_copy = []
+                    elif "##" not in line:
+                        lines_to_copy.append(line)
+
+            remove(path_to_datafile)
+
+        replace_in_files(f"{directory}/to_replace_{lowercase_model_name}.py")
+        os.rmdir(directory)

src/transformers/commands/transformers_cli.py

@@ -1,6 +1,7 @@
 #!/usr/bin/env python
 from argparse import ArgumentParser
 
+from transformers.commands.add_new_model import AddNewModelCommand
 from transformers.commands.convert import ConvertCommand
 from transformers.commands.download import DownloadCommand
 from transformers.commands.env import EnvironmentCommand
@@ -20,6 +21,7 @@ def main():
     RunCommand.register_subcommand(commands_parser)
     ServeCommand.register_subcommand(commands_parser)
     UserCommands.register_subcommand(commands_parser)
+    AddNewModelCommand.register_subcommand(commands_parser)
 
     # Let's go
     args = parser.parse_args()

src/transformers/configuration_auto.py

@@ -59,6 +59,7 @@ from .configuration_xlnet import XLNET_PRETRAINED_CONFIG_ARCHIVE_MAP, XLNetConfi
 ALL_PRETRAINED_CONFIG_ARCHIVE_MAP = dict(
     (key, value)
     for pretrained_map in [
+        # Add archive maps here
         BERT_PRETRAINED_CONFIG_ARCHIVE_MAP,
         BART_PRETRAINED_CONFIG_ARCHIVE_MAP,
         BLENDERBOT_PRETRAINED_CONFIG_ARCHIVE_MAP,
@@ -95,6 +96,7 @@ ALL_PRETRAINED_CONFIG_ARCHIVE_MAP = dict(
 
 CONFIG_MAPPING = OrderedDict(
     [
+        # Add configs here
         ("retribert", RetriBertConfig),
         ("t5", T5Config),
         ("mobilebert", MobileBertConfig),
@@ -136,6 +138,7 @@ CONFIG_MAPPING = OrderedDict(
 
 MODEL_NAMES_MAPPING = OrderedDict(
     [
+        # Add cased model names here
         ("retribert", "RetriBERT"),
         ("t5", "T5"),
         ("mobilebert", "MobileBERT"),

src/transformers/configuration_bert.py

@@ -20,6 +20,7 @@ from .utils import logging
 
 
 logger = logging.get_logger(__name__)
+logger.setLevel(logging.INFO)
 
 BERT_PRETRAINED_CONFIG_ARCHIVE_MAP = {
     "bert-base-uncased": "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-uncased-config.json",

src/transformers/modeling_auto.py

@@ -222,11 +222,14 @@ from .modeling_xlnet import (
 from .utils import logging
 
 
+# Add modeling imports here
+
 logger = logging.get_logger(__name__)
 
 
 MODEL_MAPPING = OrderedDict(
     [
+        # Base model mapping
         (RetriBertConfig, RetriBertModel),
         (T5Config, T5Model),
         (DistilBertConfig, DistilBertModel),
@@ -262,6 +265,7 @@ MODEL_MAPPING = OrderedDict(
 
 MODEL_FOR_PRETRAINING_MAPPING = OrderedDict(
     [
+        # Model for pre-training mapping
         (LayoutLMConfig, LayoutLMForMaskedLM),
         (RetriBertConfig, RetriBertModel),
         (T5Config, T5ForConditionalGeneration),
@@ -290,6 +294,7 @@ MODEL_FOR_PRETRAINING_MAPPING = OrderedDict(
 
 MODEL_WITH_LM_HEAD_MAPPING = OrderedDict(
     [
+        # Model with LM heads mapping
         (LayoutLMConfig, LayoutLMForMaskedLM),
         (T5Config, T5ForConditionalGeneration),
         (DistilBertConfig, DistilBertForMaskedLM),
@@ -320,6 +325,7 @@ MODEL_WITH_LM_HEAD_MAPPING = OrderedDict(
 
 MODEL_FOR_CAUSAL_LM_MAPPING = OrderedDict(
     [
+        # Model for Causal LM mapping
         (CamembertConfig, CamembertForCausalLM),
         (XLMRobertaConfig, XLMRobertaForCausalLM),
         (RobertaConfig, RobertaForCausalLM),
@@ -342,6 +348,7 @@ MODEL_FOR_CAUSAL_LM_MAPPING = OrderedDict(
 
 MODEL_FOR_MASKED_LM_MAPPING = OrderedDict(
     [
+        # Model for Masked LM mapping
         (LayoutLMConfig, LayoutLMForMaskedLM),
         (DistilBertConfig, DistilBertForMaskedLM),
         (AlbertConfig, AlbertForMaskedLM),
@@ -363,6 +370,7 @@ MODEL_FOR_MASKED_LM_MAPPING = OrderedDict(
 
 MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING = OrderedDict(
     [
+        # Model for Seq2Seq Causal LM mapping
         (T5Config, T5ForConditionalGeneration),
         (PegasusConfig, PegasusForConditionalGeneration),
         (MarianConfig, MarianMTModel),
@@ -378,6 +386,7 @@ MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING = OrderedDict(
 
 MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING = OrderedDict(
     [
+        # Model for Sequence Classification mapping
         (DistilBertConfig, DistilBertForSequenceClassification),
         (AlbertConfig, AlbertForSequenceClassification),
         (CamembertConfig, CamembertForSequenceClassification),
@@ -401,6 +410,7 @@ MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING = OrderedDict(
 
 MODEL_FOR_QUESTION_ANSWERING_MAPPING = OrderedDict(
     [
+        # Model for Question Answering mapping
         (DistilBertConfig, DistilBertForQuestionAnswering),
         (AlbertConfig, AlbertForQuestionAnswering),
         (CamembertConfig, CamembertForQuestionAnswering),
@@ -422,6 +432,7 @@ MODEL_FOR_QUESTION_ANSWERING_MAPPING = OrderedDict(
 
 MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING = OrderedDict(
     [
+        # Model for Token Classification mapping
         (LayoutLMConfig, LayoutLMForTokenClassification),
         (DistilBertConfig, DistilBertForTokenClassification),
         (CamembertConfig, CamembertForTokenClassification),
@@ -443,6 +454,7 @@ MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING = OrderedDict(
 
 MODEL_FOR_MULTIPLE_CHOICE_MAPPING = OrderedDict(
     [
+        # Model for Multiple Choice mapping
         (CamembertConfig, CamembertForMultipleChoice),
         (ElectraConfig, ElectraForMultipleChoice),
         (XLMRobertaConfig, XLMRobertaForMultipleChoice),

src/transformers/modeling_tf_auto.py

@@ -163,11 +163,14 @@ from .modeling_tf_xlnet import (
 from .utils import logging
 
 
+# Add modeling imports here
+
 logger = logging.get_logger(__name__)
 
 
 TF_MODEL_MAPPING = OrderedDict(
     [
+        # Base model mapping
         (T5Config, TFT5Model),
         (DistilBertConfig, TFDistilBertModel),
         (AlbertConfig, TFAlbertModel),
@@ -192,6 +195,7 @@ TF_MODEL_MAPPING = OrderedDict(
 
 TF_MODEL_FOR_PRETRAINING_MAPPING = OrderedDict(
     [
+        # Model for pre-training mapping
         (T5Config, TFT5ForConditionalGeneration),
         (DistilBertConfig, TFDistilBertForMaskedLM),
         (AlbertConfig, TFAlbertForPreTraining),
@@ -215,6 +219,7 @@ TF_MODEL_FOR_PRETRAINING_MAPPING = OrderedDict(
 
 TF_MODEL_WITH_LM_HEAD_MAPPING = OrderedDict(
     [
+        # Model with LM heads mapping
         (T5Config, TFT5ForConditionalGeneration),
         (DistilBertConfig, TFDistilBertForMaskedLM),
         (AlbertConfig, TFAlbertForMaskedLM),
@@ -240,6 +245,7 @@ TF_MODEL_WITH_LM_HEAD_MAPPING = OrderedDict(
 
 TF_MODEL_FOR_CAUSAL_LM_MAPPING = OrderedDict(
     [
+        # Model for Causal LM mapping
         (BertConfig, TFBertLMHeadModel),
         (OpenAIGPTConfig, TFOpenAIGPTLMHeadModel),
         (GPT2Config, TFGPT2LMHeadModel),
@@ -255,6 +261,7 @@ TF_MODEL_FOR_CAUSAL_LM_MAPPING = OrderedDict(
 
 TF_MODEL_FOR_MASKED_LM_MAPPING = OrderedDict(
     [
+        # Model for Masked LM mapping
         (DistilBertConfig, TFDistilBertForMaskedLM),
         (AlbertConfig, TFAlbertForMaskedLM),
         (CamembertConfig, TFCamembertForMaskedLM),
@@ -273,6 +280,7 @@ TF_MODEL_FOR_MASKED_LM_MAPPING = OrderedDict(
 
 TF_MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING = OrderedDict(
     [
+        # Model for Seq2Seq Causal LM mapping
         (T5Config, TFT5ForConditionalGeneration),
         (MarianConfig, TFMarianMTModel),
         (MBartConfig, TFMBartForConditionalGeneration),
@@ -284,6 +292,7 @@ TF_MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING = OrderedDict(
 
 TF_MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING = OrderedDict(
     [
+        # Model for Sequence Classification mapping
         (DistilBertConfig, TFDistilBertForSequenceClassification),
         (AlbertConfig, TFAlbertForSequenceClassification),
         (CamembertConfig, TFCamembertForSequenceClassification),
@@ -301,6 +310,7 @@ TF_MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING = OrderedDict(
 
 TF_MODEL_FOR_QUESTION_ANSWERING_MAPPING = OrderedDict(
     [
+        # Model for Question Answering mapping
         (DistilBertConfig, TFDistilBertForQuestionAnswering),
         (AlbertConfig, TFAlbertForQuestionAnswering),
         (CamembertConfig, TFCamembertForQuestionAnswering),
@@ -319,6 +329,7 @@ TF_MODEL_FOR_QUESTION_ANSWERING_MAPPING = OrderedDict(
 
 TF_MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING = OrderedDict(
     [
+        # Model for Token Classification mapping
         (DistilBertConfig, TFDistilBertForTokenClassification),
         (AlbertConfig, TFAlbertForTokenClassification),
         (CamembertConfig, TFCamembertForTokenClassification),
@@ -336,6 +347,7 @@ TF_MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING = OrderedDict(
 
 TF_MODEL_FOR_MULTIPLE_CHOICE_MAPPING = OrderedDict(
     [
+        # Model for Multiple Choice mapping
         (CamembertConfig, TFCamembertForMultipleChoice),
         (XLMConfig, TFXLMForMultipleChoice),
         (XLMRobertaConfig, TFXLMRobertaForMultipleChoice),

templates/cookiecutter/README.md

@@ -0,0 +1,91 @@
+# Using the `cookiecutter` to generate models
+
+This folder contains templates to generate new models that fit the current API and pass all tests. It generates
+models in both PyTorch and TensorFlow, completes the `__init__.py` and auto-modeling files, and creates the
+documentation.
+
+## Usage
+
+Using the `cookiecutter` utility requires to have all the `dev` dependencies installed. Let's first clone the 
+repository and install it in our environment:
+
+```shell script
+git clone https://github.com/huggingface/transformers
+cd transformers
+pip install -e ".[dev]"
+```
+
+Once the installation is done, you can use the CLI command `add-new-model` to generate your models:
+
+```shell script
+transformers-cli add-new-model
+```
+
+This should launch the `cookiecutter` package which should prompt you to fill in the configuration.
+
+The `modelname` should be cased according to the plain text casing, i.e., BERT, RoBERTa, DeBERTa.
+```
+modelname [LysBERT]:
+uppercase_modelname [LYSBERT]: 
+lowercase_modelname [<model_name>]: 
+camelcase_modelname [Lysbert]: 
+```
+
+Fill in the `authors` with your team members:
+```
+authors [The HuggingFace Team]: 
+```
+
+The checkpoint identifier is the checkpoint that will be used in the examples across the files. Put the name you wish,
+as it will appear on the modelhub. Do not forget to include the organisation.
+```
+checkpoint_identifier [organisation/<model_name>-base-cased]: 
+```
+
+The tokenizer should either be based on BERT if it behaves exactly like the BERT tokenizer, or a standalone otherwise.
+```
+Select tokenizer_type:
+1 - Based on BERT
+2 - Standalone
+Choose from 1, 2 [1]: 
+```
+
+Choose if your model is an encoder-decoder, or an encoder-only architecture.
+
+If your model is an encoder-only architecture, the generated architecture will be based on the BERT model. 
+If your model is an encoder-decoder architecture, the generated architecture will be based on the BART model. You can,
+of course, edit the files once the generation is complete.
+```
+Select is_encoder_decoder_model:
+1 - True
+2 - False
+Choose from 1, 2 [1]: 
+```
+
+Once the command has finished, you should have a total of 7 new files spread across the repository:
+```
+docs/source/model_doc/<model_name>.rst
+src/transformers/configuration_<model_name>.py
+src/transformers/modeling_<model_name>.py
+src/transformers/modeling_tf_<model_name>.py
+src/transformers/tokenization_<model_name>.py
+tests/test_modeling_<model_name>.py
+tests/test_modeling_tf_<model_name>.py
+```
+
+You can run the tests to ensure that they all pass:
+
+```
+python -m pytest ./tests/test_*<model_name>*.py
+```
+
+Feel free to modify each file to mimic the behavior of your model. Simply re-run the tests (and edit them if a change 
+is needed!) afterwards to make sure everything works as expected. 
+
+Once the files are generated and you are happy with your changes, here's a checklist to ensure that your contribution
+will be merged quickly:
+
+- You should run the `make fixup` utility to fix the style of the files and to ensure the code quality meets the
+  library's standards.
+- You should complete the documentation file (`docs/source/model_doc/<model_name>.rst`) so that your model may be
+  usable.

templates/cookiecutter/cookiecutter-template-{{cookiecutter.modelname}}/configuration.json

@@ -0,0 +1,11 @@
+{
+  "modelname": "{{cookiecutter.modelname}}",
+  "uppercase_modelname": "{{cookiecutter.uppercase_modelname}}",
+  "lowercase_modelname": "{{cookiecutter.lowercase_modelname}}",
+  "camelcase_modelname": "{{cookiecutter.camelcase_modelname}}",
+  "authors": "{{cookiecutter.authors}}",
+  "checkpoint_identifier": "{{cookiecutter.checkpoint_identifier}}",
+  "tokenizer_type": "{{cookiecutter.tokenizer_type}}",
+  "is_encoder_decoder_model": "{{cookiecutter.is_encoder_decoder_model}}",
+  "generate_tensorflow_and_pytorch": "{{cookiecutter.generate_tensorflow_and_pytorch}}"
+}

templates/cookiecutter/cookiecutter-template-{{cookiecutter.modelname}}/configuration_{{cookiecutter.lowercase_modelname}}.py

@@ -0,0 +1,171 @@
+# coding=utf-8
+# Copyright {{cookiecutter.authors}} and The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" {{cookiecutter.modelname}} model configuration """
+
+from .configuration_utils import PretrainedConfig
+from .utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+{{cookiecutter.uppercase_modelname}}_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "{{cookiecutter.checkpoint_identifier}}": "https://s3.amazonaws.com/models.huggingface.co/bert/{{cookiecutter.checkpoint_identifier}}/config.json",
+    # See all {{cookiecutter.modelname}} models at https://huggingface.co/models?filter={{cookiecutter.lowercase_modelname}}
+}
+
+
+class {{cookiecutter.camelcase_modelname}}Config(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a :class:`~transformers.{{cookiecutter.camelcase_modelname}}Model`.
+    It is used to instantiate an {{cookiecutter.modelname}} 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 {{cookiecutter.modelname}} `{{cookiecutter.checkpoint_identifier}} <https://huggingface.co/{{cookiecutter.checkpoint_identifier}}>`__ architecture.
+
+    Configuration objects inherit from  :class:`~transformers.PretrainedConfig` and can be used
+    to control the model outputs. Read the documentation from  :class:`~transformers.PretrainedConfig`
+    for more information.
+
+
+    Args:
+        vocab_size (:obj:`int`, optional, defaults to 30522):
+            Vocabulary size of the {{cookiecutter.modelname}} model. Defines the different tokens that
+            can be represented by the `inputs_ids` passed to the forward method of :class:`~transformers.{{cookiecutter.camelcase_modelname}}Model`.
+        hidden_size (:obj:`int`, optional, defaults to 768):
+            Dimensionality of the encoder layers and the pooler layer.
+        {% if cookiecutter.is_encoder_decoder_model == "False" -%}
+        num_hidden_layers (:obj:`int`, optional, defaults to 12):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (:obj:`int`, optional, defaults to 12):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        intermediate_size (:obj:`int`, optional, defaults to 3072):
+            Dimensionality of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
+        {% else -%}
+        encoder_layers (:obj:`int`, `optional`, defaults to 12):
+            Number of encoder layers, 6 are used for the `bart-base` model.
+        decoder_layers (:obj:`int`, `optional`, defaults to 12):
+            Number of decoder layers, 6 are used for the `bart-base` model.
+        encoder_attention_heads (:obj:`int`, `optional`, defaults to 16):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        decoder_attention_heads (:obj:`int`, `optional`, defaults to 16):
+            Number of attention heads for each attention layer in the Transformer decoder.
+        decoder_intermediate_dim (:obj:`int`, `optional`, defaults to 4096):
+            Dimensionality of the "intermediate" (often named feed-forward) layer in decoder.
+        encoder_intermediate_dim (:obj:`int`, `optional`, defaults to 4096):
+            Dimensionality of the "intermediate" (often named feed-forward) layer in decoder.
+        {% endif -%}
+        hidden_act (:obj:`str` or :obj:`function`, optional, defaults to "gelu"):
+            The non-linear activation function (function or string) in the encoder and pooler.
+            If string, "gelu", "relu", "swish" and "gelu_new" are supported.
+        hidden_dropout_prob (:obj:`float`, optional, defaults to 0.1):
+            The dropout probabilitiy for all fully connected layers in the embeddings, encoder, and pooler.
+        attention_probs_dropout_prob (:obj:`float`, optional, defaults to 0.1):
+            The dropout ratio for the attention probabilities.
+        max_position_embeddings (:obj:`int`, optional, defaults to 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 (:obj:`int`, optional, defaults to 2):
+            The vocabulary size of the `token_type_ids` passed into :class:`~transformers.{{cookiecutter.camelcase_modelname}}Model`.
+        initializer_range (:obj:`float`, optional, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        layer_norm_eps (:obj:`float`, optional, defaults to 1e-12):
+            The epsilon used by the layer normalization layers.
+
+    Example::
+
+        >>> from transformers import {{cookiecutter.camelcase_modelname}}Model, {{cookiecutter.camelcase_modelname}}Config
+
+        >>> # Initializing a {{cookiecutter.modelname}} {{cookiecutter.checkpoint_identifier}} style configuration
+        >>> configuration = {{cookiecutter.camelcase_modelname}}Config()
+
+        >>> # Initializing a model from the {{cookiecutter.checkpoint_identifier}} style configuration
+        >>> model = {{cookiecutter.camelcase_modelname}}Model(configuration)
+
+        >>> # Accessing the model configuration
+        >>> configuration = model.config
+    """
+    model_type = "{{cookiecutter.lowercase_modelname}}"
+
+    def __init__(
+        self,
+        vocab_size=30522,
+        hidden_size=768,
+        {% if cookiecutter.is_encoder_decoder_model == "False" -%}
+        is_encoder_decoder=False,
+        num_hidden_layers=12,
+        num_attention_heads=12,
+        intermediate_size=3072,
+        {% else -%}
+        is_encoder_decoder=True,
+        encoder_intermediate_dim=4096,
+        encoder_layers=12,
+        encoder_attention_heads=16,
+        decoder_intermediate_dim=4096,
+        decoder_layers=12,
+        decoder_attention_heads=16,
+        force_bos_token_to_be_generated=False,
+        {% endif -%}
+        hidden_act="gelu",
+        hidden_dropout_prob=0.1,
+        attention_probs_dropout_prob=0.1,
+        max_position_embeddings=512,
+        type_vocab_size=2,
+        initializer_range=0.02,
+        layer_norm_eps=1e-12,
+        pad_token_id=1,
+        bos_token_id=0,
+        eos_token_id=2,
+        **kwargs
+    ):
+        super().__init__(
+            pad_token_id=pad_token_id,
+            is_encoder_decoder=is_encoder_decoder,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            **kwargs
+        )
+
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        {% if cookiecutter.is_encoder_decoder_model == "False" -%}
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.intermediate_size = intermediate_size
+        {% else -%}
+        self.encoder_intermediate_dim = encoder_intermediate_dim
+        self.encoder_layers = encoder_layers
+        self.encoder_attention_heads = encoder_attention_heads
+        self.decoder_intermediate_dim = decoder_intermediate_dim
+        self.decoder_layers = decoder_layers
+        self.decoder_attention_heads = decoder_attention_heads
+        self.force_bos_token_to_be_generated = force_bos_token_to_be_generated
+        {% endif -%}
+        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.initializer_range = initializer_range
+        self.layer_norm_eps = layer_norm_eps
+
+    {% if cookiecutter.is_encoder_decoder_model == "True" %}
+    @property
+    def num_attention_heads(self) -> int:
+        return self.encoder_attention_heads
+
+    @property
+    def num_hidden_layers(self) -> int:
+        return self.encoder_layers
+    {% endif -%}
+

templates/cookiecutter/cookiecutter-template-{{cookiecutter.modelname}}/test_modeling_tf_{{cookiecutter.lowercase_modelname}}.py

@@ -0,0 +1,483 @@
+# coding=utf-8
+# Copyright 2018 The Google AI Language Team Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+{% if cookiecutter.is_encoder_decoder_model == "False" -%}
+import unittest
+
+from transformers import {{cookiecutter.camelcase_modelname}}Config, is_tf_available
+from transformers.testing_utils import require_tf, slow
+
+from .test_configuration_common import ConfigTester
+from .test_modeling_tf_common import TFModelTesterMixin, ids_tensor
+
+
+if is_tf_available():
+    import tensorflow as tf
+
+    from transformers.modeling_tf_{{cookiecutter.lowercase_modelname}} import (
+        TF{{cookiecutter.camelcase_modelname}}ForMaskedLM,
+        TF{{cookiecutter.camelcase_modelname}}ForMultipleChoice,
+        TF{{cookiecutter.camelcase_modelname}}ForQuestionAnswering,
+        TF{{cookiecutter.camelcase_modelname}}ForSequenceClassification,
+        TF{{cookiecutter.camelcase_modelname}}ForTokenClassification,
+        TF{{cookiecutter.camelcase_modelname}}Model,
+    )
+
+
+class TF{{cookiecutter.camelcase_modelname}}ModelTester:
+    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 = 13
+        self.seq_length = 7
+        self.is_training = True
+        self.use_input_mask = True
+        self.use_token_type_ids = True
+        self.use_labels = True
+        self.vocab_size = 99
+        self.hidden_size = 32
+        self.num_hidden_layers = 5
+        self.num_attention_heads = 4
+        self.intermediate_size = 37
+        self.hidden_act = "gelu"
+        self.hidden_dropout_prob = 0.1
+        self.attention_probs_dropout_prob = 0.1
+        self.max_position_embeddings = 512
+        self.type_vocab_size = 16
+        self.type_sequence_label_size = 2
+        self.initializer_range = 0.02
+        self.num_labels = 3
+        self.num_choices = 4
+        self.scope = None
+
+    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 = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
+
+        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 = {{cookiecutter.camelcase_modelname}}Config(
+            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,
+            initializer_range=self.initializer_range,
+            return_dict=True,
+        )
+
+        return config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+
+    def create_and_check_{{cookiecutter.lowercase_modelname}}_model(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = TF{{cookiecutter.camelcase_modelname}}Model(config=config)
+        inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
+
+        inputs = [input_ids, input_mask]
+        result = model(inputs)
+
+        result = model(input_ids)
+
+        self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
+
+    def create_and_check_{{cookiecutter.lowercase_modelname}}_for_masked_lm(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = TF{{cookiecutter.camelcase_modelname}}ForMaskedLM(config=config)
+        inputs = {
+            "input_ids": input_ids,
+            "attention_mask": input_mask,
+            "token_type_ids": token_type_ids,
+        }
+        result = model(inputs)
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.vocab_size))
+
+    def create_and_check_{{cookiecutter.lowercase_modelname}}_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 = TF{{cookiecutter.camelcase_modelname}}ForSequenceClassification(config=config)
+        inputs = {
+            "input_ids": input_ids,
+            "attention_mask": input_mask,
+            "token_type_ids": token_type_ids,
+        }
+
+        result = model(inputs)
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels))
+
+    def create_and_check_{{cookiecutter.lowercase_modelname}}_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 = TF{{cookiecutter.camelcase_modelname}}ForMultipleChoice(config=config)
+        multiple_choice_inputs_ids = tf.tile(tf.expand_dims(input_ids, 1), (1, self.num_choices, 1))
+        multiple_choice_input_mask = tf.tile(tf.expand_dims(input_mask, 1), (1, self.num_choices, 1))
+        multiple_choice_token_type_ids = tf.tile(tf.expand_dims(token_type_ids, 1), (1, self.num_choices, 1))
+        inputs = {
+            "input_ids": multiple_choice_inputs_ids,
+            "attention_mask": multiple_choice_input_mask,
+            "token_type_ids": multiple_choice_token_type_ids,
+        }
+        result = model(inputs)
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_choices))
+
+    def create_and_check_{{cookiecutter.lowercase_modelname}}_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 = TF{{cookiecutter.camelcase_modelname}}ForTokenClassification(config=config)
+        inputs = {
+            "input_ids": input_ids,
+            "attention_mask": input_mask,
+            "token_type_ids": token_type_ids,
+        }
+        result = model(inputs)
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.seq_length, self.num_labels))
+
+    def create_and_check_{{cookiecutter.lowercase_modelname}}_for_question_answering(
+        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = TF{{cookiecutter.camelcase_modelname}}ForQuestionAnswering(config=config)
+        inputs = {
+            "input_ids": input_ids,
+            "attention_mask": input_mask,
+            "token_type_ids": token_type_ids,
+        }
+
+        result = model(inputs)
+        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 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_tf
+class TF{{cookiecutter.camelcase_modelname}}ModelTest(TFModelTesterMixin, unittest.TestCase):
+
+    all_model_classes = (
+        (
+            TF{{cookiecutter.camelcase_modelname}}Model,
+            TF{{cookiecutter.camelcase_modelname}}ForMaskedLM,
+            TF{{cookiecutter.camelcase_modelname}}ForQuestionAnswering,
+            TF{{cookiecutter.camelcase_modelname}}ForSequenceClassification,
+            TF{{cookiecutter.camelcase_modelname}}ForTokenClassification,
+            TF{{cookiecutter.camelcase_modelname}}ForMultipleChoice,
+        )
+        if is_tf_available()
+        else ()
+    )
+
+    def setUp(self):
+        self.model_tester = TF{{cookiecutter.camelcase_modelname}}ModelTester(self)
+        self.config_tester = ConfigTester(self, config_class={{cookiecutter.camelcase_modelname}}Config, hidden_size=37)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    def test_{{cookiecutter.lowercase_modelname}}_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_{{cookiecutter.lowercase_modelname}}_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_{{cookiecutter.lowercase_modelname}}_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_{{cookiecutter.lowercase_modelname}}_for_multiple_choice(*config_and_inputs)
+
+    def test_for_question_answering(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_{{cookiecutter.lowercase_modelname}}_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_{{cookiecutter.lowercase_modelname}}_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_{{cookiecutter.lowercase_modelname}}_for_token_classification(*config_and_inputs)
+
+    @slow
+    def test_model_from_pretrained(self):
+        model = TF{{cookiecutter.camelcase_modelname}}Model.from_pretrained("{{cookiecutter.checkpoint_identifier}}")
+        self.assertIsNotNone(model)
+
+{% else -%}
+
+import tempfile
+import unittest
+
+from transformers import is_tf_available
+from transformers.testing_utils import require_tf
+
+from .test_configuration_common import ConfigTester
+from .test_modeling_tf_common import TFModelTesterMixin, ids_tensor
+
+
+if is_tf_available():
+    import tensorflow as tf
+
+    from transformers import {{cookiecutter.camelcase_modelname}}Config, TF{{cookiecutter.camelcase_modelname}}ForConditionalGeneration, TF{{cookiecutter.camelcase_modelname}}Model
+
+
+@require_tf
+class ModelTester:
+    def __init__(self, parent):
+        self.parent = parent
+        self.batch_size = 13
+        self.seq_length = 7
+        self.is_training = True
+        self.use_labels = False
+        self.vocab_size = 99
+        self.hidden_size = 32
+        self.num_hidden_layers = 5
+        self.num_attention_heads = 4
+        self.intermediate_size = 37
+        self.hidden_act = "gelu"
+        self.hidden_dropout_prob = 0.1
+        self.attention_probs_dropout_prob = 0.1
+        self.max_position_embeddings = 20
+        self.eos_token_ids = [2]
+        self.pad_token_id = 1
+        self.bos_token_id = 0
+
+    def prepare_config_and_inputs_for_common(self):
+        input_ids = ids_tensor([self.batch_size, self.seq_length - 1], self.vocab_size)
+        eos_tensor = tf.expand_dims(tf.constant([2] * self.batch_size), 1)
+        input_ids = tf.concat([input_ids, eos_tensor], axis=1)
+        input_ids = tf.clip_by_value(input_ids, 3, self.vocab_size + 1)
+
+        config = {{cookiecutter.camelcase_modelname}}Config(
+            vocab_size=self.vocab_size,
+            hidden_size=self.hidden_size,
+            encoder_layers=self.num_hidden_layers,
+            decoder_layers=self.num_hidden_layers,
+            encoder_attention_heads=self.num_attention_heads,
+            decoder_attention_heads=self.num_attention_heads,
+            encoder_intermediate_dim=self.intermediate_size,
+            decoder_intermediate_dim=self.intermediate_size,
+            dropout=self.hidden_dropout_prob,
+            attention_dropout=self.attention_probs_dropout_prob,
+            max_position_embeddings=self.max_position_embeddings,
+            eos_token_ids=[2],
+            bos_token_id=self.bos_token_id,
+            pad_token_id=self.pad_token_id,
+            decoder_start_token_id=self.pad_token_id,
+        )
+        inputs_dict = prepare_{{cookiecutter.lowercase_modelname}}_inputs_dict(config, input_ids)
+        return config, inputs_dict
+
+
+def prepare_{{cookiecutter.lowercase_modelname}}_inputs_dict(
+    config,
+    input_ids,
+    attention_mask=None,
+):
+    if attention_mask is None:
+        attention_mask = tf.cast(tf.math.not_equal(input_ids, config.pad_token_id), tf.int8)
+    return {
+        "input_ids": input_ids,
+        "decoder_input_ids": input_ids,
+        "attention_mask": attention_mask,
+    }
+
+
+@require_tf
+class TestTF{{cookiecutter.camelcase_modelname}}(TFModelTesterMixin, unittest.TestCase):
+    all_model_classes = (TF{{cookiecutter.camelcase_modelname}}ForConditionalGeneration, TF{{cookiecutter.camelcase_modelname}}Model) if is_tf_available() else ()
+    all_generative_model_classes = (TF{{cookiecutter.camelcase_modelname}}ForConditionalGeneration,) if is_tf_available() else ()
+    is_encoder_decoder = True
+    test_pruning = False
+
+    def setUp(self):
+        self.model_tester = ModelTester(self)
+        self.config_tester = ConfigTester(self, config_class={{cookiecutter.camelcase_modelname}}Config)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    def test_inputs_embeds(self):
+        # inputs_embeds not supported
+        pass
+
+    def test_compile_tf_model(self):
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+
+        optimizer = tf.keras.optimizers.Adam(learning_rate=3e-5, epsilon=1e-08, clipnorm=1.0)
+        loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)
+        metric = tf.keras.metrics.SparseCategoricalAccuracy("accuracy")
+
+        model_class = TF{{cookiecutter.camelcase_modelname}}ForConditionalGeneration
+        input_ids = {
+            "decoder_input_ids": tf.keras.Input(batch_shape=(2, 2000), name="decoder_input_ids", dtype="int32"),
+            "input_ids": tf.keras.Input(batch_shape=(2, 2000), name="input_ids", dtype="int32"),
+        }
+
+        # Prepare our model
+        model = model_class(config)
+        model(self._prepare_for_class(inputs_dict, model_class))  # Model must be called before saving.
+        # Let's load it from the disk to be sure we can use pretrained weights
+        with tempfile.TemporaryDirectory() as tmpdirname:
+            model.save_pretrained(tmpdirname)
+            model = model_class.from_pretrained(tmpdirname)
+
+        outputs_dict = model(input_ids)
+        hidden_states = outputs_dict[0]
+
+        # Add a dense layer on top to test integration with other keras modules
+        outputs = tf.keras.layers.Dense(2, activation="softmax", name="outputs")(hidden_states)
+
+        # Compile extended model
+        extendehidden_size = tf.keras.Model(inputs=[input_ids], outputs=[outputs])
+        extendehidden_size.compile(optimizer=optimizer, loss=loss, metrics=[metric])
+
+    def test_savehidden_size_with_hidden_states_output(self):
+        # Should be uncommented during patrick TF refactor
+        pass
+
+    def test_savehidden_size_with_attentions_output(self):
+        # Should be uncommented during patrick TF refactor
+        pass
+
+
+@require_tf
+class TF{{cookiecutter.camelcase_modelname}}HeadTests(unittest.TestCase):
+
+    vocab_size = 99
+
+    def _get_config_and_data(self):
+        eos_column_vector = tf.ones((4, 1), dtype=tf.int32) * 2
+        input_ids = tf.concat([ids_tensor((4, 6), self.vocab_size - 3) + 3, eos_column_vector], axis=1)
+        batch_size = input_ids.shape[0]
+        config = {{cookiecutter.camelcase_modelname}}Config(
+            vocab_size=self.vocab_size,
+            hidden_size=24,
+            encoder_layers=2,
+            decoder_layers=2,
+            encoder_attention_heads=2,
+            decoder_attention_heads=2,
+            encoder_intermediate_dim=32,
+            decoder_intermediate_dim=32,
+            max_position_embeddings=48,
+            eos_token_id=2,
+            pad_token_id=1,
+            bos_token_id=0,
+            return_dict=True,
+            decoder_start_token_id=2,
+        )
+        return config, input_ids, batch_size
+
+    def test_lm_forward(self):
+        config, input_ids, batch_size = self._get_config_and_data()
+        decoder_lm_labels = ids_tensor([batch_size, input_ids.shape[1]], self.vocab_size)
+        lm_model = TF{{cookiecutter.camelcase_modelname}}ForConditionalGeneration(config)
+        outputs = lm_model(inputs=input_ids, lm_labels=decoder_lm_labels, decoder_input_ids=input_ids, use_cache=False)
+        expected_shape = (batch_size, input_ids.shape[1], config.vocab_size)
+        self.assertEqual(outputs.logits.shape, expected_shape)
+
+    def test_lm_uneven_forward(self):
+        config = {{cookiecutter.camelcase_modelname}}Config(
+            vocab_size=10,
+            hidden_size=24,
+            encoder_layers=2,
+            decoder_layers=2,
+            encoder_attention_heads=2,
+            decoder_attention_heads=2,
+            encoder_intermediate_dim=32,
+            decoder_intermediate_dim=32,
+            max_position_embeddings=48,
+            return_dict=True,
+        )
+        lm_model = TF{{cookiecutter.camelcase_modelname}}ForConditionalGeneration(config)
+        context = tf.fill((7, 2), 4)
+        summary = tf.fill((7, 7), 6)
+        outputs = lm_model(inputs=context, decoder_input_ids=summary, use_cache=False)
+        expected_shape = (*summary.shape, config.vocab_size)
+        self.assertEqual(outputs.logits.shape, expected_shape)
+
+
+def _assert_tensors_equal(a, b, atol=1e-12, prefix=""):
+    """If tensors not close, or a and b arent both tensors, raise a nice Assertion error."""
+    if a is None and b is None:
+        return True
+    try:
+        if tf.debugging.assert_near(a, b, atol=atol):
+            return True
+        raise
+    except Exception:
+        msg = "{} != {}".format(a, b)
+        if prefix:
+            msg = prefix + ": " + msg
+        raise AssertionError(msg)
+
+
+{% endif -%}

templates/cookiecutter/cookiecutter-template-{{cookiecutter.modelname}}/test_modeling_{{cookiecutter.lowercase_modelname}}.py

@@ -0,0 +1,659 @@
+# coding=utf-8
+# Copyright 2018 The Google AI Language Team Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" Testing suite for the PyTorch {{cookiecutter.modelname}} model. """
+
+
+{% if cookiecutter.is_encoder_decoder_model == "False" -%}
+import unittest
+
+from transformers import is_torch_available
+from transformers.testing_utils import require_torch, slow, torch_device
+from .test_configuration_common import ConfigTester
+
+from .test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
+
+
+if is_torch_available():
+    from transformers import (
+        {{cookiecutter.camelcase_modelname}}Config,
+        {{cookiecutter.camelcase_modelname}}ForMaskedLM,
+        {{cookiecutter.camelcase_modelname}}ForMultipleChoice,
+        {{cookiecutter.camelcase_modelname}}ForQuestionAnswering,
+        {{cookiecutter.camelcase_modelname}}ForSequenceClassification,
+        {{cookiecutter.camelcase_modelname}}ForTokenClassification,
+        {{cookiecutter.camelcase_modelname}}Model,
+    )
+    from transformers.modeling_{{cookiecutter.lowercase_modelname}} import {{cookiecutter.uppercase_modelname}}_PRETRAINED_MODEL_ARCHIVE_LIST
+
+
+class {{cookiecutter.camelcase_modelname}}ModelTester:
+    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 = {{cookiecutter.camelcase_modelname}}Config(
+            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,
+            return_dict=True,
+        )
+
+        return config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+
+    def create_and_check_model(
+            self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = {{cookiecutter.camelcase_modelname}}Model(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(input_ids, attention_mask=input_mask, token_type_ids=token_type_ids)
+        result = model(input_ids, token_type_ids=token_type_ids)
+        result = model(input_ids)
+        self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
+
+    def create_and_check_for_masked_lm(
+            self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = {{cookiecutter.camelcase_modelname}}ForMaskedLM(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_question_answering(
+            self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+    ):
+        model = {{cookiecutter.camelcase_modelname}}ForQuestionAnswering(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 = {{cookiecutter.camelcase_modelname}}ForSequenceClassification(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 = {{cookiecutter.camelcase_modelname}}ForTokenClassification(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 = {{cookiecutter.camelcase_modelname}}ForMultipleChoice(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 {{cookiecutter.camelcase_modelname}}ModelTest(ModelTesterMixin, unittest.TestCase):
+
+    all_model_classes = (
+        (
+            {{cookiecutter.camelcase_modelname}}Model,
+            {{cookiecutter.camelcase_modelname}}ForMaskedLM,
+            {{cookiecutter.camelcase_modelname}}ForMultipleChoice,
+            {{cookiecutter.camelcase_modelname}}ForQuestionAnswering,
+            {{cookiecutter.camelcase_modelname}}ForSequenceClassification,
+            {{cookiecutter.camelcase_modelname}}ForTokenClassification,
+        )
+        if is_torch_available()
+        else ()
+    )
+
+    def setUp(self):
+        self.model_tester = {{cookiecutter.camelcase_modelname}}ModelTester(self)
+        self.config_tester = ConfigTester(self, config_class={{cookiecutter.camelcase_modelname}}Config, hidden_size=37)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    def test_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_model(*config_and_inputs)
+
+    def test_for_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_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 {{cookiecutter.uppercase_modelname}}_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
+            model = {{cookiecutter.camelcase_modelname}}Model.from_pretrained(model_name)
+            self.assertIsNotNone(model)
+
+
+{% else -%}
+import tempfile
+import unittest
+
+import timeout_decorator  # noqa
+
+from transformers import is_torch_available
+from transformers.testing_utils import require_torch, torch_device
+
+from .test_configuration_common import ConfigTester
+from .test_modeling_common import ModelTesterMixin, ids_tensor
+
+
+if is_torch_available():
+    import torch
+
+    from transformers import (
+        {{cookiecutter.camelcase_modelname}}Config,
+        {{cookiecutter.camelcase_modelname}}ForConditionalGeneration,
+        {{cookiecutter.camelcase_modelname}}ForQuestionAnswering,
+        {{cookiecutter.camelcase_modelname}}ForSequenceClassification,
+        {{cookiecutter.camelcase_modelname}}Model,
+    )
+    from transformers.modeling_{{cookiecutter.lowercase_modelname}} import (
+        _prepare_{{cookiecutter.lowercase_modelname}}_decoder_inputs,
+        invert_mask,
+        shift_tokens_right,
+    )
+PGE_ARTICLE = """ PG&E stated it scheduled the blackouts in response to forecasts for high winds amid dry conditions. The aim is to reduce the risk of wildfires. Nearly 800 thousand customers were scheduled to be affected by the shutoffs which were expected to last through at least midday tomorrow."""
+
+
+@require_torch
+class ModelTester:
+    def __init__(
+        self,
+        parent,
+    ):
+        self.parent = parent
+        self.batch_size = 13
+        self.seq_length = 7
+        self.is_training = True
+        self.use_labels = False
+        self.vocab_size = 99
+        self.hidden_size = 16
+        self.num_hidden_layers = 2
+        self.num_attention_heads = 4
+        self.intermediate_size = 4
+        self.hidden_act = "gelu"
+        self.hidden_dropout_prob = 0.1
+        self.attention_probs_dropout_prob = 0.1
+        self.max_position_embeddings = 20
+        self.eos_token_id = 2
+        self.pad_token_id = 1
+        self.bos_token_id = 0
+        torch.manual_seed(0)
+
+    def prepare_config_and_inputs(self):
+        input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size).clamp(
+            3,
+        )
+        input_ids[:, -1] = 2  # Eos Token
+
+        config = {{cookiecutter.camelcase_modelname}}Config(
+            vocab_size=self.vocab_size,
+            hidden_size=self.hidden_size,
+            encoder_layers=self.num_hidden_layers,
+            decoder_layers=self.num_hidden_layers,
+            encoder_attention_heads=self.num_attention_heads,
+            decoder_attention_heads=self.num_attention_heads,
+            encoder_ffn_dim=self.intermediate_size,
+            decoder_ffn_dim=self.intermediate_size,
+            dropout=self.hidden_dropout_prob,
+            attention_dropout=self.attention_probs_dropout_prob,
+            max_position_embeddings=self.max_position_embeddings,
+            eos_token_id=self.eos_token_id,
+            bos_token_id=self.bos_token_id,
+            pad_token_id=self.pad_token_id,
+        )
+        inputs_dict = prepare_{{cookiecutter.lowercase_modelname}}_inputs_dict(config, input_ids)
+        return config, inputs_dict
+
+    def prepare_config_and_inputs_for_common(self):
+        config, inputs_dict = self.prepare_config_and_inputs()
+        inputs_dict["decoder_input_ids"] = inputs_dict["input_ids"]
+        inputs_dict["decoder_attention_mask"] = inputs_dict["attention_mask"]
+        inputs_dict["use_cache"] = False
+        return config, inputs_dict
+
+
+def prepare_{{cookiecutter.lowercase_modelname}}_inputs_dict(
+    config,
+    input_ids,
+    attention_mask=None,
+):
+    if attention_mask is None:
+        attention_mask = input_ids.ne(config.pad_token_id)
+    return {
+        "input_ids": input_ids,
+        "attention_mask": attention_mask,
+    }
+
+
+@require_torch
+class {{cookiecutter.uppercase_modelname}}ModelTest(ModelTesterMixin, unittest.TestCase):
+    all_model_classes = (
+        ({{cookiecutter.camelcase_modelname}}Model, {{cookiecutter.camelcase_modelname}}ForConditionalGeneration, {{cookiecutter.camelcase_modelname}}ForSequenceClassification, {{cookiecutter.camelcase_modelname}}ForQuestionAnswering)
+        if is_torch_available()
+        else ()
+    )
+    all_generative_model_classes = ({{cookiecutter.camelcase_modelname}}ForConditionalGeneration,) if is_torch_available() else ()
+    is_encoder_decoder = True
+    test_pruning = False
+    test_head_masking = False
+    test_missing_keys = False
+
+    def setUp(self):
+        self.model_tester = ModelTester(self)
+        self.config_tester = ConfigTester(self, config_class={{cookiecutter.camelcase_modelname}}Config)
+
+    def test_config(self):
+        self.config_tester.run_common_tests()
+
+    def test_initialization_more(self):
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs()
+        model = {{cookiecutter.camelcase_modelname}}Model(config)
+        model.to(torch_device)
+        model.eval()
+        # test init
+        self.assertTrue((model.encoder.embed_tokens.weight == model.shared.weight).all().item())
+
+        def _check_var(module):
+            """Check that we initialized various parameters from N(0, config.initializer_range)."""
+            self.assertAlmostEqual(torch.std(module.weight).item(), config.initializer_range, 2)
+
+        _check_var(model.encoder.embed_tokens)
+        _check_var(model.encoder.layers[0].self_attn.k_proj)
+        _check_var(model.encoder.layers[0].fc1)
+        _check_var(model.encoder.embed_positions)
+
+    def test_advanced_inputs(self):
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs()
+        config.use_cache = False
+        inputs_dict["input_ids"][:, -2:] = config.pad_token_id
+        decoder_input_ids, decoder_attn_mask, causal_mask = _prepare_{{cookiecutter.lowercase_modelname}}_decoder_inputs(
+            config, inputs_dict["input_ids"]
+        )
+        model = {{cookiecutter.camelcase_modelname}}Model(config).to(torch_device).eval()
+
+        decoder_features_with_created_mask = model(**inputs_dict)[0]
+        decoder_features_with_passed_mask = model(
+            decoder_attention_mask=invert_mask(decoder_attn_mask), decoder_input_ids=decoder_input_ids, **inputs_dict
+        )[0]
+        assert_tensors_close(decoder_features_with_passed_mask, decoder_features_with_created_mask)
+        useless_mask = torch.zeros_like(decoder_attn_mask)
+        decoder_features = model(decoder_attention_mask=useless_mask, **inputs_dict)[0]
+        self.assertTrue(isinstance(decoder_features, torch.Tensor))  # no hidden states or attentions
+        self.assertEqual(
+            decoder_features.size(), (self.model_tester.batch_size, self.model_tester.seq_length, config.hidden_size)
+        )
+        if decoder_attn_mask.min().item() < -1e3:  # some tokens were masked
+            self.assertFalse((decoder_features_with_created_mask == decoder_features).all().item())
+
+        # Test different encoder attention masks
+        decoder_features_with_long_encoder_mask = model(
+            inputs_dict["input_ids"], attention_mask=inputs_dict["attention_mask"].long()
+        )[0]
+        assert_tensors_close(decoder_features_with_long_encoder_mask, decoder_features_with_created_mask)
+
+    def test_save_load_strict(self):
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs()
+        for model_class in self.all_model_classes:
+            model = model_class(config)
+
+            with tempfile.TemporaryDirectory() as tmpdirname:
+                model.save_pretrained(tmpdirname)
+                model2, info = model_class.from_pretrained(tmpdirname, output_loading_info=True)
+            self.assertEqual(info["missing_keys"], [])
+
+    @unittest.skip("Passing inputs_embeds not implemented for {{cookiecutter.camelcase_modelname}}.")
+    def test_inputs_embeds(self):
+        pass
+
+
+@require_torch
+class {{cookiecutter.camelcase_modelname}}HeadTests(unittest.TestCase):
+    vocab_size = 99
+
+    def _get_config_and_data(self):
+        input_ids = torch.tensor(
+            [
+                [71, 82, 18, 33, 46, 91, 2],
+                [68, 34, 26, 58, 30, 82, 2],
+                [5, 97, 17, 39, 94, 40, 2],
+                [76, 83, 94, 25, 70, 78, 2],
+                [87, 59, 41, 35, 48, 66, 2],
+                [55, 13, 16, 58, 5, 2, 1],  # note padding
+                [64, 27, 31, 51, 12, 75, 2],
+                [52, 64, 86, 17, 83, 39, 2],
+                [48, 61, 9, 24, 71, 82, 2],
+                [26, 1, 60, 48, 22, 13, 2],
+                [21, 5, 62, 28, 14, 76, 2],
+                [45, 98, 37, 86, 59, 48, 2],
+                [70, 70, 50, 9, 28, 0, 2],
+            ],
+            dtype=torch.long,
+            device=torch_device,
+        )
+
+        batch_size = input_ids.shape[0]
+        config = {{cookiecutter.camelcase_modelname}}Config(
+            vocab_size=self.vocab_size,
+            hidden_size=24,
+            encoder_layers=2,
+            decoder_layers=2,
+            encoder_attention_heads=2,
+            decoder_attention_heads=2,
+            encoder_ffn_dim=32,
+            decoder_ffn_dim=32,
+            max_position_embeddings=48,
+            eos_token_id=2,
+            pad_token_id=1,
+            bos_token_id=0,
+            return_dict=True,
+        )
+        return config, input_ids, batch_size
+
+    def test_sequence_classification_forward(self):
+        config, input_ids, batch_size = self._get_config_and_data()
+        labels = _long_tensor([2] * batch_size).to(torch_device)
+        config.num_labels = 3
+        model = {{cookiecutter.camelcase_modelname}}ForSequenceClassification(config)
+        model.to(torch_device)
+        outputs = model(input_ids=input_ids, decoder_input_ids=input_ids, labels=labels)
+        expected_shape = torch.Size((batch_size, config.num_labels))
+        self.assertEqual(outputs["logits"].shape, expected_shape)
+        self.assertIsInstance(outputs["loss"].item(), float)
+
+    def test_question_answering_forward(self):
+        config, input_ids, batch_size = self._get_config_and_data()
+        sequence_labels = ids_tensor([batch_size], 2).to(torch_device)
+        model = {{cookiecutter.camelcase_modelname}}ForQuestionAnswering(config)
+        model.to(torch_device)
+        outputs = model(
+            input_ids=input_ids,
+            start_positions=sequence_labels,
+            end_positions=sequence_labels,
+        )
+
+        self.assertEqual(outputs["start_logits"].shape, input_ids.shape)
+        self.assertEqual(outputs["end_logits"].shape, input_ids.shape)
+        self.assertIsInstance(outputs["loss"].item(), float)
+
+    @timeout_decorator.timeout(1)
+    def test_lm_forward(self):
+        config, input_ids, batch_size = self._get_config_and_data()
+        lm_labels = ids_tensor([batch_size, input_ids.shape[1]], self.vocab_size).to(torch_device)
+        lm_model = {{cookiecutter.camelcase_modelname}}ForConditionalGeneration(config)
+        lm_model.to(torch_device)
+        outputs = lm_model(input_ids=input_ids, labels=lm_labels)
+        expected_shape = (batch_size, input_ids.shape[1], config.vocab_size)
+        self.assertEqual(outputs["logits"].shape, expected_shape)
+        self.assertIsInstance(outputs["loss"].item(), float)
+
+    def test_lm_uneven_forward(self):
+        config = {{cookiecutter.camelcase_modelname}}Config(
+            vocab_size=self.vocab_size,
+            hidden_size=14,
+            encoder_layers=2,
+            decoder_layers=2,
+            encoder_attention_heads=2,
+            decoder_attention_heads=2,
+            encoder_ffn_dim=8,
+            decoder_ffn_dim=8,
+            max_position_embeddings=48,
+            return_dict=True,
+        )
+        lm_model = {{cookiecutter.camelcase_modelname}}ForConditionalGeneration(config).to(torch_device)
+        context = torch.Tensor([[71, 82, 18, 33, 46, 91, 2], [68, 34, 26, 58, 30, 2, 1]]).long().to(torch_device)
+        summary = torch.Tensor([[82, 71, 82, 18, 2], [58, 68, 2, 1, 1]]).long().to(torch_device)
+        outputs = lm_model(input_ids=context, decoder_input_ids=summary, labels=summary)
+        expected_shape = (*summary.shape, config.vocab_size)
+        self.assertEqual(outputs["logits"].shape, expected_shape)
+
+    def test_generate_beam_search(self):
+        input_ids = torch.Tensor([[71, 82, 2], [68, 34, 2]]).long().to(torch_device)
+        config = {{cookiecutter.camelcase_modelname}}Config(
+            vocab_size=self.vocab_size,
+            hidden_size=24,
+            encoder_layers=2,
+            decoder_layers=2,
+            encoder_attention_heads=2,
+            decoder_attention_heads=2,
+            encoder_ffn_dim=32,
+            decoder_ffn_dim=32,
+            max_position_embeddings=48,
+            eos_token_id=2,
+            pad_token_id=1,
+            bos_token_id=0,
+        )
+        lm_model = {{cookiecutter.camelcase_modelname}}ForConditionalGeneration(config).to(torch_device)
+        lm_model.eval()
+
+        max_length = 5
+        generated_ids = lm_model.generate(
+            input_ids.clone(),
+            do_sample=True,
+            num_return_sequences=1,
+            num_beams=2,
+            no_repeat_ngram_size=3,
+            max_length=max_length,
+        )
+        self.assertEqual(generated_ids.shape, (input_ids.shape[0], max_length))
+
+    def test_shift_tokens_right(self):
+        input_ids = torch.Tensor([[71, 82, 18, 33, 2, 1, 1], [68, 34, 26, 58, 30, 82, 2]]).long()
+        shifted = shift_tokens_right(input_ids, 1)
+        n_pad_before = input_ids.eq(1).float().sum()
+        n_pad_after = shifted.eq(1).float().sum()
+        self.assertEqual(shifted.shape, input_ids.shape)
+        self.assertEqual(n_pad_after, n_pad_before - 1)
+        self.assertTrue(torch.eq(shifted[:, 0], 2).all())
+
+    def test_generate_fp16(self):
+        config, input_ids, batch_size = self._get_config_and_data()
+        attention_mask = input_ids.ne(1).to(torch_device)
+        model = {{cookiecutter.camelcase_modelname}}ForConditionalGeneration(config).eval().to(torch_device)
+        if torch_device == "cuda":
+            model.half()
+        model.generate(input_ids, attention_mask=attention_mask)
+        model.generate(num_beams=4, do_sample=True, early_stopping=False, num_return_sequences=3)
+
+    def test_dummy_inputs(self):
+        config, *_ = self._get_config_and_data()
+        model = {{cookiecutter.camelcase_modelname}}ForConditionalGeneration(config).eval().to(torch_device)
+        model(**model.dummy_inputs)
+
+    def test_prepare_{{cookiecutter.lowercase_modelname}}_decoder_inputs(self):
+        config, *_ = self._get_config_and_data()
+        input_ids = _long_tensor(([4, 4, 2]))
+        decoder_input_ids = _long_tensor([[26388, 2, config.pad_token_id]])
+        ignore = float("-inf")
+        decoder_input_ids, decoder_attn_mask, causal_mask = _prepare_{{cookiecutter.lowercase_modelname}}_decoder_inputs(
+            config, input_ids, decoder_input_ids
+        )
+        expected_causal_mask = torch.tensor(
+            [[0, ignore, ignore], [0, 0, ignore], [0, 0, 0]]  # never attend to the final token, because its pad
+        ).to(input_ids.device)
+        self.assertEqual(decoder_attn_mask.size(), decoder_input_ids.size())
+        self.assertTrue(torch.eq(expected_causal_mask, causal_mask).all())
+
+    def test_resize_tokens_embeddings_more(self):
+        config, input_ids, _ = self._get_config_and_data()
+
+        def _get_embs(m):
+            return (m.get_input_embeddings().weight.data.clone(), m.get_output_embeddings().weight.data.clone())
+
+        model = {{cookiecutter.camelcase_modelname}}ForConditionalGeneration(config).eval().to(torch_device)
+        input, output = _get_embs(model)
+        self.assertTrue(torch.eq(input, output).all())
+        new_vocab_size = 45
+        model.resize_token_embeddings(new_vocab_size)
+        input_new, output_new = _get_embs(model)
+        self.assertEqual(input_new.shape, (new_vocab_size, config.hidden_size))
+        self.assertEqual(output_new.shape, (new_vocab_size, config.hidden_size))
+        self.assertTrue(torch.eq(input_new, output_new).all())
+
+
+def assert_tensors_close(a, b, atol=1e-12, prefix=""):
+    """If tensors have different shapes, different values or a and b are not both tensors, raise a nice Assertion error."""
+    if a is None and b is None:
+        return True
+    try:
+        if torch.allclose(a, b, atol=atol):
+            return True
+        raise
+    except Exception:
+        pct_different = (torch.gt((a - b).abs(), atol)).float().mean().item()
+        if a.numel() > 100:
+            msg = f"tensor values are {pct_different:.1%} percent different."
+        else:
+            msg = f"{a} != {b}"
+        if prefix:
+            msg = prefix + ": " + msg
+        raise AssertionError(msg)
+
+
+def _long_tensor(tok_lst):
+    return torch.tensor(tok_lst, dtype=torch.long, device=torch_device)
+
+{% endif -%}

templates/cookiecutter/cookiecutter-template-{{cookiecutter.modelname}}/to_replace_{{cookiecutter.lowercase_modelname}}.py

@@ -0,0 +1,269 @@
+## This file is made so that specific statements may be copied inside existing files. This is useful to copy
+## import statements in __init__.py, or to complete model lists in the AUTO files.
+##
+## It is to be used as such:
+## Put '# To replace in: "FILE_PATH"' in order to indicate the contents will be copied in the file at path FILE_PATH
+## Put '# Below: "STATEMENT"' in order to copy the contents below **the first occurence** of that line in the file at FILE_PATH
+## Put '# Replace with:' followed by the lines containing the content to define the content
+## End a statement with '# End.'. If starting a new statement without redefining the FILE_PATH, it will continue pasting
+## content in that file.
+##
+## Put '## COMMENT' to comment on the file.
+
+
+# To replace in: "src/transformers/__init__.py"
+# Below: "if is_torch_available():" if generating PyTorch
+# Replace with:
+{% if cookiecutter.is_encoder_decoder_model == "False" %}
+    from .modeling_{{cookiecutter.lowercase_modelname}} import (
+        {{cookiecutter.uppercase_modelname}}_PRETRAINED_MODEL_ARCHIVE_LIST,
+        {{cookiecutter.camelcase_modelname}}ForMaskedLM,
+        {{cookiecutter.camelcase_modelname}}ForMultipleChoice,
+        {{cookiecutter.camelcase_modelname}}ForQuestionAnswering,
+        {{cookiecutter.camelcase_modelname}}ForSequenceClassification,
+        {{cookiecutter.camelcase_modelname}}ForTokenClassification,
+        {{cookiecutter.camelcase_modelname}}Layer,
+        {{cookiecutter.camelcase_modelname}}Model,
+        {{cookiecutter.camelcase_modelname}}PreTrainedModel,
+        load_tf_weights_in_{{cookiecutter.lowercase_modelname}},
+    )
+{% else %}
+    from .modeling_{{cookiecutter.lowercase_modelname}} import (
+        {{cookiecutter.uppercase_modelname}}_PRETRAINED_MODEL_ARCHIVE_LIST,
+        {{cookiecutter.camelcase_modelname}}ForConditionalGeneration,
+        {{cookiecutter.camelcase_modelname}}ForQuestionAnswering,
+        {{cookiecutter.camelcase_modelname}}ForSequenceClassification,
+        {{cookiecutter.camelcase_modelname}}Model,
+        Pretrained{{cookiecutter.camelcase_modelname}}Model,
+    )
+{% endif -%}
+# End.
+
+# Below: "if is_tf_available():" if generating TensorFlow
+# Replace with:
+{% if cookiecutter.is_encoder_decoder_model == "False" %}
+    from .modeling_tf_{{cookiecutter.lowercase_modelname}} import (
+        TF_{{cookiecutter.uppercase_modelname}}_PRETRAINED_MODEL_ARCHIVE_LIST,
+        TF{{cookiecutter.camelcase_modelname}}ForMaskedLM,
+        TF{{cookiecutter.camelcase_modelname}}ForMultipleChoice,
+        TF{{cookiecutter.camelcase_modelname}}ForQuestionAnswering,
+        TF{{cookiecutter.camelcase_modelname}}ForSequenceClassification,
+        TF{{cookiecutter.camelcase_modelname}}ForTokenClassification,
+        TF{{cookiecutter.camelcase_modelname}}Layer,
+        TF{{cookiecutter.camelcase_modelname}}Model,
+        TF{{cookiecutter.camelcase_modelname}}PreTrainedModel,
+    )
+{% else %}
+    from .modeling_tf_{{cookiecutter.lowercase_modelname}} import (
+        TF_{{cookiecutter.uppercase_modelname}}_PRETRAINED_MODEL_ARCHIVE_LIST,
+        TF{{cookiecutter.camelcase_modelname}}ForConditionalGeneration,
+        TF{{cookiecutter.camelcase_modelname}}Model,
+        TFPretrained{{cookiecutter.camelcase_modelname}}Model,
+    )
+{% endif -%}
+# End.
+
+
+# Below: "from .configuration_albert import ALBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, AlbertConfig"
+# Replace with:
+from .configuration_{{cookiecutter.lowercase_modelname}} import {{cookiecutter.uppercase_modelname}}_PRETRAINED_CONFIG_ARCHIVE_MAP, {{cookiecutter.camelcase_modelname}}Config
+# End.
+
+
+
+# To replace in: "src/transformers/configuration_auto.py"
+# Below: "# Add configs here"
+# Replace with:
+        ("{{cookiecutter.lowercase_modelname}}", {{cookiecutter.camelcase_modelname}}Config),
+# End.
+
+# Below: "# Add archive maps here"
+# Replace with:
+        {{cookiecutter.uppercase_modelname}}_PRETRAINED_CONFIG_ARCHIVE_MAP,
+# End.
+
+# Below: "from .configuration_albert import ALBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, AlbertConfig",
+# Replace with:
+from .configuration_{{cookiecutter.lowercase_modelname}} import {{cookiecutter.uppercase_modelname}}_PRETRAINED_CONFIG_ARCHIVE_MAP, {{cookiecutter.camelcase_modelname}}Config
+# End.
+
+# Below: "# Add cased model names here"
+# Replace with:
+        ("{{cookiecutter.lowercase_modelname}}", "{{cookiecutter.camelcase_modelname}}"),
+# End.
+
+
+
+# To replace in: "src/transformers/modeling_auto.py" if generating PyTorch
+# Below: "from .configuration_auto import ("
+# Replace with:
+    {{cookiecutter.camelcase_modelname}}Config,
+# End.
+
+# Below: "# Add modeling imports here"
+# Replace with:
+
+{% if cookiecutter.is_encoder_decoder_model == "False" -%}
+from .modeling_{{cookiecutter.lowercase_modelname}} import (
+    {{cookiecutter.camelcase_modelname}}ForMaskedLM,
+    {{cookiecutter.camelcase_modelname}}ForMultipleChoice,
+    {{cookiecutter.camelcase_modelname}}ForQuestionAnswering,
+    {{cookiecutter.camelcase_modelname}}ForSequenceClassification,
+    {{cookiecutter.camelcase_modelname}}ForTokenClassification,
+    {{cookiecutter.camelcase_modelname}}Model,
+)
+{% else -%}
+from .modeling_{{cookiecutter.lowercase_modelname}} import (
+    {{cookiecutter.camelcase_modelname}}ForConditionalGeneration,
+    {{cookiecutter.camelcase_modelname}}ForQuestionAnswering,
+    {{cookiecutter.camelcase_modelname}}ForSequenceClassification,
+    {{cookiecutter.camelcase_modelname}}Model,
+)
+{% endif -%}
+# End.
+
+# Below: "# Base model mapping"
+# Replace with:
+        ({{cookiecutter.camelcase_modelname}}Config, {{cookiecutter.camelcase_modelname}}Model),
+# End.
+
+# Below: "# Model with LM heads mapping"
+# Replace with:
+{% if cookiecutter.is_encoder_decoder_model == "False" -%}
+        ({{cookiecutter.camelcase_modelname}}Config, {{cookiecutter.camelcase_modelname}}ForMaskedLM),
+{% else -%}
+        ({{cookiecutter.camelcase_modelname}}Config, {{cookiecutter.camelcase_modelname}}ForConditionalGeneration),
+{% endif -%}
+# End.
+
+# Below: "# Model for Masked LM mapping"
+# Replace with:
+{% if cookiecutter.is_encoder_decoder_model == "False" -%}
+        ({{cookiecutter.camelcase_modelname}}Config, {{cookiecutter.camelcase_modelname}}ForMaskedLM),
+{% else -%}
+{% endif -%}
+# End.
+
+# Below: "# Model for Sequence Classification mapping"
+# Replace with:
+        ({{cookiecutter.camelcase_modelname}}Config, {{cookiecutter.camelcase_modelname}}ForSequenceClassification),
+# End.
+
+# Below: "# Model for Question Answering mapping"
+# Replace with:
+        ({{cookiecutter.camelcase_modelname}}Config, {{cookiecutter.camelcase_modelname}}ForQuestionAnswering),
+# End.
+
+# Below: "# Model for Token Classification mapping"
+# Replace with:
+{% if cookiecutter.is_encoder_decoder_model == "False" -%}
+        ({{cookiecutter.camelcase_modelname}}Config, {{cookiecutter.camelcase_modelname}}ForTokenClassification),
+{% else -%}
+{% endif -%}
+# End.
+
+# Below: "# Model for Multiple Choice mapping"
+# Replace with:
+{% if cookiecutter.is_encoder_decoder_model == "False" -%}
+        ({{cookiecutter.camelcase_modelname}}Config, {{cookiecutter.camelcase_modelname}}ForMultipleChoice),
+{% else -%}
+{% endif -%}
+# End.
+
+
+# Below: "# Model for Seq2Seq Causal LM mapping"
+# Replace with:
+{% if cookiecutter.is_encoder_decoder_model == "False" -%}
+{% else -%}
+        ({{cookiecutter.camelcase_modelname}}Config, {{cookiecutter.camelcase_modelname}}ForConditionalGeneration),
+{% endif -%}
+# End.
+
+# To replace in: "src/transformers/modeling_tf_auto.py" if generating TensorFlow
+# Below: "from .configuration_auto import ("
+# Replace with:
+    {{cookiecutter.camelcase_modelname}}Config,
+# End.
+
+# Below: "# Add modeling imports here"
+# Replace with:
+
+{% if cookiecutter.is_encoder_decoder_model == "False" -%}
+from .modeling_tf_{{cookiecutter.lowercase_modelname}} import (
+    TF{{cookiecutter.camelcase_modelname}}ForMaskedLM,
+    TF{{cookiecutter.camelcase_modelname}}ForMultipleChoice,
+    TF{{cookiecutter.camelcase_modelname}}ForQuestionAnswering,
+    TF{{cookiecutter.camelcase_modelname}}ForSequenceClassification,
+    TF{{cookiecutter.camelcase_modelname}}ForTokenClassification,
+    TF{{cookiecutter.camelcase_modelname}}Model,
+)
+{% else -%}
+from .modeling_tf_{{cookiecutter.lowercase_modelname}} import (
+    TF{{cookiecutter.camelcase_modelname}}ForConditionalGeneration,
+    TF{{cookiecutter.camelcase_modelname}}Model,
+)
+{% endif -%}
+# End.
+
+# Below: "# Base model mapping"
+# Replace with:
+        ({{cookiecutter.camelcase_modelname}}Config, TF{{cookiecutter.camelcase_modelname}}Model),
+# End.
+
+# Below: "# Model with LM heads mapping"
+# Replace with:
+{% if cookiecutter.is_encoder_decoder_model == "False" -%}
+        ({{cookiecutter.camelcase_modelname}}Config, TF{{cookiecutter.camelcase_modelname}}ForMaskedLM),
+{% else -%}
+        ({{cookiecutter.camelcase_modelname}}Config, TF{{cookiecutter.camelcase_modelname}}ForConditionalGeneration),
+{% endif -%}
+# End.
+
+# Below: "# Model for Masked LM mapping"
+# Replace with:
+{% if cookiecutter.is_encoder_decoder_model == "False" -%}
+        ({{cookiecutter.camelcase_modelname}}Config, TF{{cookiecutter.camelcase_modelname}}ForMaskedLM),
+{% else -%}
+{% endif -%}
+# End.
+
+# Below: "# Model for Sequence Classification mapping"
+# Replace with:
+{% if cookiecutter.is_encoder_decoder_model == "False" -%}
+        ({{cookiecutter.camelcase_modelname}}Config, TF{{cookiecutter.camelcase_modelname}}ForSequenceClassification),
+{% else -%}
+{% endif -%}
+# End.
+
+# Below: "# Model for Question Answering mapping"
+# Replace with:
+{% if cookiecutter.is_encoder_decoder_model == "False" -%}
+        ({{cookiecutter.camelcase_modelname}}Config, TF{{cookiecutter.camelcase_modelname}}ForQuestionAnswering),
+{% else -%}
+{% endif -%}
+# End.
+
+# Below: "# Model for Token Classification mapping"
+# Replace with:
+{% if cookiecutter.is_encoder_decoder_model == "False" -%}
+        ({{cookiecutter.camelcase_modelname}}Config, TF{{cookiecutter.camelcase_modelname}}ForTokenClassification),
+{% else -%}
+{% endif -%}
+# End.
+
+# Below: "# Model for Multiple Choice mapping"
+# Replace with:
+{% if cookiecutter.is_encoder_decoder_model == "False" -%}
+        ({{cookiecutter.camelcase_modelname}}Config, TF{{cookiecutter.camelcase_modelname}}ForMultipleChoice),
+{% else -%}
+{% endif -%}
+# End.
+
+
+# Below: "# Model for Seq2Seq Causal LM mapping"
+# Replace with:
+{% if cookiecutter.is_encoder_decoder_model == "False" -%}
+{% else -%}
+        ({{cookiecutter.camelcase_modelname}}Config, TF{{cookiecutter.camelcase_modelname}}ForConditionalGeneration),
+{% endif -%}
+# End.

templates/cookiecutter/cookiecutter-template-{{cookiecutter.modelname}}/tokenization_{{cookiecutter.lowercase_modelname}}.py

@@ -0,0 +1,342 @@
+# coding=utf-8
+# Copyright 2018 The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Tokenization classes for {{cookiecutter.modelname}}."""
+
+{%- if cookiecutter.tokenizer_type == "Based on BERT" %}
+from .tokenization_bert import BertTokenizer, BertTokenizerFast
+from .utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+VOCAB_FILES_NAMES = {"vocab_file": "vocab.txt"}
+
+PRETRAINED_VOCAB_FILES_MAP = {
+    "vocab_file": {
+        "{{cookiecutter.checkpoint_identifier}}": "https://s3.amazonaws.com/models.huggingface.co/bert/{{cookiecutter.checkpoint_identifier}}/vocab.txt",
+    }
+}
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    "{{cookiecutter.checkpoint_identifier}}": 512,
+}
+
+
+PRETRAINED_INIT_CONFIGURATION = {
+    "{{cookiecutter.checkpoint_identifier}}": {"do_lower_case": False},
+}
+
+
+class {{cookiecutter.camelcase_modelname}}Tokenizer(BertTokenizer):
+    r"""
+    Construct a {{cookiecutter.modelname}} tokenizer.
+
+    :class:`~transformers.{{cookiecutter.camelcase_modelname}}Tokenizer` is identical to :class:`~transformers.BertTokenizer` and runs end-to-end
+    tokenization: punctuation splitting and wordpiece.
+
+    Refer to superclass :class:`~transformers.BertTokenizer` for usage examples and documentation concerning
+    parameters.
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+    pretrained_init_configuration = PRETRAINED_INIT_CONFIGURATION
+
+
+class {{cookiecutter.camelcase_modelname}}TokenizerFast(BertTokenizerFast):
+    r"""
+    Construct a "fast" {{cookiecutter.modelname}} tokenizer (backed by HuggingFace's `tokenizers` library).
+
+    :class:`~transformers.{{cookiecutter.camelcase_modelname}}TokenizerFast` is identical to :class:`~transformers.BertTokenizerFast` and runs
+    end-to-end tokenization: punctuation splitting and wordpiece.
+
+    Refer to superclass :class:`~transformers.BertTokenizerFast` for usage examples and documentation concerning
+    parameters.
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+    pretrained_init_configuration = PRETRAINED_INIT_CONFIGURATION
+{%- elif cookiecutter.tokenizer_type == "Standalone" %}
+import warnings
+
+from tokenizers import ByteLevelBPETokenizer
+
+from .tokenization_utils import AddedToken, PreTrainedTokenizer
+from .tokenization_utils_base import BatchEncoding
+from .tokenization_utils_fast import PreTrainedTokenizerFast
+from typing import List, Optional
+from .utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+VOCAB_FILES_NAMES = {}
+
+PRETRAINED_VOCAB_FILES_MAP = {}
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    "{{cookiecutter.checkpoint_identifier}}": 1024,
+}
+
+class {{cookiecutter.camelcase_modelname}}Tokenizer(PreTrainedTokenizer):
+    """
+    Construct a {{cookiecutter.modelname}} tokenizer. Based on byte-level Byte-Pair-Encoding.
+
+    Args:
+        vocab_file (:obj:`str`):
+            Path to the vocabulary file.
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+    model_input_names = ["attention_mask"]
+
+    def __init__(
+            self,
+            vocab_file,
+            unk_token="<|endoftext|>",
+            bos_token="<|endoftext|>",
+            eos_token="<|endoftext|>",
+            **kwargs
+    ):
+        bos_token = AddedToken(bos_token, lstrip=False, rstrip=False) if isinstance(bos_token, str) else bos_token
+        eos_token = AddedToken(eos_token, lstrip=False, rstrip=False) if isinstance(eos_token, str) else eos_token
+        unk_token = AddedToken(unk_token, lstrip=False, rstrip=False) if isinstance(unk_token, str) else unk_token
+        super().__init__(bos_token=bos_token, eos_token=eos_token, unk_token=unk_token, **kwargs)
+
+        "Initialisation"
+
+    @property
+    def vocab_size(self):
+        "Returns vocab size"
+
+    def get_vocab(self):
+        "Returns vocab as a dict"
+
+    def _tokenize(self, text):
+        """ Returns a tokenized string. """
+
+    def _convert_token_to_id(self, token):
+        """ Converts a token (str) in an id using the vocab. """
+
+    def _convert_id_to_token(self, index):
+        """Converts an index (integer) in a token (str) using the vocab."""
+
+    def convert_tokens_to_string(self, tokens):
+        """ Converts a sequence of tokens (string) in a single string. """
+
+    def save_vocabulary(self, save_directory):
+        """
+        Save the vocabulary and special tokens file to a directory.
+
+        Args:
+            save_directory (:obj:`str`):
+                The directory in which to save the vocabulary.
+
+        Returns:
+            :obj:`Tuple(str)`: Paths to the files saved.
+        """
+
+    def build_inputs_with_special_tokens(
+            self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
+    ) -> List[int]:
+        """
+        Build model inputs from a sequence or a pair of sequence for sequence classification tasks
+        by concatenating and adding special tokens.
+        A {{cookiecutter.modelname}} sequence has the following format:
+
+        - single sequence: ``<s> X </s>``
+        - pair of sequences: ``<s> A </s></s> B </s>``
+
+        Args:
+            token_ids_0 (:obj:`List[int]`):
+                List of IDs to which the special tokens will be added.
+            token_ids_1 (:obj:`List[int]`, `optional`):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            :obj:`List[int]`: List of `input IDs <../glossary.html#input-ids>`__ with the appropriate special tokens.
+        """
+        if token_ids_1 is None:
+            return [self.cls_token_id] + token_ids_0 + [self.sep_token_id]
+        cls = [self.cls_token_id]
+        sep = [self.sep_token_id]
+        return cls + token_ids_0 + sep + sep + token_ids_1 + sep
+
+    def get_special_tokens_mask(
+            self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None, already_has_special_tokens: bool = False
+    ) -> List[int]:
+        """
+        Retrieve sequence ids from a token list that has no special tokens added. This method is called when adding
+        special tokens using the tokenizer ``prepare_for_model`` method.
+
+        Args:
+            token_ids_0 (:obj:`List[int]`):
+                List of IDs.
+            token_ids_1 (:obj:`List[int]`, `optional`):
+                Optional second list of IDs for sequence pairs.
+            already_has_special_tokens (:obj:`bool`, `optional`, defaults to :obj:`False`):
+                Whether or not the token list is already formatted with special tokens for the model.
+
+        Returns:
+            :obj:`List[int]`: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
+        """
+        if already_has_special_tokens:
+            if token_ids_1 is not None:
+                raise ValueError(
+                    "You should not supply a second sequence if the provided sequence of "
+                    "ids is already formatted with special tokens for the model."
+                )
+            return list(map(lambda x: 1 if x in [self.sep_token_id, self.cls_token_id] else 0, token_ids_0))
+
+        if token_ids_1 is None:
+            return [1] + ([0] * len(token_ids_0)) + [1]
+        return [1] + ([0] * len(token_ids_0)) + [1, 1] + ([0] * len(token_ids_1)) + [1]
+
+    def create_token_type_ids_from_sequences(
+            self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
+    ) -> List[int]:
+        """
+        Create a mask from the two sequences passed to be used in a sequence-pair classification task.
+        {{cookiecutter.modelname}} does not make use of token type ids, therefore a list of zeros is returned.
+
+        Args:
+            token_ids_0 (:obj:`List[int]`):
+                List of IDs.
+            token_ids_1 (:obj:`List[int]`, `optional`):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            :obj:`List[int]`:  List of zeros.
+        """
+        sep = [self.sep_token_id]
+        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 + sep + token_ids_1 + sep) * [0]
+
+    def prepare_for_tokenization(self, text, is_split_into_words=False, **kwargs):
+        if "is_pretokenized" in kwargs:
+            warnings.warn(
+                "`is_pretokenized` is deprecated and will be removed in a future version, use `is_split_into_words` instead.",
+                FutureWarning,
+            )
+            is_split_into_words = kwargs.pop("is_pretokenized")
+
+        add_prefix_space = kwargs.pop("add_prefix_space", self.add_prefix_space)
+        if (is_split_into_words or add_prefix_space) and (len(text) > 0 and not text[0].isspace()):
+            text = " " + text
+        return (text, kwargs)
+
+class {{cookiecutter.camelcase_modelname}}TokenizerFast(PreTrainedTokenizerFast):
+    """
+    Construct a "fast" {{cookiecutter.modelname}} tokenizer (backed by HuggingFace's `tokenizers` library).
+
+    Args:
+        vocab_file (:obj:`str`):
+            Path to the vocabulary file.
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+    model_input_names = ["attention_mask"]
+
+    def __init__(
+            self,
+            vocab_file,
+            merges_file,
+            unk_token="<|endoftext|>",
+            bos_token="<|endoftext|>",
+            eos_token="<|endoftext|>",
+            add_prefix_space=False,
+            trim_offsets=True,
+            **kwargs
+    ):
+        super().__init__(
+            ByteLevelBPETokenizer(
+                vocab_file=vocab_file,
+                merges_file=merges_file,
+                add_prefix_space=add_prefix_space,
+                trim_offsets=trim_offsets,
+            ),
+            bos_token=bos_token,
+            eos_token=eos_token,
+            unk_token=unk_token,
+            **kwargs,
+        )
+        self.add_prefix_space = add_prefix_space
+
+    def _batch_encode_plus(self, *args, **kwargs) -> BatchEncoding:
+        is_split_into_words = None
+        if "is_pretokenized" in kwargs:
+            warnings.warn(
+                "`is_pretokenized` is deprecated and will be removed in a future version, use `is_split_into_words` instead.",
+                FutureWarning,
+            )
+            is_split_into_words = kwargs.pop("is_pretokenized")
+
+        is_split_into_words = kwargs.get("is_split_into_words", False) if is_split_into_words is None else is_split_into_words
+        return super()._batch_encode_plus(*args, **kwargs)
+
+    def _encode_plus(self, *args, **kwargs) -> BatchEncoding:
+        is_split_into_words = None
+        if "is_pretokenized" in kwargs:
+            warnings.warn(
+                "`is_pretokenized` is deprecated and will be removed in a future version, use `is_split_into_words` instead.",
+                FutureWarning,
+            )
+        is_split_into_words = kwargs.get("is_split_into_words", False) if is_split_into_words is None else is_split_into_words
+        return super()._encode_plus(*args, **kwargs)
+
+
+    def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
+        output = [self.bos_token_id] + token_ids_0 + [self.eos_token_id]
+        if token_ids_1 is None:
+            return output
+
+        return output + [self.eos_token_id] + token_ids_1 + [self.eos_token_id]
+
+
+    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.
+        {{cookiecutter.modelname}} does not make use of token type ids, therefore a list of zeros is returned.
+
+        Args:
+            token_ids_0 (:obj:`List[int]`):
+                List of IDs.
+            token_ids_1 (:obj:`List[int]`, `optional`):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            :obj:`List[int]`:  List of zeros.
+        """
+        sep = [self.sep_token_id]
+        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 + sep + token_ids_1 + sep) * [0]
+
+
+{% endif %}

templates/cookiecutter/cookiecutter-template-{{cookiecutter.modelname}}/{{cookiecutter.lowercase_modelname}}.rst

@@ -0,0 +1,179 @@
+{{cookiecutter.uppercase_modelname}}
+-----------------------------------------------------------------------------------------------------------------------
+
+Overview
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The {{cookiecutter.modelname}} model was proposed in `<INSERT PAPER NAME HERE>
+<<INSERT PAPER LINK HERE>>`__  by <INSERT AUTHORS HERE>. <INSERT SHORT SUMMARY HERE>
+
+The abstract from the paper is the following:
+
+*<INSERT PAPER ABSTRACT HERE>*
+
+Tips:
+
+<INSERT TIPS ABOUT MODEL HERE>
+
+{{cookiecutter.camelcase_modelname}}Config
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.{{cookiecutter.camelcase_modelname}}Config
+    :members:
+
+
+{{cookiecutter.camelcase_modelname}}Tokenizer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.{{cookiecutter.camelcase_modelname}}Tokenizer
+    :members: build_inputs_with_special_tokens, get_special_tokens_mask,
+        create_token_type_ids_from_sequences, save_vocabulary
+
+
+{{cookiecutter.camelcase_modelname}}TokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.{{cookiecutter.camelcase_modelname}}TokenizerFast
+    :members: build_inputs_with_special_tokens, get_special_tokens_mask,
+        create_token_type_ids_from_sequences, save_vocabulary
+
+
+{% if cookiecutter.is_encoder_decoder_model == "False" -%}
+{% if "PyTorch" in cookiecutter.generate_tensorflow_and_pytorch -%}
+{{cookiecutter.camelcase_modelname}}Model
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.{{cookiecutter.camelcase_modelname}}Model
+    :members: forward
+
+
+{{cookiecutter.camelcase_modelname}}ForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.{{cookiecutter.camelcase_modelname}}ForMaskedLM
+    :members: forward
+
+
+{{cookiecutter.camelcase_modelname}}ForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.{{cookiecutter.camelcase_modelname}}ForSequenceClassification
+    :members: forward
+
+
+{{cookiecutter.camelcase_modelname}}ForMultipleChoice
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.{{cookiecutter.camelcase_modelname}}ForMultipleChoice
+    :members:
+
+
+{{cookiecutter.camelcase_modelname}}ForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.{{cookiecutter.camelcase_modelname}}ForTokenClassification
+    :members: forward
+
+
+{{cookiecutter.camelcase_modelname}}ForQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.{{cookiecutter.camelcase_modelname}}ForQuestionAnswering
+    :members: forward
+
+{% endif -%}
+{% if "TensorFlow" in cookiecutter.generate_tensorflow_and_pytorch -%}
+
+TF{{cookiecutter.camelcase_modelname}}Model
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TF{{cookiecutter.camelcase_modelname}}Model
+    :members: call
+
+
+TF{{cookiecutter.camelcase_modelname}}ForMaskedLM
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TF{{cookiecutter.camelcase_modelname}}ForMaskedLM
+    :members: call
+
+
+TF{{cookiecutter.camelcase_modelname}}ForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TF{{cookiecutter.camelcase_modelname}}ForSequenceClassification
+    :members: call
+
+
+TF{{cookiecutter.camelcase_modelname}}ForMultipleChoice
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TF{{cookiecutter.camelcase_modelname}}ForMultipleChoice
+    :members: call
+
+
+TF{{cookiecutter.camelcase_modelname}}ForTokenClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TF{{cookiecutter.camelcase_modelname}}ForTokenClassification
+    :members: call
+
+
+TF{{cookiecutter.camelcase_modelname}}ForQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TF{{cookiecutter.camelcase_modelname}}ForQuestionAnswering
+    :members: call
+
+
+{% endif -%}
+{% else -%}
+{% if "PyTorch" in cookiecutter.generate_tensorflow_and_pytorch -%}
+
+{{cookiecutter.camelcase_modelname}}Model
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.{{cookiecutter.camelcase_modelname}}Model
+    :members: forward
+
+.. autofunction:: transformers.modeling_{{cookiecutter.lowercase_modelname}}._prepare_{{cookiecutter.lowercase_modelname}}_decoder_inputs
+
+
+{{cookiecutter.camelcase_modelname}}ForConditionalGeneration
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.{{cookiecutter.camelcase_modelname}}ForConditionalGeneration
+    :members: forward
+
+
+{{cookiecutter.camelcase_modelname}}ForSequenceClassification
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.{{cookiecutter.camelcase_modelname}}ForSequenceClassification
+    :members: forward
+
+
+{{cookiecutter.camelcase_modelname}}ForQuestionAnswering
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.{{cookiecutter.camelcase_modelname}}ForQuestionAnswering
+    :members: forward
+
+
+{% endif -%}
+{% if "TensorFlow" in cookiecutter.generate_tensorflow_and_pytorch -%}
+
+TF{{cookiecutter.camelcase_modelname}}Model
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TF{{cookiecutter.camelcase_modelname}}Model
+    :members: call
+
+
+TF{{cookiecutter.camelcase_modelname}}ForConditionalGeneration
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.TF{{cookiecutter.camelcase_modelname}}ForConditionalGeneration
+    :members: call
+{% endif -%}
+{% endif -%}

templates/cookiecutter/cookiecutter.json

@@ -0,0 +1,11 @@
+{
+  "modelname": "BrandNewBERT",
+  "uppercase_modelname": "BRAND_NEW_BERT",
+  "lowercase_modelname": "brand_new_bert",
+  "camelcase_modelname": "BrandNewBert",
+  "authors": "The HuggingFace Team",
+  "checkpoint_identifier": "brand-new-bert-base-cased",
+  "tokenizer_type": ["Based on BERT", "Standalone"],
+  "is_encoder_decoder_model": ["True", "False"],
+  "generate_tensorflow_and_pytorch": ["PyTorch & TensorFlow", "PyTorch", "TensorFlow"]
+}

templates/cookiecutter/tests/encoder-bert-tokenizer.json

@@ -0,0 +1,11 @@
+{
+  "modelname": "EncoderBERT",
+  "uppercase_modelname": "ENCODER_BERT",
+  "lowercase_modelname": "encoder_bert",
+  "camelcase_modelname": "EncoderBert",
+  "authors": "The HuggingFace Team",
+  "checkpoint_identifier": "brand-new-bert-base-cased",
+  "tokenizer_type": "Based on BERT",
+  "is_encoder_decoder_model": "False",
+  "generate_tensorflow_and_pytorch": "PyTorch & TensorFlow"
+}

templates/cookiecutter/tests/pt-encoder-bert-tokenizer.json

@@ -0,0 +1,11 @@
+{
+  "modelname": "PTEncoderBERT",
+  "uppercase_modelname": "PT_ENCODER_BERT",
+  "lowercase_modelname": "pt_encoder_bert",
+  "camelcase_modelname": "PtEncoderBert",
+  "authors": "The HuggingFace Team",
+  "checkpoint_identifier": "brand-new-bert-base-cased",
+  "tokenizer_type": "Based on BERT",
+  "is_encoder_decoder_model": "False",
+  "generate_tensorflow_and_pytorch": "PyTorch"
+}

templates/cookiecutter/tests/seq-to-seq-roberta-tokenizer.json

@@ -0,0 +1,11 @@
+{
+  "modelname": "SeqToSeqRoBERTa",
+  "uppercase_modelname": "SEQ_TO_SEQ_ROBERTA",
+  "lowercase_modelname": "seq_to_seq_roberta",
+  "camelcase_modelname": "SeqToSeqRoberta",
+  "authors": "The HuggingFace Team",
+  "checkpoint_identifier": "brand-new-roberta-base-cased",
+  "tokenizer_type": "Based on BERT",
+  "is_encoder_decoder_model": "True",
+  "generate_tensorflow_and_pytorch": "PyTorch & TensorFlow"
+}

templates/cookiecutter/tests/standalone.json

@@ -0,0 +1,11 @@
+{
+  "modelname": "BIEncoderBERT",
+  "uppercase_modelname": "BI_ENCODER_BERT",
+  "lowercase_modelname": "bi_encoder_bert",
+  "camelcase_modelname": "BiEncoderBert",
+  "authors": "The HuggingFace Team",
+  "checkpoint_identifier": "bi-brand-new-bert-base-cased",
+  "tokenizer_type": "Standalone",
+  "is_encoder_decoder_model": "False",
+  "generate_tensorflow_and_pytorch": "PyTorch & TensorFlow"
+}

templates/cookiecutter/tests/tf-encoder-bert-tokenizer.json

@@ -0,0 +1,11 @@
+{
+  "modelname": "TFEncoderBERT",
+  "uppercase_modelname": "TF_ENCODER_BERT",
+  "lowercase_modelname": "tf_encoder_bert",
+  "camelcase_modelname": "TfEncoderBert",
+  "authors": "The HuggingFace Team",
+  "checkpoint_identifier": "brand-new-bert-base-cased",
+  "tokenizer_type": "Based on BERT",
+  "is_encoder_decoder_model": "False",
+  "generate_tensorflow_and_pytorch": "TensorFlow"
+}

utils/check_repo.py

@@ -245,20 +245,9 @@ def check_models_are_documented(module, doc_file):
 def _get_model_name(module):
     """ Get the model name for the module defining it."""
     splits = module.__name__.split("_")
+    splits = splits[(2 if splits[1] in ["flax", "tf"] else 1) :]
 
-    # Secial case for transfo_xl
-    if splits[-1] == "xl":
-        return "_".join(splits[-2:])
-    # Special case for xlm_prophetnet
-    if splits[-1] == "prophetnet" and splits[-2] == "xlm":
-        return "_".join(splits[-2:])
-    # Secial case for xlm_roberta
-    if splits[-1] == "roberta" and splits[-2] == "xlm":
-        return "_".join(splits[-2:])
-    # Special case for bert_generation
-    if splits[-1] == "generation" and splits[-2] == "bert":
-        return "_".join(splits[-2:])
-    return splits[-1]
+    return "_".join(splits)
 
 
 def check_all_models_are_documented():