Release: v4.22.1

* Fix tokenizer load from one file

* Add a test

* Style

Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>

.circleci/TROUBLESHOOT.md

@@ -1,6 +1,6 @@
 # Troubleshooting
 
-This is a document explaining how to deal with various issues on Circle-CI. The entries may include actual solutions or pointers to Issues that cover those.
+This is a document explaining how to deal with various issues on Circle-CI. The entries may include actually solutions or pointers to Issues that cover those.
 
 ## Circle CI
 

.circleci/create_circleci_config.py

@@ -1,547 +0,0 @@
-# coding=utf-8
-# Copyright 2022 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.
-
-import argparse
-import copy
-import os
-import random
-from dataclasses import dataclass
-from typing import Any, Dict, List, Optional
-import glob
-import yaml
-
-
-COMMON_ENV_VARIABLES = {
-    "OMP_NUM_THREADS": 1,
-    "TRANSFORMERS_IS_CI": True,
-    "PYTEST_TIMEOUT": 120,
-    "RUN_PIPELINE_TESTS": False,
-    "RUN_PT_TF_CROSS_TESTS": False,
-    "RUN_PT_FLAX_CROSS_TESTS": False,
-}
-# Disable the use of {"s": None} as the output is way too long, causing the navigation on CircleCI impractical
-COMMON_PYTEST_OPTIONS = {"max-worker-restart": 0, "dist": "loadfile", "v": None}
-DEFAULT_DOCKER_IMAGE = [{"image": "cimg/python:3.8.12"}]
-
-
-class EmptyJob:
-    job_name = "empty"
-
-    def to_dict(self):
-        return {
-            "docker": copy.deepcopy(DEFAULT_DOCKER_IMAGE),
-            "steps":["checkout"],
-        }
-
-
-@dataclass
-class CircleCIJob:
-    name: str
-    additional_env: Dict[str, Any] = None
-    cache_name: str = None
-    cache_version: str = "0.8.2"
-    docker_image: List[Dict[str, str]] = None
-    install_steps: List[str] = None
-    marker: Optional[str] = None
-    parallelism: Optional[int] = 1
-    pytest_num_workers: int = 12
-    pytest_options: Dict[str, Any] = None
-    resource_class: Optional[str] = "2xlarge"
-    tests_to_run: Optional[List[str]] = None
-    # This should be only used for doctest job!
-    command_timeout: Optional[int] = None
-
-    def __post_init__(self):
-        # Deal with defaults for mutable attributes.
-        if self.additional_env is None:
-            self.additional_env = {}
-        if self.cache_name is None:
-            self.cache_name = self.name
-        if self.docker_image is None:
-            # Let's avoid changing the default list and make a copy.
-            self.docker_image = copy.deepcopy(DEFAULT_DOCKER_IMAGE)
-        else:
-            # BIG HACK WILL REMOVE ONCE FETCHER IS UPDATED
-            print(os.environ.get("GIT_COMMIT_MESSAGE"))
-            if "[build-ci-image]" in os.environ.get("GIT_COMMIT_MESSAGE", "") or os.environ.get("GIT_COMMIT_MESSAGE", "") == "dev-ci":
-                self.docker_image[0]["image"] = f"{self.docker_image[0]['image']}:dev"
-            print(f"Using {self.docker_image} docker image")
-        if self.install_steps is None:
-            self.install_steps = []
-        if self.pytest_options is None:
-            self.pytest_options = {}
-        if isinstance(self.tests_to_run, str):
-            self.tests_to_run = [self.tests_to_run]
-        if self.parallelism is None:
-            self.parallelism = 1
-
-    def to_dict(self):
-        env = COMMON_ENV_VARIABLES.copy()
-        env.update(self.additional_env)
-
-        cache_branch_prefix = os.environ.get("CIRCLE_BRANCH", "pull")
-        if cache_branch_prefix != "main":
-            cache_branch_prefix = "pull"
-
-        job = {
-            "docker": self.docker_image,
-            "environment": env,
-        }
-        if self.resource_class is not None:
-            job["resource_class"] = self.resource_class
-        if self.parallelism is not None:
-            job["parallelism"] = self.parallelism
-        steps = [
-            "checkout",
-            {"attach_workspace": {"at": "test_preparation"}},
-        ]
-        steps.extend([{"run": l} for l in self.install_steps])
-        steps.append({"run": {"name": "Show installed libraries and their size", "command": """du -h -d 1 "$(pip -V | cut -d ' ' -f 4 | sed 's/pip//g')" | grep -vE "dist-info|_distutils_hack|__pycache__" | sort -h | tee installed.txt || true"""}})
-        steps.append({"run": {"name": "Show installed libraries and their versions", "command": """pip list --format=freeze | tee installed.txt || true"""}})
-
-        steps.append({"run":{"name":"Show biggest libraries","command":"""dpkg-query --show --showformat='${Installed-Size}\t${Package}\n' | sort -rh | head -25 | sort -h | awk '{ package=$2; sub(".*/", "", package); printf("%.5f GB %s\n", $1/1024/1024, package)}' || true"""}})
-        steps.append({"store_artifacts": {"path": "installed.txt"}})
-
-        all_options = {**COMMON_PYTEST_OPTIONS, **self.pytest_options}
-        pytest_flags = [f"--{key}={value}" if (value is not None or key in ["doctest-modules"]) else f"-{key}" for key, value in all_options.items()]
-        pytest_flags.append(
-            f"--make-reports={self.name}" if "examples" in self.name else f"--make-reports=tests_{self.name}"
-        )
-
-        steps.append({"run": {"name": "Create `test-results` directory", "command": "mkdir test-results"}})
-        test_command = ""
-        if self.command_timeout:
-            test_command = f"timeout {self.command_timeout} "
-        # junit familiy xunit1 is necessary to support splitting on test name or class name with circleci split
-        test_command += f"python3 -m pytest -rsfE -p no:warnings -o junit_family=xunit1 --tb=short --junitxml=test-results/junit.xml -n {self.pytest_num_workers} " + " ".join(pytest_flags)
-
-        if self.parallelism == 1:
-            if self.tests_to_run is None:
-                test_command += " << pipeline.parameters.tests_to_run >>"
-            else:
-                test_command += " " + " ".join(self.tests_to_run)
-        else:
-            # We need explicit list instead of `pipeline.parameters.tests_to_run` (only available at job runtime)
-            tests = self.tests_to_run
-            if tests is None:
-                folder = os.environ["test_preparation_dir"]
-                test_file = os.path.join(folder, "filtered_test_list.txt")
-                if os.path.exists(test_file): # We take this job's tests from the filtered test_list.txt
-                    with open(test_file) as f:
-                        tests = f.read().split(" ")
-
-            # expand the test list
-            if tests == ["tests"]:
-                tests = [os.path.join("tests", x) for x in os.listdir("tests")]
-            expanded_tests = []
-            for test in tests:
-                if test.endswith(".py"):
-                    expanded_tests.append(test)
-                elif test == "tests/models":
-                    if "tokenization" in self.name:
-                        expanded_tests.extend(glob.glob("tests/models/**/test_tokenization*.py", recursive=True))
-                    elif self.name in ["flax","torch","tf"]:
-                        name = self.name if self.name != "torch" else ""
-                        if self.name == "torch":
-                            all_tests = glob.glob(f"tests/models/**/test_modeling_{name}*.py", recursive=True)
-                            filtered = [k for k in all_tests if ("_tf_") not in k and "_flax_" not in k]
-                            expanded_tests.extend(filtered)
-                        else:
-                            expanded_tests.extend(glob.glob(f"tests/models/**/test_modeling_{name}*.py", recursive=True))
-                    else:
-                        expanded_tests.extend(glob.glob("tests/models/**/test_modeling*.py", recursive=True))
-                elif test == "tests/pipelines":
-                    expanded_tests.extend(glob.glob("tests/models/**/test_modeling*.py", recursive=True))
-                else:
-                    expanded_tests.append(test)
-            tests = " ".join(expanded_tests)
-
-            # Each executor to run ~10 tests
-            n_executors = max(len(expanded_tests) // 10, 1)
-            # Avoid empty test list on some executor(s) or launching too many executors
-            if n_executors > self.parallelism:
-                n_executors = self.parallelism
-            job["parallelism"] = n_executors
-
-            # Need to be newline separated for the command `circleci tests split` below
-            command = f'echo {tests} | tr " " "\\n" >> tests.txt'
-            steps.append({"run": {"name": "Get tests", "command": command}})
-
-            command = 'TESTS=$(circleci tests split tests.txt) && echo $TESTS > splitted_tests.txt'
-            steps.append({"run": {"name": "Split tests", "command": command}})
-
-            steps.append({"store_artifacts": {"path": "tests.txt"}})
-            steps.append({"store_artifacts": {"path": "splitted_tests.txt"}})
-
-            test_command = ""
-            if self.command_timeout:
-                test_command = f"timeout {self.command_timeout} "
-            test_command += f"python3 -m pytest -rsfE -p no:warnings --tb=short  -o junit_family=xunit1 --junitxml=test-results/junit.xml -n {self.pytest_num_workers} " + " ".join(pytest_flags)
-            test_command += " $(cat splitted_tests.txt)"
-        if self.marker is not None:
-            test_command += f" -m {self.marker}"
-
-        if self.name == "pr_documentation_tests":
-            # can't use ` | tee tee tests_output.txt` as usual
-            test_command += " > tests_output.txt"
-            # Save the return code, so we can check if it is timeout in the next step.
-            test_command += '; touch "$?".txt'
-            # Never fail the test step for the doctest job. We will check the results in the next step, and fail that
-            # step instead if the actual test failures are found. This is to avoid the timeout being reported as test
-            # failure.
-            test_command = f"({test_command}) || true"
-        else:
-            test_command = f"({test_command} | tee tests_output.txt)"
-        steps.append({"run": {"name": "Run tests", "command": test_command}})
-
-        steps.append({"run": {"name": "Skipped tests", "when": "always", "command": f"python3 .circleci/parse_test_outputs.py --file tests_output.txt --skip"}})
-        steps.append({"run": {"name": "Failed tests",  "when": "always", "command": f"python3 .circleci/parse_test_outputs.py --file tests_output.txt --fail"}})
-        steps.append({"run": {"name": "Errors",        "when": "always", "command": f"python3 .circleci/parse_test_outputs.py --file tests_output.txt --errors"}})
-
-        steps.append({"store_test_results": {"path": "test-results"}})
-        steps.append({"store_artifacts": {"path": "tests_output.txt"}})
-        steps.append({"store_artifacts": {"path": "test-results/junit.xml"}})
-        steps.append({"store_artifacts": {"path": "reports"}})
-
-        job["steps"] = steps
-        return job
-
-    @property
-    def job_name(self):
-        return self.name if "examples" in self.name else f"tests_{self.name}"
-
-
-# JOBS
-torch_and_tf_job = CircleCIJob(
-    "torch_and_tf",
-    docker_image=[{"image":"huggingface/transformers-torch-tf-light"}],
-    install_steps=["uv venv && uv pip install ."],
-    additional_env={"RUN_PT_TF_CROSS_TESTS": True},
-    marker="is_pt_tf_cross_test",
-    pytest_options={"rA": None, "durations": 0},
-)
-
-
-torch_and_flax_job = CircleCIJob(
-    "torch_and_flax",
-    additional_env={"RUN_PT_FLAX_CROSS_TESTS": True},
-    docker_image=[{"image":"huggingface/transformers-torch-jax-light"}],
-    install_steps=["uv venv && uv pip install ."],
-    marker="is_pt_flax_cross_test",
-    pytest_options={"rA": None, "durations": 0},
-)
-
-torch_job = CircleCIJob(
-    "torch",
-    docker_image=[{"image": "huggingface/transformers-torch-light"}],
-    install_steps=["uv venv && uv pip install ."],
-    parallelism=6,
-    pytest_num_workers=4
-)
-
-tokenization_job = CircleCIJob(
-    "tokenization",
-    docker_image=[{"image": "huggingface/transformers-torch-light"}],
-    install_steps=["uv venv && uv pip install ."],
-    parallelism=6,
-    pytest_num_workers=4
-)
-
-
-tf_job = CircleCIJob(
-    "tf",
-    docker_image=[{"image":"huggingface/transformers-tf-light"}],
-    install_steps=["uv venv", "uv pip install -e."],
-    parallelism=6,
-    pytest_num_workers=4,
-)
-
-
-flax_job = CircleCIJob(
-    "flax",
-    docker_image=[{"image":"huggingface/transformers-jax-light"}],
-    install_steps=["uv venv && uv pip install ."],
-    parallelism=6,
-    pytest_num_workers=4
-)
-
-
-pipelines_torch_job = CircleCIJob(
-    "pipelines_torch",
-    additional_env={"RUN_PIPELINE_TESTS": True},
-    docker_image=[{"image":"huggingface/transformers-torch-light"}],
-    install_steps=["uv venv && uv pip install ."],
-    marker="is_pipeline_test",
-)
-
-
-pipelines_tf_job = CircleCIJob(
-    "pipelines_tf",
-    additional_env={"RUN_PIPELINE_TESTS": True},
-    docker_image=[{"image":"huggingface/transformers-tf-light"}],
-    install_steps=["uv venv && uv pip install ."],
-    marker="is_pipeline_test",
-)
-
-
-custom_tokenizers_job = CircleCIJob(
-    "custom_tokenizers",
-    additional_env={"RUN_CUSTOM_TOKENIZERS": True},
-    docker_image=[{"image": "huggingface/transformers-custom-tokenizers"}],
-    install_steps=["uv venv","uv pip install -e ."],
-    parallelism=None,
-    resource_class=None,
-    tests_to_run=[
-        "./tests/models/bert_japanese/test_tokenization_bert_japanese.py",
-        "./tests/models/openai/test_tokenization_openai.py",
-        "./tests/models/clip/test_tokenization_clip.py",
-    ],
-)
-
-
-examples_torch_job = CircleCIJob(
-    "examples_torch",
-    additional_env={"OMP_NUM_THREADS": 8},
-    cache_name="torch_examples",
-    docker_image=[{"image":"huggingface/transformers-examples-torch"}],
-    # TODO @ArthurZucker remove this once docker is easier to build
-    install_steps=["uv venv && uv pip install . && uv pip install -r examples/pytorch/_tests_requirements.txt"],
-    pytest_num_workers=1,
-)
-
-
-examples_tensorflow_job = CircleCIJob(
-    "examples_tensorflow",
-    cache_name="tensorflow_examples",
-    docker_image=[{"image":"huggingface/transformers-examples-tf"}],
-    install_steps=["uv venv && uv pip install . && uv pip install -r examples/tensorflow/_tests_requirements.txt"],
-    parallelism=8
-)
-
-
-hub_job = CircleCIJob(
-    "hub",
-    additional_env={"HUGGINGFACE_CO_STAGING": True},
-    docker_image=[{"image":"huggingface/transformers-torch-light"}],
-    install_steps=[
-        "uv venv && uv pip install .",
-        'git config --global user.email "ci@dummy.com"',
-        'git config --global user.name "ci"',
-    ],
-    marker="is_staging_test",
-    pytest_num_workers=1,
-)
-
-
-onnx_job = CircleCIJob(
-    "onnx",
-    docker_image=[{"image":"huggingface/transformers-torch-tf-light"}],
-    install_steps=[
-        "uv venv && uv pip install .",
-        "uv pip install --upgrade eager pip",
-        "uv pip install .[torch,tf,testing,sentencepiece,onnxruntime,vision,rjieba]",
-    ],
-    pytest_options={"k onnx": None},
-    pytest_num_workers=1,
-)
-
-
-exotic_models_job = CircleCIJob(
-    "exotic_models",
-    install_steps=["uv venv && uv pip install ."],
-    docker_image=[{"image":"huggingface/transformers-exotic-models"}],
-    tests_to_run=[
-        "tests/models/*layoutlmv*",
-        "tests/models/*nat",
-        "tests/models/deta",
-        "tests/models/udop",
-        "tests/models/nougat",
-    ],
-    pytest_num_workers=12,
-    parallelism=4,
-    pytest_options={"durations": 100},
-)
-
-
-repo_utils_job = CircleCIJob(
-    "repo_utils",
-    docker_image=[{"image":"huggingface/transformers-consistency"}],
-    install_steps=["uv venv && uv pip install ."],
-    parallelism=None,
-    pytest_num_workers=1,
-    resource_class="large",
-    tests_to_run="tests/repo_utils",
-)
-
-
-# We also include a `dummy.py` file in the files to be doc-tested to prevent edge case failure. Otherwise, the pytest
-# hangs forever during test collection while showing `collecting 0 items / 21 errors`. (To see this, we have to remove
-# the bash output redirection.)
-py_command = 'from utils.tests_fetcher import get_doctest_files; to_test = get_doctest_files() + ["dummy.py"]; to_test = " ".join(to_test); print(to_test)'
-py_command = f"$(python3 -c '{py_command}')"
-command = f'echo "{py_command}" > pr_documentation_tests_temp.txt'
-doc_test_job = CircleCIJob(
-    "pr_documentation_tests",
-    docker_image=[{"image":"huggingface/transformers-consistency"}],
-    additional_env={"TRANSFORMERS_VERBOSITY": "error", "DATASETS_VERBOSITY": "error", "SKIP_CUDA_DOCTEST": "1"},
-    install_steps=[
-        # Add an empty file to keep the test step running correctly even no file is selected to be tested.
-        "touch dummy.py",
-        {
-            "name": "Get files to test",
-            "command": command,
-        },
-        {
-            "name": "Show information in `Get files to test`",
-            "command":
-                "cat pr_documentation_tests_temp.txt"
-        },
-        {
-            "name": "Get the last line in `pr_documentation_tests.txt`",
-            "command":
-                "tail -n1 pr_documentation_tests_temp.txt | tee pr_documentation_tests.txt"
-        },
-    ],
-    tests_to_run="$(cat pr_documentation_tests.txt)",  # noqa
-    pytest_options={"-doctest-modules": None, "doctest-glob": "*.md", "dist": "loadfile", "rvsA": None},
-    command_timeout=1200,  # test cannot run longer than 1200 seconds
-    pytest_num_workers=1,
-)
-
-REGULAR_TESTS = [
-    torch_and_tf_job,
-    torch_and_flax_job,
-    torch_job,
-    tf_job,
-    flax_job,
-    custom_tokenizers_job,
-    hub_job,
-    onnx_job,
-    exotic_models_job,
-    tokenization_job
-]
-EXAMPLES_TESTS = [
-    examples_torch_job,
-    examples_tensorflow_job,
-]
-PIPELINE_TESTS = [
-    pipelines_torch_job,
-    pipelines_tf_job,
-]
-REPO_UTIL_TESTS = [repo_utils_job]
-DOC_TESTS = [doc_test_job]
-
-
-def create_circleci_config(folder=None):
-    if folder is None:
-        folder = os.getcwd()
-    # Used in CircleCIJob.to_dict() to expand the test list (for using parallelism)
-    os.environ["test_preparation_dir"] = folder
-    jobs = []
-    all_test_file = os.path.join(folder, "test_list.txt")
-    if os.path.exists(all_test_file):
-        with open(all_test_file) as f:
-            all_test_list = f.read()
-    else:
-        all_test_list = []
-    if len(all_test_list) > 0:
-        jobs.extend(PIPELINE_TESTS)
-
-    test_file = os.path.join(folder, "filtered_test_list.txt")
-    if os.path.exists(test_file):
-        with open(test_file) as f:
-            test_list = f.read()
-    else:
-        test_list = []
-    if len(test_list) > 0:
-        jobs.extend(REGULAR_TESTS)
-
-        extended_tests_to_run = set(test_list.split())
-        # Extend the test files for cross test jobs
-        for job in jobs:
-            if job.job_name in ["tests_torch_and_tf", "tests_torch_and_flax"]:
-                for test_path in copy.copy(extended_tests_to_run):
-                    dir_path, fn = os.path.split(test_path)
-                    if fn.startswith("test_modeling_tf_"):
-                        fn = fn.replace("test_modeling_tf_", "test_modeling_")
-                    elif fn.startswith("test_modeling_flax_"):
-                        fn = fn.replace("test_modeling_flax_", "test_modeling_")
-                    else:
-                        if job.job_name == "test_torch_and_tf":
-                            fn = fn.replace("test_modeling_", "test_modeling_tf_")
-                        elif job.job_name == "test_torch_and_flax":
-                            fn = fn.replace("test_modeling_", "test_modeling_flax_")
-                    new_test_file = str(os.path.join(dir_path, fn))
-                    if os.path.isfile(new_test_file):
-                        if new_test_file not in extended_tests_to_run:
-                            extended_tests_to_run.add(new_test_file)
-        extended_tests_to_run = sorted(extended_tests_to_run)
-        for job in jobs:
-            if job.job_name in ["tests_torch_and_tf", "tests_torch_and_flax"]:
-                job.tests_to_run = extended_tests_to_run
-        fn = "filtered_test_list_cross_tests.txt"
-        f_path = os.path.join(folder, fn)
-        with open(f_path, "w") as fp:
-            fp.write(" ".join(extended_tests_to_run))
-
-    example_file = os.path.join(folder, "examples_test_list.txt")
-    if os.path.exists(example_file) and os.path.getsize(example_file) > 0:
-        with open(example_file, "r", encoding="utf-8") as f:
-            example_tests = f.read()
-        for job in EXAMPLES_TESTS:
-            framework = job.name.replace("examples_", "").replace("torch", "pytorch")
-            if example_tests == "all":
-                job.tests_to_run = [f"examples/{framework}"]
-            else:
-                job.tests_to_run = [f for f in example_tests.split(" ") if f.startswith(f"examples/{framework}")]
-
-            if len(job.tests_to_run) > 0:
-                jobs.append(job)
-
-    doctest_file = os.path.join(folder, "doctest_list.txt")
-    if os.path.exists(doctest_file):
-        with open(doctest_file) as f:
-            doctest_list = f.read()
-    else:
-        doctest_list = []
-    if len(doctest_list) > 0:
-        jobs.extend(DOC_TESTS)
-
-    repo_util_file = os.path.join(folder, "test_repo_utils.txt")
-    if os.path.exists(repo_util_file) and os.path.getsize(repo_util_file) > 0:
-        jobs.extend(REPO_UTIL_TESTS)
-
-    if len(jobs) == 0:
-        jobs = [EmptyJob()]
-    config = {"version": "2.1"}
-    config["parameters"] = {
-        # Only used to accept the parameters from the trigger
-        "nightly": {"type": "boolean", "default": False},
-        "tests_to_run": {"type": "string", "default": test_list},
-    }
-    config["jobs"] = {j.job_name: j.to_dict() for j in jobs}
-    config["workflows"] = {"version": 2, "run_tests": {"jobs": [j.job_name for j in jobs]}}
-    with open(os.path.join(folder, "generated_config.yml"), "w") as f:
-        f.write(yaml.dump(config, indent=2, width=1000000, sort_keys=False))
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser()
-    parser.add_argument(
-        "--fetcher_folder", type=str, default=None, help="Only test that all tests and modules are accounted for."
-    )
-    args = parser.parse_args()
-
-    create_circleci_config(args.fetcher_folder)

.circleci/parse_test_outputs.py

@@ -1,70 +0,0 @@
-import re
-import argparse
-
-def parse_pytest_output(file_path):
-    skipped_tests = {}
-    skipped_count = 0
-    with open(file_path, 'r') as file:
-        for line in file:
-            match = re.match(r'^SKIPPED \[(\d+)\] (tests/.*): (.*)$', line)
-            if match:
-                skipped_count += 1
-                test_file, test_line, reason = match.groups()
-                skipped_tests[reason] = skipped_tests.get(reason, []) + [(test_file, test_line)]
-    for k,v in sorted(skipped_tests.items(), key=lambda x:len(x[1])):
-        print(f"{len(v):4} skipped because: {k}")
-    print("Number of skipped tests:", skipped_count)
-
-def parse_pytest_failure_output(file_path):
-    failed_tests = {}
-    failed_count = 0
-    with open(file_path, 'r') as file:
-        for line in file:
-            match = re.match(r'^FAILED (tests/.*) - (.*): (.*)$', line)
-            if match:
-                failed_count += 1
-                _, error, reason = match.groups()
-                failed_tests[reason] = failed_tests.get(reason, []) + [error]
-    for k,v in sorted(failed_tests.items(), key=lambda x:len(x[1])):
-        print(f"{len(v):4} failed because `{v[0]}` -> {k}")
-    print("Number of failed tests:", failed_count)
-    if failed_count>0:
-        exit(1)
-
-def parse_pytest_errors_output(file_path):
-    print(file_path)
-    error_tests = {}
-    error_count = 0
-    with open(file_path, 'r') as file:
-        for line in file:
-            match = re.match(r'^ERROR (tests/.*) - (.*): (.*)$', line)
-            if match:
-                error_count += 1
-                _, test_error, reason = match.groups()
-                error_tests[reason] = error_tests.get(reason, []) + [test_error]
-    for k,v in sorted(error_tests.items(), key=lambda x:len(x[1])):
-        print(f"{len(v):4} errored out because of `{v[0]}` -> {k}")
-    print("Number of errors:", error_count)
-    if error_count>0:
-        exit(1)
-
-def main():
-    parser = argparse.ArgumentParser()
-    parser.add_argument("--file", help="file to parse")
-    parser.add_argument("--skip", action="store_true", help="show skipped reasons")
-    parser.add_argument("--fail", action="store_true", help="show failed tests")
-    parser.add_argument("--errors", action="store_true", help="show failed tests")
-    args = parser.parse_args()
-
-    if args.skip:
-        parse_pytest_output(args.file)
-
-    if args.fail:
-        parse_pytest_failure_output(args.file)
-
-    if args.errors:
-        parse_pytest_errors_output(args.file)
-
-
-if __name__ == "__main__":
-    main()

.github/ISSUE_TEMPLATE/bug-report.yml

@@ -2,16 +2,6 @@ name: "\U0001F41B Bug Report"
 description: Submit a bug report to help us improve transformers
 labels: [ "bug" ]
 body:
-  - type: markdown
-    attributes:
-      value: |
-        Thanks for taking the time to fill out this bug report! 🤗
-
-        Before you submit your bug report:
-
-          - If it is your first time submitting, be sure to check our [bug report guidelines](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md#did-you-find-a-bug)
-          - Try our [docs bot](https://huggingface.co/spaces/huggingchat/hf-docs-chat) -- it might be able to help you with your issue
-
   - type: textarea
     id: system-info
     attributes:
@@ -28,36 +18,40 @@ body:
       description: |
         Your issue will be replied to more quickly if you can figure out the right person to tag with @
         If you know how to use git blame, that is the easiest way, otherwise, here is a rough guide of **who to tag**.
-
-        All issues are read by one of the core maintainers, so if you don't know who to tag, just leave this blank and
-        a core maintainer will ping the right person.
-
         Please tag fewer than 3 people.
-
+        
         Models:
 
-          - text models: @ArthurZucker
-          - vision models: @amyeroberts
-          - speech models: @sanchit-gandhi
-          - graph models: @clefourrier
+          - ALBERT, BERT, XLM, DeBERTa, DeBERTa-v2, ELECTRA, MobileBert, SqueezeBert: `@LysandreJik`
+          - T5, Pegasus, EncoderDecoder: `@patrickvonplaten`
+          - Blenderbot, MBART, BART, Marian, Pegasus: `@patil-suraj`
+          - Reformer, TransfoXL, XLNet, FNet: `@patrickvonplaten`
+          - Longformer, BigBird: `@ydshieh`
+          - FSMT: `@stas00`
+          - Funnel: `@sgugger`
+          - GPT-2, GPT: `@patil-suraj`, `@patrickvonplaten`, `@LysandreJik`
+          - RAG, DPR: `@patrickvonplaten`, `@lhoestq`
+          - TensorFlow: `@Rocketknight1`
+          - JAX/Flax: `@patil-suraj`
+          - TAPAS, LayoutLM, LayoutLMv2, LUKE, ViT, BEiT, DEiT, DETR, CANINE: `@NielsRogge`
+          - GPT-Neo, GPT-J, CLIP: `@patil-suraj`
+          - Wav2Vec2, HuBERT, UniSpeech, UniSpeechSAT, SEW, SEW-D: `@patrickvonplaten`, `@anton-l`
+          - SpeechEncoderDecoder, Speech2Text, Speech2Text2: `@sanchit-gandhi`, `@patrickvonplaten`, `@anton-l`
+          
+          If the model isn't in the list, ping `@LysandreJik` who will redirect you to the correct contributor.
 
         Library:
+          - Benchmarks: `@patrickvonplaten`
+          - Deepspeed: `@stas00`
+          - Ray/raytune: `@richardliaw`, `@amogkam`
+          - Text generation: `@patrickvonplaten`, `@Narsil`, `@gante`
+          - Tokenizers: `@SaulLu`
+          - Trainer: `@sgugger`
+          - Pipelines: `@Narsil`
+          - Speech: `@patrickvonplaten`, `@anton-l`, `@sanchit-gandhi`
+          - Vision: `@NielsRogge`, `@sgugger`
 
-          - flax: @sanchit-gandhi
-          - generate: @zucchini-nlp (visual-language models) or @gante (all others)
-          - pipelines: @Narsil
-          - tensorflow: @gante and @Rocketknight1
-          - tokenizers: @ArthurZucker
-          - trainer: @muellerzr @SunMarc
-
-        Integrations:
-
-          - deepspeed: HF Trainer/Accelerate: @muellerzr
-          - ray/raytune: @richardliaw, @amogkam
-          - Big Model Inference: @SunMarc
-          - quantization (bitsandbytes, autogpt): @SunMarc
-
-        Documentation: @stevhliu
+        Documentation: `@sgugger`, `@stevhliu`
 
         Model hub:
 
@@ -65,19 +59,17 @@ body:
 
         HF projects:
 
-          - accelerate: [different repo](https://github.com/huggingface/accelerate)
           - datasets: [different repo](https://github.com/huggingface/datasets)
-          - diffusers: [different repo](https://github.com/huggingface/diffusers)
           - rust tokenizers: [different repo](https://github.com/huggingface/tokenizers)
 
-        Maintained examples (not research project or legacy):
+        Examples:
 
-          - Flax: @sanchit-gandhi
-          - PyTorch: See Models above and tag the person corresponding to the modality of the example.
-          - TensorFlow: @Rocketknight1
+          - maintained examples (not research project or legacy): `@sgugger`, `@patil-suraj`
 
-        Research projects are not maintained and should be taken as is.
+        For research projetcs, please ping the contributor directly. For example, on the following projects:
 
+          - research_projects/bert-loses-patience: `@JetRunner`
+          - research_projects/distillation: `@VictorSanh`
       placeholder: "@Username ..."
 
   - type: checkboxes
@@ -112,11 +104,11 @@ body:
 
       placeholder: |
         Steps to reproduce the behavior:
-
+          
           1.
           2.
           3.
-
+          
 
   - type: textarea
     id: expected-behavior

.github/ISSUE_TEMPLATE/feature-request.yml

@@ -1,6 +1,6 @@
 name: "\U0001F680 Feature request"
 description: Submit a proposal/request for a new transformers feature
-labels: [ "Feature request" ]
+labels: [ "feature" ]
 body:
   - type: textarea
     id: feature-request
@@ -19,7 +19,7 @@ body:
       label: Motivation
       description: |
         Please outline the motivation for the proposal. Is your feature request related to a problem? e.g., I'm always frustrated when [...]. If this is related to another GitHub issue, please link here too.
-
+        
 
   - type: textarea
     id: contribution

.github/ISSUE_TEMPLATE/i18n.md

@@ -1,46 +0,0 @@
----
-name: 🌐 Translating a new language?
-about: Start a new translation effort in your language
-title: '[i18n-<languageCode>] Translating docs to <languageName>'
-labels: WIP
-assignees: ''
-
----
-
-<!--
-Note: Please search to see if an issue already exists for the language you are trying to translate.
--->
-
-Hi!
-
-Let's bring the documentation to all the <languageName>-speaking community 🌐 (currently 0 out of 267 complete)
-
-Who would want to translate? Please follow the 🤗 [TRANSLATING guide](https://github.com/huggingface/transformers/blob/main/docs/TRANSLATING.md). Here is a list of the files ready for translation. Let us know in this issue if you'd like to translate any, and we'll add your name to the list.
-
-Some notes:
-
-* Please translate using an informal tone (imagine you are talking with a friend about transformers 🤗).
-* Please translate in a gender-neutral way.
-* Add your translations to the folder called `<languageCode>` inside the [source folder](https://github.com/huggingface/transformers/tree/main/docs/source).
-* Register your translation in `<languageCode>/_toctree.yml`; please follow the order of the [English version](https://github.com/huggingface/transformers/blob/main/docs/source/en/_toctree.yml).
-* Once you're finished, open a pull request and tag this issue by including #issue-number in the description, where issue-number is the number of this issue. Please ping @stevhliu and @MKhalusova for review.
-* 🙋 If you'd like others to help you with the translation, you can also post in the 🤗 [forums](https://discuss.huggingface.co/).
-
-## Get Started section
-
-- [ ] [index.md](https://github.com/huggingface/transformers/blob/main/docs/source/en/index.md) https://github.com/huggingface/transformers/pull/20180
-- [ ] [quicktour.md](https://github.com/huggingface/transformers/blob/main/docs/source/en/quicktour.md) (waiting for initial PR to go through)
-- [ ] [installation.md](https://github.com/huggingface/transformers/blob/main/docs/source/en/installation.md).
-
-## Tutorial section
-- [ ] [pipeline_tutorial.md](https://github.com/huggingface/transformers/blob/main/docs/source/en/pipeline_tutorial.md)
-- [ ]  [autoclass_tutorial.md](https://github.com/huggingface/transformers/blob/main/docs/source/en/autoclass_tutorial.md)
-- [ ]  [preprocessing.md](https://github.com/huggingface/transformers/blob/main/docs/source/en/preprocessing.md)
-- [ ]  [training.md](https://github.com/huggingface/transformers/blob/main/docs/source/en/training.md)
-- [ ]  [accelerate.md](https://github.com/huggingface/transformers/blob/main/docs/source/en/accelerate.md)
-- [ ]  [model_sharing.md](https://github.com/huggingface/transformers/blob/main/docs/source/en/model_sharing.md)
-- [ ]  [multilingual.md](https://github.com/huggingface/transformers/blob/main/docs/source/en/multilingual.md)
-
-<!--
-Keep on adding more as you go 🔥
--->

.github/PULL_REQUEST_TEMPLATE.md

@@ -17,7 +17,7 @@ Fixes # (issue)
 
 ## Before submitting
 - [ ] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
-- [ ] Did you read the [contributor guideline](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md#create-a-pull-request),
+- [ ] Did you read the [contributor guideline](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md#start-contributing-pull-requests),
       Pull Request section?
 - [ ] Was this discussed/approved via a Github issue or the [forum](https://discuss.huggingface.co/)? Please add a link
       to it if that's the case.
@@ -39,40 +39,36 @@ members/contributors who may be interested in your PR.
 
 Models:
 
-- text models: @ArthurZucker
-- vision models: @amyeroberts
-- speech models: @sanchit-gandhi
-- graph models: @clefourrier
+- albert, bert, xlm: @LysandreJik
+- blenderbot, bart, marian, pegasus, encoderdecoder,  t5: @patrickvonplaten, @patil-suraj
+- longformer, reformer, transfoxl, xlnet: @patrickvonplaten
+- fsmt: @stas00
+- funnel: @sgugger
+- gpt2: @patrickvonplaten, @LysandreJik
+- rag: @patrickvonplaten, @lhoestq
+- tensorflow: @LysandreJik
 
 Library:
 
-- flax: @sanchit-gandhi
-- generate: @zucchini-nlp (visual-language models) or @gante (all others)
-- pipelines: @Narsil
-- tensorflow: @gante and @Rocketknight1
-- tokenizers: @ArthurZucker
-- trainer: @muellerzr and @SunMarc
-
-Integrations:
-
-- deepspeed: HF Trainer/Accelerate: @muellerzr
+- benchmarks: @patrickvonplaten
+- deepspeed: @stas00
 - ray/raytune: @richardliaw, @amogkam
-- Big Model Inference: @SunMarc
-- quantization (bitsandbytes, autogpt): @SunMarc
+- text generation: @patrickvonplaten
+- tokenizers: @n1t0, @LysandreJik
+- trainer: @sgugger
+- pipelines: @LysandreJik
 
-Documentation: @stevhliu
+Documentation: @sgugger
 
 HF projects:
 
-- accelerate: [different repo](https://github.com/huggingface/accelerate)
 - datasets: [different repo](https://github.com/huggingface/datasets)
-- diffusers: [different repo](https://github.com/huggingface/diffusers)
 - rust tokenizers: [different repo](https://github.com/huggingface/tokenizers)
 
-Maintained examples (not research project or legacy):
+Examples:
 
-- Flax: @sanchit-gandhi
-- PyTorch: See Models above and tag the person corresponding to the modality of the example.
-- TensorFlow: @Rocketknight1
+- maintained examples (not research project or legacy): @sgugger, @patil-suraj
+- research_projects/bert-loses-patience: @JetRunner
+- research_projects/distillation: @VictorSanh
 
  -->

.github/conda/meta.yaml

@@ -16,6 +16,7 @@ requirements:
     - pip
     - numpy >=1.17
     - dataclasses
+    - importlib_metadata
     - huggingface_hub
     - packaging
     - filelock
@@ -26,12 +27,11 @@ requirements:
     - protobuf
     - tokenizers >=0.11.1,!=0.11.3,<0.13
     - pyyaml >=5.1
-    - safetensors
-    - fsspec
   run:
     - python
     - numpy >=1.17
     - dataclasses
+    - importlib_metadata
     - huggingface_hub
     - packaging
     - filelock
@@ -42,8 +42,6 @@ requirements:
     - protobuf
     - tokenizers >=0.11.1,!=0.11.3,<0.13
     - pyyaml >=5.1
-    - safetensors
-    - fsspec
 
 test:
   imports:

.github/workflows/TROUBLESHOOT.md

@@ -1,6 +1,6 @@
 # Troubleshooting
 
-This is a document explaining how to deal with various issues on github-actions self-hosted CI. The entries may include actual solutions or pointers to Issues that cover those.
+This is a document explaining how to deal with various issues on github-actions self-hosted CI. The entries may include actually solutions or pointers to Issues that cover those.
 
 ## GitHub Actions (self-hosted CI)
 

.github/workflows/add-model-like.yml

@@ -3,20 +3,20 @@ name: Add model like runner
 on:
   push:
     branches:
-      - none # put main here when this is fixed
-  #pull_request:
-  #  paths:
-  #    - "src/**"
-  #    - "tests/**"
-  #    - ".github/**"
-  #  types: [opened, synchronize, reopened]
+      - main
+  pull_request:
+    paths:
+      - "src/**"
+      - "tests/**"
+      - ".github/**"
+    types: [opened, synchronize, reopened]
 
 jobs:
   run_tests_templates_like:
     name: "Add new model like template tests"
-    runs-on: ubuntu-22.04
+    runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v2
 
       - name: Install dependencies
         run: |
@@ -74,7 +74,7 @@ jobs:
 
       - name: Test suite reports artifacts
         if: ${{ always() }}
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v2
         with:
           name: run_all_tests_new_models_test_reports
           path: reports/tests_new_models

.github/workflows/benchmark.yml

@@ -1,42 +0,0 @@
-name: Self-hosted runner (benchmark)
-
-on:
-  schedule:
-    - cron: "17 2 * * *"
-  workflow_call:
-
-env:
-  HF_HOME: /mnt/cache
-  TF_FORCE_GPU_ALLOW_GROWTH: true
-
-
-jobs:
-  benchmark:
-    name: Benchmark
-    runs-on: [single-gpu, nvidia-gpu, a10, ci]
-    container:
-      image: huggingface/transformers-all-latest-gpu
-      options: --gpus all --privileged --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: |
-          git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: Benchmark (daily)
-        if: github.event_name == 'schedule'
-        working-directory: /transformers
-        run: |
-          python3 -m pip install optimum-benchmark>=0.2.0
-          HF_TOKEN=${{ secrets.TRANSFORMERS_BENCHMARK_TOKEN }} python3 benchmark/benchmark.py --repo_id hf-internal-testing/benchmark_results --path_in_repo $(date +'%Y-%m-%d') --config-dir benchmark/config --config-name generation --commit=${{ github.sha }} backend.model=google/gemma-2b backend.cache_implementation=null,static backend.torch_compile=false,true --multirun
-
-      - name: Benchmark (merged to main event)
-        if: github.event_name == 'push' && github.ref_name == 'main'
-        working-directory: /transformers
-        run: |
-          python3 -m pip install optimum-benchmark>=0.2.0
-          HF_TOKEN=${{ secrets.TRANSFORMERS_BENCHMARK_TOKEN }} python3 benchmark/benchmark.py --repo_id hf-internal-testing/benchmark_results_merge_event --path_in_repo $(date +'%Y-%m-%d') --config-dir benchmark/config --config-name generation --commit=${{ github.sha }} backend.model=google/gemma-2b backend.cache_implementation=null,static backend.torch_compile=false,true --multirun

.github/workflows/build-ci-docker-images.yml

@@ -1,77 +0,0 @@
-name: Build pr ci-docker
-
-on:
-  push:
-    branches:
-      - push-ci-image # for now let's only build on this branch
-  repository_dispatch:
-  workflow_call:
-    inputs:
-      image_postfix:
-        required: true
-        type: string
-  schedule:
-    - cron: "6 0 * * *"
-
-
-concurrency:
-  group: ${{ github.workflow }}
-  cancel-in-progress: true
-
-jobs:
-  build:
-    runs-on: ubuntu-22.04
-
-    if: ${{ contains(github.event.head_commit.message, '[build-ci-image]') || contains(github.event.head_commit.message, '[push-ci-image]') && '!cancelled()' || github.event_name == 'schedule' }}
-
-    strategy:
-      matrix:
-        file: ["quality", "consistency", "custom-tokenizers", "torch-light", "tf-light", "exotic-models", "torch-tf-light", "torch-jax-light", "jax-light", "examples-torch",  "examples-tf"]
-    continue-on-error: true
-
-    steps:
-      -
-        name: Set tag
-        run: |
-              if ${{contains(github.event.head_commit.message, '[build-ci-image]')}}; then
-                  echo "TAG=huggingface/transformers-${{ matrix.file }}:dev" >> "$GITHUB_ENV" 
-                  echo "setting it to DEV!"
-              else
-                  echo "TAG=huggingface/transformers-${{ matrix.file }}" >> "$GITHUB_ENV"
-                  
-              fi
-      -
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
-      -
-        name: Check out code
-        uses: actions/checkout@v4
-      -
-        name: Login to DockerHub
-        uses: docker/login-action@v3
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      -
-        name: Build ${{ matrix.file }}.dockerfile
-        uses: docker/build-push-action@v5
-        with:
-          context: ./docker
-          build-args: |
-            REF=${{ github.sha }}
-          file: "./docker/${{ matrix.file }}.dockerfile"
-          push: ${{ contains(github.event.head_commit.message, 'ci-image]') ||  github.event_name == 'schedule' }}
-          tags: ${{ env.TAG }}
-
-  notify:
-    runs-on: ubuntu-22.04
-    if: ${{ contains(github.event.head_commit.message, '[build-ci-image]') || contains(github.event.head_commit.message, '[push-ci-image]') && '!cancelled()' || github.event_name == 'schedule' }}
-    steps:
-      - name: Post to Slack
-        if: ${{ contains(github.event.head_commit.message, '[push-ci-image]') && github.event_name != 'schedule' }}
-        uses: huggingface/hf-workflows/.github/actions/post-slack@main
-        with:
-          slack_channel: "#transformers-ci-circleci-images"
-          title: 🤗 New docker images for CircleCI are pushed.
-          status: ${{ job.status }}
-          slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}

.github/workflows/build-docker-images.yml

@@ -3,15 +3,11 @@ name: Build docker images (scheduled)
 on:
   push:
     branches:
-      - build_ci_docker_image*
+      - docker-image*
   repository_dispatch:
   workflow_call:
-    inputs:
-      image_postfix:
-        required: true
-        type: string
   schedule:
-    - cron: "17 0 * * *"
+    - cron: "0 1 * * *"
 
 concurrency:
   group: docker-images-builds
@@ -20,183 +16,152 @@ concurrency:
 jobs:
   latest-docker:
     name: "Latest PyTorch + TensorFlow [dev]"
-    runs-on: [intel-cpu, 8-cpu, ci]
+    runs-on: ubuntu-latest
     steps:
       -
         name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
+        uses: docker/setup-buildx-action@v1
       -
         name: Check out code
-        uses: actions/checkout@v4
+        uses: actions/checkout@v2
       -
         name: Login to DockerHub
-        uses: docker/login-action@v3
+        uses: docker/login-action@v1
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
       -
         name: Build and push
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v2
         with:
           context: ./docker/transformers-all-latest-gpu
           build-args: |
             REF=main
           push: true
-          tags: huggingface/transformers-all-latest-gpu${{ inputs.image_postfix }}
-      # Push CI images still need to be re-built daily
-      -
-        name: Build and push (for Push CI) in a daily basis
-        # This condition allows `schedule` events, or `push` events that trigger this workflow NOT via `workflow_call`.
-        # The later case is useful for manual image building for debugging purpose. Use another tag in this case!
-        if: inputs.image_postfix != '-push-ci'
-        uses: docker/build-push-action@v5
-        with:
-          context: ./docker/transformers-all-latest-gpu
-          build-args: |
-            REF=main
-          push: true
-          tags: huggingface/transformers-all-latest-gpu-push-ci
+          tags: huggingface/transformers-all-latest-gpu
 
-      - name: Post to Slack
-        if: always()
-        uses: huggingface/hf-workflows/.github/actions/post-slack@main
+  latest-with-torch-nightly-docker:
+    name: "Nightly PyTorch + Stable TensorFlow"
+    runs-on: ubuntu-latest
+    steps:
+      -
+        name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v1
+      -
+        name: Check out code
+        uses: actions/checkout@v2
+      -
+        name: Login to DockerHub
+        uses: docker/login-action@v1
         with:
-          slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the transformers-all-latest-gpu-push-ci docker build 
-          status: ${{ job.status }}
-          slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_PASSWORD }}
+      -
+        name: Build and push
+        uses: docker/build-push-action@v2
+        with:
+          context: ./docker/transformers-all-latest-gpu
+          build-args: |
+            REF=main
+            PYTORCH=pre
+          push: true
+          tags: huggingface/transformers-all-latest-torch-nightly-gpu
 
   latest-torch-deepspeed-docker:
     name: "Latest PyTorch + DeepSpeed"
-    runs-on: [intel-cpu, 8-cpu, ci]
+    runs-on: ubuntu-latest
     steps:
       -
         name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
+        uses: docker/setup-buildx-action@v1
       -
         name: Check out code
-        uses: actions/checkout@v4
+        uses: actions/checkout@v2
       -
         name: Login to DockerHub
-        uses: docker/login-action@v3
+        uses: docker/login-action@v1
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
       -
         name: Build and push
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v2
         with:
           context: ./docker/transformers-pytorch-deepspeed-latest-gpu
           build-args: |
             REF=main
           push: true
-          tags: huggingface/transformers-pytorch-deepspeed-latest-gpu${{ inputs.image_postfix }}
+          tags: huggingface/transformers-pytorch-deepspeed-latest-gpu
 
-      - name: Post to Slack
-        if: always()
-        uses: huggingface/hf-workflows/.github/actions/post-slack@main
-        with:
-          slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER}}
-          title: 🤗 Results of the transformers-pytorch-deepspeed-latest-gpu docker build 
-          status: ${{ job.status }}
-          slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
-
-  # Can't build 2 images in a single job `latest-torch-deepspeed-docker` (for `nvcr.io/nvidia`)
-  latest-torch-deepspeed-docker-for-push-ci-daily-build:
-    name: "Latest PyTorch + DeepSpeed (Push CI - Daily Build)"
-    runs-on: [intel-cpu, 8-cpu, ci]
+  nightly-torch-deepspeed-docker:
+    name: "Nightly PyTorch + DeepSpeed"
+    runs-on: ubuntu-latest
     steps:
       -
         name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
+        uses: docker/setup-buildx-action@v1
       -
         name: Check out code
-        uses: actions/checkout@v4
+        uses: actions/checkout@v2
       -
         name: Login to DockerHub
-        uses: docker/login-action@v3
+        uses: docker/login-action@v1
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      # Push CI images still need to be re-built daily
       -
-        name: Build and push (for Push CI) in a daily basis
-        # This condition allows `schedule` events, or `push` events that trigger this workflow NOT via `workflow_call`.
-        # The later case is useful for manual image building for debugging purpose. Use another tag in this case!
-        if: inputs.image_postfix != '-push-ci'
-        uses: docker/build-push-action@v5
+        name: Build and push
+        uses: docker/build-push-action@v2
         with:
-          context: ./docker/transformers-pytorch-deepspeed-latest-gpu
+          context: ./docker/transformers-pytorch-deepspeed-nightly-gpu
           build-args: |
             REF=main
           push: true
-          tags: huggingface/transformers-pytorch-deepspeed-latest-gpu-push-ci
-
-      - name: Post to Slack
-        if: always()
-        uses: huggingface/hf-workflows/.github/actions/post-slack@main
-        with:
-          slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the transformers-pytorch-deepspeed-latest-gpu-push-ci docker build 
-          status: ${{ job.status }}
-          slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
+          tags: huggingface/transformers-pytorch-deepspeed-nightly-gpu
 
   doc-builder:
     name: "Doc builder"
-    # Push CI doesn't need this image
-    if: inputs.image_postfix != '-push-ci'
-    runs-on: [intel-cpu, 8-cpu, ci]
+    runs-on: ubuntu-latest
     steps:
       -
         name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
+        uses: docker/setup-buildx-action@v1
       -
         name: Check out code
-        uses: actions/checkout@v4
+        uses: actions/checkout@v2
       -
         name: Login to DockerHub
-        uses: docker/login-action@v3
+        uses: docker/login-action@v1
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
       -
         name: Build and push
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v2
         with:
           context: ./docker/transformers-doc-builder
           push: true
           tags: huggingface/transformers-doc-builder
 
-      - name: Post to Slack
-        if: always()
-        uses: huggingface/hf-workflows/.github/actions/post-slack@main
-        with:
-          slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the huggingface/transformers-doc-builder docker build 
-          status: ${{ job.status }}
-          slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
-
   latest-pytorch:
     name: "Latest PyTorch [dev]"
-    # Push CI doesn't need this image
-    if: inputs.image_postfix != '-push-ci'
-    runs-on: [intel-cpu, 8-cpu, ci]
+    runs-on: ubuntu-latest
     steps:
       -
         name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
+        uses: docker/setup-buildx-action@v1
       -
         name: Check out code
-        uses: actions/checkout@v4
+        uses: actions/checkout@v2
       -
         name: Login to DockerHub
-        uses: docker/login-action@v3
+        uses: docker/login-action@v1
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
       -
         name: Build and push
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v2
         with:
           context: ./docker/transformers-pytorch-gpu
           build-args: |
@@ -204,181 +169,28 @@ jobs:
           push: true
           tags: huggingface/transformers-pytorch-gpu
 
-      - name: Post to Slack
-        if: always()
-        uses: huggingface/hf-workflows/.github/actions/post-slack@main
-        with:
-          slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the huggingface/transformers-pytorch-gpudocker build 
-          status: ${{ job.status }}
-          slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
-
-  latest-pytorch-amd:
-    name: "Latest PyTorch (AMD) [dev]"
-    runs-on: [intel-cpu, 8-cpu, ci]
-    steps:
-      - 
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
-      - 
-        name: Check out code
-        uses: actions/checkout@v4
-      - 
-        name: Login to DockerHub
-        uses: docker/login-action@v3
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      - 
-        name: Build and push
-        uses: docker/build-push-action@v5
-        with:
-          context: ./docker/transformers-pytorch-amd-gpu
-          build-args: |
-            REF=main
-          push: true
-          tags: huggingface/transformers-pytorch-amd-gpu${{ inputs.image_postfix }}
-      # Push CI images still need to be re-built daily
-      -
-        name: Build and push (for Push CI) in a daily basis
-        # This condition allows `schedule` events, or `push` events that trigger this workflow NOT via `workflow_call`.
-        # The later case is useful for manual image building for debugging purpose. Use another tag in this case!
-        if: inputs.image_postfix != '-push-ci'
-        uses: docker/build-push-action@v5
-        with:
-          context: ./docker/transformers-pytorch-amd-gpu
-          build-args: |
-            REF=main
-          push: true
-          tags: huggingface/transformers-pytorch-amd-gpu-push-ci
-
-      - name: Post to Slack
-        if: always()
-        uses: huggingface/hf-workflows/.github/actions/post-slack@main
-        with:
-          slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the huggingface/transformers-pytorch-amd-gpu-push-ci build 
-          status: ${{ job.status }}
-          slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
-
   latest-tensorflow:
     name: "Latest TensorFlow [dev]"
-    # Push CI doesn't need this image
-    if: inputs.image_postfix != '-push-ci'
-    runs-on: [intel-cpu, 8-cpu, ci]
+    runs-on: ubuntu-latest
     steps:
       -
         name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
+        uses: docker/setup-buildx-action@v1
       -
         name: Check out code
-        uses: actions/checkout@v4
+        uses: actions/checkout@v2
       -
         name: Login to DockerHub
-        uses: docker/login-action@v3
+        uses: docker/login-action@v1
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
       -
         name: Build and push
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v2
         with:
           context: ./docker/transformers-tensorflow-gpu
           build-args: |
             REF=main
           push: true
           tags: huggingface/transformers-tensorflow-gpu
-
-      - name: Post to Slack
-        if: always()
-        uses: huggingface/hf-workflows/.github/actions/post-slack@main
-        with:
-          slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the huggingface/transformers-tensorflow-gpu build 
-          status: ${{ job.status }}
-          slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
-
-  latest-pytorch-deepspeed-amd:
-    name: "PyTorch + DeepSpeed (AMD) [dev]"
-    runs-on: [intel-cpu, 8-cpu, ci]
-    steps:
-      - 
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
-      - 
-        name: Check out code
-        uses: actions/checkout@v4
-      - 
-        name: Login to DockerHub
-        uses: docker/login-action@v3
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      - 
-        name: Build and push
-        uses: docker/build-push-action@v5
-        with:
-          context: ./docker/transformers-pytorch-deepspeed-amd-gpu
-          build-args: |
-            REF=main
-          push: true
-          tags: huggingface/transformers-pytorch-deepspeed-amd-gpu${{ inputs.image_postfix }}
-      # Push CI images still need to be re-built daily
-      -
-        name: Build and push (for Push CI) in a daily basis
-        # This condition allows `schedule` events, or `push` events that trigger this workflow NOT via `workflow_call`.
-        # The later case is useful for manual image building for debugging purpose. Use another tag in this case!
-        if: inputs.image_postfix != '-push-ci'
-        uses: docker/build-push-action@v5
-        with:
-          context: ./docker/transformers-pytorch-deepspeed-amd-gpu
-          build-args: |
-            REF=main
-          push: true
-          tags: huggingface/transformers-pytorch-deepspeed-amd-gpu-push-ci
-
-      - name: Post to Slack
-        if: always()
-        uses: huggingface/hf-workflows/.github/actions/post-slack@main
-        with:
-          slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the transformers-pytorch-deepspeed-amd-gpu build 
-          status: ${{ job.status }}
-          slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
-
-  latest-quantization-torch-docker:
-    name: "Latest Pytorch + Quantization [dev]"
-     # Push CI doesn't need this image
-    if: inputs.image_postfix != '-push-ci'
-    runs-on: [intel-cpu, 8-cpu, ci]
-    steps:
-      -
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
-      -
-        name: Check out code
-        uses: actions/checkout@v4
-      -
-        name: Login to DockerHub
-        uses: docker/login-action@v3
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      -
-        name: Build and push
-        uses: docker/build-push-action@v5
-        with:
-          context: ./docker/transformers-quantization-latest-gpu
-          build-args: |
-            REF=main
-          push: true
-          tags: huggingface/transformers-quantization-latest-gpu${{ inputs.image_postfix }}
-
-      - name: Post to Slack
-        if: always()
-        uses: huggingface/hf-workflows/.github/actions/post-slack@main
-        with:
-          slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the transformers-quantization-latest-gpu build 
-          status: ${{ job.status }}
-          slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}

.github/workflows/build-nightly-ci-docker-images.yml

@@ -1,65 +0,0 @@
-name: Build docker images (Nightly CI)
-
-on:
-  workflow_call:
-  push:
-    branches:
-      - build_nightly_ci_docker_image*
-
-concurrency:
-  group: docker-images-builds
-  cancel-in-progress: false
-
-jobs:
-  latest-with-torch-nightly-docker:
-    name: "Nightly PyTorch + Stable TensorFlow"
-    runs-on: [intel-cpu, 8-cpu, ci]
-    steps:
-      -
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
-      -
-        name: Check out code
-        uses: actions/checkout@v4
-      -
-        name: Login to DockerHub
-        uses: docker/login-action@v2
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      -
-        name: Build and push
-        uses: docker/build-push-action@v3
-        with:
-          context: ./docker/transformers-all-latest-gpu
-          build-args: |
-            REF=main
-            PYTORCH=pre
-          push: true
-          tags: huggingface/transformers-all-latest-torch-nightly-gpu
-
-  nightly-torch-deepspeed-docker:
-    name: "Nightly PyTorch + DeepSpeed"
-    runs-on: [intel-cpu, 8-cpu, ci]
-    steps:
-      -
-        name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
-      -
-        name: Check out code
-        uses: actions/checkout@v4
-      -
-        name: Login to DockerHub
-        uses: docker/login-action@v2
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      -
-        name: Build and push
-        uses: docker/build-push-action@v3
-        with:
-          context: ./docker/transformers-pytorch-deepspeed-nightly-gpu
-          build-args: |
-            REF=main
-          push: true
-          tags: huggingface/transformers-pytorch-deepspeed-nightly-gpu

.github/workflows/build-past-ci-docker-images.yml

@@ -3,7 +3,7 @@ name: Build docker images (Past CI)
 on:
   push:
     branches:
-      - build_past_ci_docker_image*
+      - past-ci-docker-image*
 
 concurrency:
   group: docker-images-builds
@@ -15,40 +15,28 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        version: ["1.13", "1.12", "1.11"]
-    runs-on: [intel-cpu, 8-cpu, ci]
+        version: ["1.11", "1.10", "1.9", "1.8", "1.7", "1.6", "1.5", "1.4"]
+    runs-on: ubuntu-latest
     steps:
       -
         name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
+        uses: docker/setup-buildx-action@v1
       -
         name: Check out code
-        uses: actions/checkout@v4
-      -
-        id: get-base-image
-        name: Get Base Image
-        env:
-          framework_version: ${{ matrix.version }}
-        run: |
-          echo "base_image=$(python3 -c 'import os; from utils.past_ci_versions import past_versions_testing; base_image = past_versions_testing["pytorch"][os.environ["framework_version"]]["base_image"]; print(base_image)')" >> $GITHUB_OUTPUT
-      -
-        name: Print Base Image
-        run: |
-          echo ${{ steps.get-base-image.outputs.base_image }}
+        uses: actions/checkout@v2
       -
         name: Login to DockerHub
-        uses: docker/login-action@v2
+        uses: docker/login-action@v1
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
       -
         name: Build and push
-        uses: docker/build-push-action@v3
+        uses: docker/build-push-action@v2
         with:
           context: ./docker/transformers-past-gpu
           build-args: |
             REF=main
-            BASE_DOCKER_IMAGE=${{ steps.get-base-image.outputs.base_image }}
             FRAMEWORK=pytorch
             VERSION=${{ matrix.version }}
           push: true
@@ -59,41 +47,62 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        version: ["2.11", "2.10", "2.9", "2.8", "2.7", "2.6", "2.5"]
-    runs-on: [intel-cpu, 8-cpu, ci]
+        version: ["2.8", "2.7", "2.6", "2.5"]
+    runs-on: ubuntu-latest
     steps:
       -
         name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
+        uses: docker/setup-buildx-action@v1
       -
         name: Check out code
-        uses: actions/checkout@v4
-      -
-        id: get-base-image
-        name: Get Base Image
-        env:
-          framework_version: ${{ matrix.version }}
-        run: |
-          echo "base_image=$(python3 -c 'import os; from utils.past_ci_versions import past_versions_testing; base_image = past_versions_testing["tensorflow"][os.environ["framework_version"]]["base_image"]; print(base_image)')" >> $GITHUB_OUTPUT
-      -
-        name: Print Base Image
-        run: |
-          echo ${{ steps.get-base-image.outputs.base_image }}
+        uses: actions/checkout@v2
       -
         name: Login to DockerHub
-        uses: docker/login-action@v2
+        uses: docker/login-action@v1
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
       -
         name: Build and push
-        uses: docker/build-push-action@v3
+        uses: docker/build-push-action@v2
         with:
           context: ./docker/transformers-past-gpu
           build-args: |
             REF=main
-            BASE_DOCKER_IMAGE=${{ steps.get-base-image.outputs.base_image }}
             FRAMEWORK=tensorflow
             VERSION=${{ matrix.version }}
           push: true
           tags: huggingface/transformers-tensorflow-past-${{ matrix.version }}-gpu
+
+  past-tensorflow-docker-2-4:
+    name: "Past TensorFlow Docker"
+    strategy:
+      fail-fast: false
+      matrix:
+        version: ["2.4"]
+    runs-on: ubuntu-latest
+    steps:
+      -
+        name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v1
+      -
+        name: Check out code
+        uses: actions/checkout@v2
+      -
+        name: Login to DockerHub
+        uses: docker/login-action@v1
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_PASSWORD }}
+      -
+        name: Build and push
+        uses: docker/build-push-action@v2
+        with:
+          context: ./docker/transformers-past-gpu
+          build-args: |
+            REF=main
+            BASE_DOCKER_IMAGE=nvidia/cuda:11.0.3-cudnn8-devel-ubuntu20.04
+            FRAMEWORK=tensorflow
+            VERSION=${{ matrix.version }}
+          push: true
+          tags: huggingface/transformers-tensorflow-past-${{ matrix.version }}-gpu

.github/workflows/build_documentation.yml

@@ -15,8 +15,6 @@ jobs:
       commit_sha: ${{ github.sha }}
       package: transformers
       notebook_folder: transformers_doc
-      languages: de en es fr hi it ko pt tr zh ja te
-      custom_container: huggingface/transformers-doc-builder
+      languages: de en es it pt
     secrets:
       token: ${{ secrets.HUGGINGFACE_PUSH }}
-      hf_token: ${{ secrets.HF_DOC_BUILD_PUSH }}

.github/workflows/build_pr_documentation.yml

@@ -14,5 +14,4 @@ jobs:
       commit_sha: ${{ github.event.pull_request.head.sha }}
       pr_number: ${{ github.event.number }}
       package: transformers
-      languages: de en es fr hi it ko pt tr zh ja te
-      custom_container: huggingface/transformers-doc-builder
+      languages: de en es it pt

.github/workflows/check_tiny_models.yml

@@ -1,82 +0,0 @@
-name: Check Tiny Models
-
-on:
-  push:
-    branches:
-      - check_tiny_models*
-  repository_dispatch:
-  schedule:
-    - cron: "0 2 * * *"
-
-env:
-  TOKEN: ${{ secrets.TRANSFORMERS_HUB_BOT_HF_TOKEN }}
-
-jobs:
-  check_tiny_models:
-    name: Check tiny models
-    runs-on: ubuntu-22.04
-    steps:
-      - name: Checkout transformers
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 2
-
-      - uses: actions/checkout@v4
-      - name: Set up Python 3.8
-        uses: actions/setup-python@v4
-        with:
-          # Semantic version range syntax or exact version of a Python version
-          python-version: '3.8'
-          # Optional - x64 or x86 architecture, defaults to x64
-          architecture: 'x64'
-
-      - name: Install
-        run: |
-          sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng cmake
-          pip install --upgrade pip
-          python -m pip install -U .[sklearn,torch,testing,sentencepiece,torch-speech,vision,timm,video,tf-cpu]
-          pip install tensorflow_probability
-          python -m pip install -U 'natten<0.15.0'
-
-      - name: Create all tiny models (locally)
-        run: |
-          python utils/create_dummy_models.py tiny_local_models --all --num_workers 2
-
-      - name: Local tiny model reports artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: tiny_local_model_creation_reports
-          path: tiny_local_models/reports
-
-      # GitHub-hosted runners have 2-core CPUs
-      - name: Run pipeline tests against all new (local) tiny models
-        run: |
-          OMP_NUM_THREADS=1 TRANSFORMERS_TINY_MODEL_PATH=tiny_local_models python -m pytest --max-worker-restart=0 -n 2 --dist=loadfile -s -rA --make-reports=tests_pipelines tests/models -m is_pipeline_test -k "test_pipeline_" | tee tests_output.txt
-
-      - name: Test suite reports artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: tiny_local_model_creation_reports
-          path: reports/tests_pipelines
-
-      - name: Create + Upload tiny models for new model architecture(s)
-        run: |
-          python utils/update_tiny_models.py --num_workers 2
-
-      - name: Full report
-        run: cat tiny_models/reports/tiny_model_creation_report.json
-
-      - name: Failure report
-        run: cat tiny_models/reports/simple_failed_report.txt
-
-      - name: Summary report
-        run: cat tiny_models/reports/tiny_model_summary.json
-
-      - name: New tiny model creation reports artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: tiny_model_creation_reports
-          path: tiny_models/reports

.github/workflows/delete_doc_comment.yml

@@ -0,0 +1,13 @@
+name: Delete dev documentation
+
+on:
+  pull_request:
+    types: [ closed ]
+
+
+jobs:
+  delete:
+    uses: huggingface/doc-builder/.github/workflows/delete_doc_comment.yml@main
+    with:
+      pr_number: ${{ github.event.number }}
+      package: transformers

.github/workflows/doctest_job.yml

@@ -1,82 +0,0 @@
-name: Doctest job
-
-on:
-  workflow_call:
-    inputs:
-      job_splits:
-        required: true
-        type: string
-      split_keys:
-        required: true
-        type: string
-
-env:
-  HF_HOME: /mnt/cache
-  TRANSFORMERS_IS_CI: yes
-  RUN_SLOW: yes
-  OMP_NUM_THREADS: 16
-  MKL_NUM_THREADS: 16
-  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
-  TF_FORCE_GPU_ALLOW_GROWTH: true
-
-jobs:
-  run_doctests:
-    name: " "
-    strategy:
-      max-parallel: 8  # 8 jobs at a time
-      fail-fast: false
-      matrix:
-        split_keys: ${{ fromJson(inputs.split_keys) }}
-    runs-on: [single-gpu, nvidia-gpu, t4, ci]
-    container:
-      image: huggingface/transformers-all-latest-gpu
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .[flax]
-
-      - name: GPU visibility
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        run: pip freeze
-
-      - name: Get doctest files
-        working-directory: /transformers
-        run: |
-          echo "${{ toJson(fromJson(inputs.job_splits)[matrix.split_keys]) }}" > doc_tests.txt
-          cat doc_tests.txt
-
-      - name: Set `split_keys`
-        shell: bash
-        run: |
-          echo "${{ matrix.split_keys }}"
-          split_keys=${{ matrix.split_keys }}
-          split_keys=${split_keys//'/'/'_'}
-          echo "split_keys"
-          echo "split_keys=$split_keys" >> $GITHUB_ENV
-
-      - name: Run doctests
-        working-directory: /transformers
-        run: |
-          cat doc_tests.txt
-          python3 -m pytest -v --make-reports doc_tests_gpu_${{ env.split_keys }} --doctest-modules $(cat doc_tests.txt) -sv --doctest-continue-on-failure --doctest-glob="*.md"
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/doc_tests_gpu_${{ env.split_keys }}/failures_short.txt
-
-      - name: "Test suite reports artifacts: doc_tests_gpu_test_reports_${{ env.split_keys }}"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: doc_tests_gpu_test_reports_${{ env.split_keys }}
-          path: /transformers/reports/doc_tests_gpu_${{ env.split_keys }}

.github/workflows/doctests.yml

@@ -3,86 +3,76 @@ name: Doctests
 on:
   push:
     branches:
-      - run_doctest*
+      - doctest*
   repository_dispatch:
   schedule:
-    - cron: "17 2 * * *"
+    - cron: "0 0 * * *"
+
 
 env:
-  NUM_SLICES: 3
+  HF_HOME: /mnt/cache
+  TRANSFORMERS_IS_CI: yes
+  RUN_SLOW: yes
+  OMP_NUM_THREADS: 16
+  MKL_NUM_THREADS: 16
+  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
+  TF_FORCE_GPU_ALLOW_GROWTH: true
 
 jobs:
-  setup:
-    name: Setup
-    runs-on: [single-gpu, nvidia-gpu, t4, ci]
+  run_doctests:
+    runs-on: [self-hosted, doc-tests-gpu]
     container:
       image: huggingface/transformers-all-latest-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    outputs:
-      job_splits: ${{ steps.set-matrix.outputs.job_splits }}
-      split_keys: ${{ steps.set-matrix.outputs.split_keys }}
     steps:
-      - name: Update clone
-        working-directory: /transformers
+      - uses: actions/checkout@v2
+      - name: NVIDIA-SMI
         run: |
-          git fetch && git checkout ${{ github.sha }}
+          nvidia-smi
 
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Check values for matrix
-        working-directory: /transformers
+      - name: GPU visibility
         run: |
-          python3 utils/split_doctest_jobs.py
-          python3 utils/split_doctest_jobs.py --only_return_keys --num_splits ${{ env.NUM_SLICES }}
+          python3 utils/print_env.py
 
-      - id: set-matrix
-        working-directory: /transformers
-        name: Set values for matrix
+      - name: Prepare files for doctests
         run: |
-          echo "job_splits=$(python3 utils/split_doctest_jobs.py)" >> $GITHUB_OUTPUT
-          echo "split_keys=$(python3 utils/split_doctest_jobs.py --only_return_keys --num_splits ${{ env.NUM_SLICES }})" >> $GITHUB_OUTPUT
+          python3 utils/prepare_for_doc_test.py src docs
+
+      - name: Run doctests
+        run: |
+          python3 -m pytest -v --make-reports doc_tests_gpu --doctest-modules $(cat utils/documentation_tests.txt) -sv --doctest-continue-on-failure --doctest-glob="*.mdx"
+
+      - name: Clean files after doctests
+        run: |
+          python3 utils/prepare_for_doc_test.py src docs --remove_new_line
+
+      - name: Failure short reports
+        if: ${{ failure() }}
+        continue-on-error: true
+        run: cat reports/doc_tests_gpu/failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: doc_tests_gpu_test_reports
+          path: reports/doc_tests_gpu
 
-  call_doctest_job:
-    name: "Call doctest jobs"
-    needs: setup
-    strategy:
-      max-parallel: 1  # 1 split at a time (in `doctest_job.yml`, we set `8` to run 8 jobs at the same time)
-      fail-fast: false
-      matrix:
-        split_keys: ${{ fromJson(needs.setup.outputs.split_keys) }}
-    uses: ./.github/workflows/doctest_job.yml
-    with:
-      job_splits: ${{ needs.setup.outputs.job_splits }}
-      split_keys: ${{ toJson(matrix.split_keys) }}
-    secrets: inherit
 
   send_results:
     name: Send results to webhook
-    runs-on: ubuntu-22.04
+    runs-on: ubuntu-latest
     if: always()
-    needs: [call_doctest_job]
+    needs: [run_doctests]
     steps:
-      - uses: actions/checkout@v4
-      - uses: actions/download-artifact@v4
+      - uses: actions/checkout@v2
+      - uses: actions/download-artifact@v2
       - name: Send message to Slack
         env:
           CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
-          ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-          # Use `CI_SLACK_CHANNEL_DUMMY_TESTS` when doing experimentation
-          SLACK_REPORT_CHANNEL: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY_DOCS }}
+          CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY_DOCS }}
+          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY_DOCS }}
+          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
         run: |
           pip install slack_sdk
           python utils/notification_service_doc_tests.py
-
-      - name: "Upload results"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: doc_test_results
-          path: doc_test_results

.github/workflows/model-templates.yml

@@ -0,0 +1,81 @@
+name: Model templates runner
+
+on:
+  repository_dispatch:
+  schedule:
+    - cron: "0 2 * * *"
+
+jobs:
+  run_tests_templates:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v2
+
+      - name: Install dependencies
+        run: |
+          sudo apt -y update && sudo apt install -y libsndfile1-dev
+
+      - name: Load cached virtual environment
+        uses: actions/cache@v2
+        id: cache
+        with:
+          path: ~/venv/
+          key: v4-tests_templates-${{ hashFiles('setup.py') }}
+
+      - name: Create virtual environment on cache miss
+        if: steps.cache.outputs.cache-hit != 'true'
+        run: |
+          python -m venv ~/venv && . ~/venv/bin/activate
+          pip install --upgrade pip!=21.3
+          pip install -e .[dev]
+
+      - name: Check transformers location
+        # make `transformers` available as package (required since we use `-e` flag) and check it's indeed from the repo.
+        run: |
+          . ~/venv/bin/activate
+          python setup.py develop
+          transformer_loc=$(pip show transformers | grep "Location: " | cut -c11-)
+          transformer_repo_loc=$(pwd .)
+          if [ "$transformer_loc" != "$transformer_repo_loc/src" ]; then
+              echo "transformers is from $transformer_loc but it shoud be from $transformer_repo_loc/src."
+              echo "A fix is required. Stop testing."
+              exit 1
+          fi
+
+      - name: Create model files
+        run: |
+          . ~/venv/bin/activate
+          transformers-cli add-new-model --testing --testing_file=templates/adding_a_new_model/tests/encoder-bert-tokenizer.json --path=templates/adding_a_new_model
+          transformers-cli add-new-model --testing --testing_file=templates/adding_a_new_model/tests/pt-encoder-bert-tokenizer.json --path=templates/adding_a_new_model
+          transformers-cli add-new-model --testing --testing_file=templates/adding_a_new_model/tests/standalone.json --path=templates/adding_a_new_model
+          transformers-cli add-new-model --testing --testing_file=templates/adding_a_new_model/tests/tf-encoder-bert-tokenizer.json --path=templates/adding_a_new_model
+          transformers-cli add-new-model --testing --testing_file=templates/adding_a_new_model/tests/tf-seq-2-seq-bart-tokenizer.json --path=templates/adding_a_new_model
+          transformers-cli add-new-model --testing --testing_file=templates/adding_a_new_model/tests/pt-seq-2-seq-bart-tokenizer.json --path=templates/adding_a_new_model
+          transformers-cli add-new-model --testing --testing_file=templates/adding_a_new_model/tests/flax-encoder-bert-tokenizer.json --path=templates/adding_a_new_model
+          transformers-cli add-new-model --testing --testing_file=templates/adding_a_new_model/tests/flax-seq-2-seq-bart-tokenizer.json --path=templates/adding_a_new_model
+          make style
+          python utils/check_table.py --fix_and_overwrite
+          python utils/check_dummies.py --fix_and_overwrite
+          python utils/check_copies.py --fix_and_overwrite
+
+      - name: Run all non-slow tests
+        run: |
+          . ~/venv/bin/activate
+          python -m pytest -n 2 --dist=loadfile -s --make-reports=tests_templates tests/*template*
+
+      - name: Run style changes
+        run: |
+          . ~/venv/bin/activate
+          make style && make quality && make repo-consistency
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: cat reports/tests_templates/failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: run_all_tests_templates_test_reports
+          path: reports/tests_templates

.github/workflows/model_jobs.yml

@@ -1,121 +0,0 @@
-name: model jobs
-
-on:
-  workflow_call:
-    inputs:
-      folder_slices:
-        required: true
-        type: string
-      machine_type:
-        required: true
-        type: string
-      slice_id:
-        required: true
-        type: number
-      runner:
-        required: true
-        type: string
-      docker:
-        required: true
-        type: string
-
-env:
-  HF_HOME: /mnt/cache
-  TRANSFORMERS_IS_CI: yes
-  OMP_NUM_THREADS: 8
-  MKL_NUM_THREADS: 8
-  RUN_SLOW: yes
-  # For gated repositories, we still need to agree to share information on the Hub repo. page in order to get access.
-  # This token is created under the bot `hf-transformers-bot`.
-  HF_HUB_READ_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}
-  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
-  TF_FORCE_GPU_ALLOW_GROWTH: true
-  RUN_PT_TF_CROSS_TESTS: 1
-  CUDA_VISIBLE_DEVICES: 0,1
-
-jobs:
-  run_models_gpu:
-    name: " "
-    strategy:
-      max-parallel: 8
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(inputs.folder_slices)[inputs.slice_id] }}
-    runs-on: ['${{ inputs.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
-    container:
-      image: ${{ inputs.docker }}
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: Echo input and matrix info
-        shell: bash
-        run: |
-          echo "${{ inputs.folder_slices }}"
-          echo "${{ matrix.folders }}"
-          echo "${{ toJson(fromJson(inputs.folder_slices)[inputs.slice_id]) }}"
-
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: Update / Install some packages (for Past CI)
-        if: ${{ contains(inputs.docker, '-past-') }}
-        working-directory: /transformers
-        run: |
-          python3 -m pip install -U datasets
-
-      - name: Update / Install some packages (for Past CI)
-        if: ${{ contains(inputs.docker, '-past-') && contains(inputs.docker, '-pytorch-') }}
-        working-directory: /transformers
-        run: |
-          python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -rsfE -v --make-reports=${{ inputs.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports tests/${{ matrix.folders }}
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ inputs.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports/failures_short.txt
-
-      - name: Run test
-        shell: bash
-        run: |
-          mkdir -p /transformers/reports/${{ inputs.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports
-          echo "hello" > /transformers/reports/${{ inputs.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports/hello.txt
-          echo "${{ inputs.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports"
-
-      - name: "Test suite reports artifacts: ${{ inputs.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ inputs.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports
-          path: /transformers/reports/${{ inputs.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports

.github/workflows/push-important-models.yml

@@ -1,142 +0,0 @@
-name: Slow tests on important models (on Push - A10)
-
-on:
-  push:
-    branches: [ main ]
-
-env:
-  OUTPUT_SLACK_CHANNEL_ID: "C06L2SGMEEA"
-  HF_HUB_READ_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}
-  HF_HOME: /mnt/cache 
-  TRANSFORMERS_IS_CI: yes 
-  OMP_NUM_THREADS: 8 
-  MKL_NUM_THREADS: 8 
-  RUN_SLOW: yes # For gated repositories, we still need to agree to share information on the Hub repo. page in order to get access. # This token is created under the bot `hf-transformers-bot`. 
-  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }} 
-  TF_FORCE_GPU_ALLOW_GROWTH: true 
-  RUN_PT_TF_CROSS_TESTS: 1
-
-jobs:
-  get_modified_models:
-    name: "Get all modified files"
-    runs-on: ubuntu-latest
-    outputs:
-      matrix: ${{ steps.set-matrix.outputs.matrix }}
-    steps:
-      - name: Check out code
-        uses: actions/checkout@v4
-      
-      - name: Get changed files
-        id: changed-files
-        uses: tj-actions/changed-files@3f54ebb830831fc121d3263c1857cfbdc310cdb9 #v42
-        with:
-          files: src/transformers/models/**
-      
-      - name: Run step if only the files listed above change
-        if: steps.changed-files.outputs.any_changed == 'true'
-        id: set-matrix
-        env:
-          ALL_CHANGED_FILES: ${{ steps.changed-files.outputs.all_changed_files }}
-        run: |
-            model_arrays=()
-            for file in $ALL_CHANGED_FILES; do
-                model_path="${file#*models/}"
-                model_path="models/${model_path%%/*}"
-                if grep -qFx "$model_path" utils/important_models.txt; then
-                    # Append the file to the matrix string
-                    model_arrays+=("$model_path")
-                fi
-            done
-            matrix_string=$(printf '"%s", ' "${model_arrays[@]}" | sed 's/, $//')
-            echo "matrix=[$matrix_string]" >> $GITHUB_OUTPUT
-  test_modified_files:
-    needs: get_modified_models
-    name: Slow & FA2 tests
-    runs-on: [single-gpu, nvidia-gpu, a10, ci]
-    container:
-      image: huggingface/transformers-all-latest-gpu
-      options: --gpus all --privileged --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    if: ${{ needs.get_modified_models.outputs.matrix != '[]' && needs.get_modified_models.outputs.matrix != '' && fromJson(needs.get_modified_models.outputs.matrix)[0] != null }}
-    strategy:
-      fail-fast: false
-      matrix: 
-        model-name: ${{ fromJson(needs.get_modified_models.outputs.matrix) }}
-
-    steps:
-      - name: Check out code
-        uses: actions/checkout@v4
-      
-      - name: Install locally transformers & other libs
-        run: |
-          apt install sudo
-          sudo -H pip install --upgrade pip
-          sudo -H pip uninstall -y transformers 
-          sudo -H pip install -U -e ".[testing]" 
-          MAX_JOBS=4 pip install flash-attn --no-build-isolation
-          pip install bitsandbytes
-      
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-      
-      - name: Show installed libraries and their versions
-        run: pip freeze
-      
-      - name: Run FA2 tests
-        id: run_fa2_tests
-        run:
-          pytest -rsfE -m "flash_attn_test" --make-reports=${{ matrix.model-name }}_fa2_tests/ tests/${{ matrix.model-name }}/test_modeling_*
-      
-      - name: "Test suite reports artifacts: ${{ matrix.model-name }}_fa2_tests"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.model-name }}_fa2_tests
-          path: /transformers/reports/${{ matrix.model-name }}_fa2_tests
-      
-      - name: Post to Slack
-        if: always()
-        uses: huggingface/hf-workflows/.github/actions/post-slack@main
-        with:
-          slack_channel: ${{ env.OUTPUT_SLACK_CHANNEL_ID }}
-          title: 🤗 Results of the FA2 tests - ${{ matrix.model-name }}
-          status: ${{ steps.run_fa2_tests.conclusion}}
-          slack_token: ${{ secrets.CI_SLACK_BOT_TOKEN }}
-      
-      - name: Run integration tests
-        id: run_integration_tests
-        if: always()
-        run:
-          pytest -rsfE -k "IntegrationTest"  --make-reports=tests_integration_${{ matrix.model-name }} tests/${{ matrix.model-name }}/test_modeling_*
-      
-      - name: "Test suite reports artifacts: tests_integration_${{ matrix.model-name }}"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: tests_integration_${{ matrix.model-name }}
-          path: /transformers/reports/tests_integration_${{ matrix.model-name }}
-
-      - name: Post to Slack
-        if: always()
-        uses: huggingface/hf-workflows/.github/actions/post-slack@main 
-        with:
-          slack_channel: ${{ env.OUTPUT_SLACK_CHANNEL_ID }}
-          title: 🤗 Results of the Integration tests - ${{ matrix.model-name }}
-          status: ${{ steps.run_integration_tests.conclusion}}
-          slack_token: ${{ secrets.CI_SLACK_BOT_TOKEN }}
-
-      - name: Tailscale # In order to be able to SSH when a test fails
-        if: ${{ runner.debug == '1'}}
-        uses: huggingface/tailscale-action@v1
-        with:
-          authkey: ${{ secrets.TAILSCALE_SSH_AUTHKEY }}
-          slackChannel: ${{ secrets.SLACK_CIFEEDBACK_CHANNEL }}
-          slackToken: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
-          waitForSSH: true
-
-  benchmark:
-    name: Benchmark workflow
-    needs: get_modified_models
-    if: ${{ needs.get_modified_models.outputs.matrix != '[]' && needs.get_modified_models.outputs.matrix != '' && fromJson(needs.get_modified_models.outputs.matrix)[0] != null }}
-    uses: ./.github/workflows/benchmark.yml
-    secrets: inherit

.github/workflows/release-conda.yml

@@ -12,7 +12,7 @@ env:
 
 jobs:
   build_and_package:
-    runs-on: ubuntu-22.04
+    runs-on: ubuntu-latest
     defaults:
       run:
         shell: bash -l {0}

.github/workflows/self-nightly-caller.yml

@@ -1,43 +0,0 @@
-name: Self-hosted runner (nightly-ci)
-
-
-on:
-  repository_dispatch:
-  schedule:
-    - cron: "17 2 * * *"
-  push:
-    branches:
-      - run_nightly_ci*
-
-jobs:
-  build_nightly_ci_images:
-    name: Build Nightly CI Docker Images
-    if: (github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_nightly_ci'))
-    uses: ./.github/workflows/build-nightly-ci-docker-images.yml
-    secrets: inherit
-
-  model-ci:
-    name: Model CI
-    needs: [build_nightly_ci_images]
-    uses: ./.github/workflows/self-scheduled.yml
-    with:
-      job: run_models_gpu
-      slack_report_channel: "#transformers-ci-past-future"
-      runner: ci
-      docker: huggingface/transformers-all-latest-torch-nightly-gpu
-      ci_event: Nightly CI
-    secrets: inherit
-
-  deepspeed-ci:
-    name: DeepSpeed CI
-    needs: [build_nightly_ci_images]
-    uses: ./.github/workflows/self-scheduled.yml
-    with:
-      job: run_torch_cuda_extensions_gpu
-      slack_report_channel: "#transformers-ci-past-future"
-      runner: ci
-      # test deepspeed nightly build with the latest release torch
-      docker: huggingface/transformers-pytorch-deepspeed-latest-gpu
-      ci_event: Nightly CI
-      working-directory-prefix: /workspace
-    secrets: inherit

.github/workflows/self-nightly-past-ci-caller.yml

@@ -1,132 +0,0 @@
-name: Self-hosted runner (nightly-past-ci-caller)
-
-on:
-  schedule:
-    - cron: "17 2,14 * * *"
-  push:
-    branches:
-      - run_past_ci*
-
-jobs:
-  get_number:
-    name: Get number
-    runs-on: ubuntu-22.04
-    outputs:
-      run_number: ${{ steps.get_number.outputs.run_number }}
-    steps:
-      - name: Get number
-        id: get_number
-        run: |
-          echo "${{ github.run_number }}"
-          echo "$(python3 -c 'print(int(${{ github.run_number }}) % 10)')"
-          echo "run_number=$(python3 -c 'print(int(${{ github.run_number }}) % 10)')" >> $GITHUB_OUTPUT
-
-  run_past_ci_pytorch_1-13:
-    name: PyTorch 1.13
-    needs: get_number
-    if: needs.get_number.outputs.run_number == 0 && (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
-    uses: ./.github/workflows/self-past-caller.yml
-    with:
-      framework: pytorch
-      version: "1.13"
-      sha: ${{ github.sha }}
-    secrets: inherit
-
-  run_past_ci_pytorch_1-12:
-    name: PyTorch 1.12
-    needs: get_number
-    if: needs.get_number.outputs.run_number == 1 && (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
-    uses: ./.github/workflows/self-past-caller.yml
-    with:
-      framework: pytorch
-      version: "1.12"
-      sha: ${{ github.sha }}
-    secrets: inherit
-
-  run_past_ci_pytorch_1-11:
-    name: PyTorch 1.11
-    needs: get_number
-    if: needs.get_number.outputs.run_number == 2 && (cancelled() != true) && ((github.event_name == 'schedule') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci')))
-    uses: ./.github/workflows/self-past-caller.yml
-    with:
-      framework: pytorch
-      version: "1.11"
-      sha: ${{ github.sha }}
-    secrets: inherit
-
-  run_past_ci_tensorflow_2-11:
-    name: TensorFlow 2.11
-    needs: get_number
-    if: needs.get_number.outputs.run_number == 3 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    uses: ./.github/workflows/self-past-caller.yml
-    with:
-      framework: tensorflow
-      version: "2.11"
-      sha: ${{ github.sha }}
-    secrets: inherit
-
-  run_past_ci_tensorflow_2-10:
-    name: TensorFlow 2.10
-    needs: get_number
-    if: needs.get_number.outputs.run_number == 4 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    uses: ./.github/workflows/self-past-caller.yml
-    with:
-      framework: tensorflow
-      version: "2.10"
-      sha: ${{ github.sha }}
-    secrets: inherit
-
-  run_past_ci_tensorflow_2-9:
-    name: TensorFlow 2.9
-    needs: get_number
-    if: needs.get_number.outputs.run_number == 5 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    uses: ./.github/workflows/self-past-caller.yml
-    with:
-      framework: tensorflow
-      version: "2.9"
-      sha: ${{ github.sha }}
-    secrets: inherit
-
-  run_past_ci_tensorflow_2-8:
-    name: TensorFlow 2.8
-    needs: get_number
-    if: needs.get_number.outputs.run_number == 6 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    uses: ./.github/workflows/self-past-caller.yml
-    with:
-      framework: tensorflow
-      version: "2.8"
-      sha: ${{ github.sha }}
-    secrets: inherit
-
-  run_past_ci_tensorflow_2-7:
-    name: TensorFlow 2.7
-    needs: get_number
-    if: needs.get_number.outputs.run_number == 7 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    uses: ./.github/workflows/self-past-caller.yml
-    with:
-      framework: tensorflow
-      version: "2.7"
-      sha: ${{ github.sha }}
-    secrets: inherit
-
-  run_past_ci_tensorflow_2-6:
-    name: TensorFlow 2.6
-    needs: get_number
-    if: needs.get_number.outputs.run_number == 8 && (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    uses: ./.github/workflows/self-past-caller.yml
-    with:
-      framework: tensorflow
-      version: "2.6"
-      sha: ${{ github.sha }}
-    secrets: inherit
-
-  run_past_ci_tensorflow_2-5:
-    name: TensorFlow 2.5
-    needs: get_number
-    if: needs.get_number.outputs.run_number == 9 &&  (cancelled() != true) && ((github.event_name == 'push') && startsWith(github.ref_name, 'run_past_ci'))
-    uses: ./.github/workflows/self-past-caller.yml
-    with:
-      framework: tensorflow
-      version: "2.5"
-      sha: ${{ github.sha }}
-    secrets: inherit

.github/workflows/self-nightly-scheduled.yml

@@ -0,0 +1,264 @@
+name: Self-hosted runner (nightly)
+
+# Note that each job's dependencies go into a corresponding docker file.
+#
+# For example for `run_all_tests_torch_cuda_extensions_gpu` the docker image is
+# `huggingface/transformers-pytorch-deepspeed-latest-gpu`, which can be found at
+# `docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile`
+
+on:
+  repository_dispatch:
+# Disable temporarily until the test suite can be run under 12 hours.
+#  schedule:
+#    - cron: "0 16 * * *"
+
+env:
+  HF_HOME: /mnt/cache
+  TRANSFORMERS_IS_CI: yes
+  OMP_NUM_THREADS: 8
+  MKL_NUM_THREADS: 8
+  RUN_SLOW: yes
+  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
+  TF_FORCE_GPU_ALLOW_GROWTH: true
+  RUN_PT_TF_CROSS_TESTS: 1
+
+jobs:
+  run_check_runners:
+    name: Check Runners
+    strategy:
+      matrix:
+        machine_type: [single-gpu, multi-gpu]
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    container:
+      image: huggingface/transformers-all-latest-torch-nightly-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+  setup:
+    name: Setup
+    needs: run_check_runners
+    strategy:
+      matrix:
+        machine_type: [single-gpu, multi-gpu]
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    container:
+      image: huggingface/transformers-all-latest-torch-nightly-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    outputs:
+      matrix: ${{ steps.set-matrix.outputs.matrix }}
+    steps:
+      - name: Update clone
+        working-directory: /transformers
+        run: |
+          git fetch && git checkout ${{ github.sha }}
+
+      - name: Cleanup
+        working-directory: /transformers
+        run: |
+          rm -rf tests/__pycache__
+          rm -rf tests/models/__pycache__
+          rm -rf reports
+
+      - id: set-matrix
+        name: Identify models to test
+        working-directory: /transformers/tests
+        run: |
+          echo "::set-output name=matrix::$(python3 -c 'import os; tests = os.getcwd(); model_tests = os.listdir(os.path.join(tests, "models")); d1 = sorted(list(filter(os.path.isdir, os.listdir(tests)))); d2 = sorted(list(filter(os.path.isdir, [f"models/{x}" for x in model_tests]))); d1.remove("models"); d = d2 + d1; print(d)')"
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+  run_tests_single_gpu:
+    name: Model tests
+    strategy:
+      fail-fast: false
+      matrix:
+        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
+        machine_type: [single-gpu]
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    container:
+      image: huggingface/transformers-all-latest-torch-nightly-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    needs: [run_check_runners, setup]
+    steps:
+      - name: Echo folder ${{ matrix.folders }}
+        shell: bash
+        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
+        # set the artifact folder names (because the character `/` is not allowed).
+        run: |
+          echo "${{ matrix.folders }}"
+          matrix_folders=${{ matrix.folders }}
+          matrix_folders=${matrix_folders/'models/'/'models_'}
+          echo "$matrix_folders"
+          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
+
+      - name: Update clone
+        working-directory: /transformers
+        run: git fetch && git checkout ${{ github.sha }}
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Environment
+        working-directory: /transformers
+        run: |
+          python3 utils/print_env.py
+
+      - name: Run all tests on GPU
+        working-directory: /transformers
+        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
+
+      - name: Failure short reports
+        if: ${{ failure() }}
+        continue-on-error: true
+        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
+          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
+
+  run_tests_multi_gpu:
+    name: Model tests
+    strategy:
+      fail-fast: false
+      matrix:
+        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
+        machine_type: [multi-gpu]
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    container:
+      image: huggingface/transformers-all-latest-torch-nightly-gpu
+      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    needs: [run_check_runners, setup]
+    steps:
+      - name: Echo folder ${{ matrix.folders }}
+        shell: bash
+        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
+        # set the artifact folder names (because the character `/` is not allowed).
+        run: |
+          echo "${{ matrix.folders }}"
+          matrix_folders=${{ matrix.folders }}
+          matrix_folders=${matrix_folders/'models/'/'models_'}
+          echo "$matrix_folders"
+          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
+
+      - name: Update clone
+        working-directory: /transformers
+        run: git fetch && git checkout ${{ github.sha }}
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Environment
+        working-directory: /transformers
+        run: |
+          python3 utils/print_env.py
+
+      - name: Run all tests on GPU
+        working-directory: /transformers
+        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
+
+      - name: Failure short reports
+        if: ${{ failure() }}
+        continue-on-error: true
+        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
+          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
+
+  run_all_tests_torch_cuda_extensions_gpu:
+    name: Torch CUDA extension tests
+    strategy:
+      fail-fast: false
+      matrix:
+        machine_type: [single-gpu, multi-gpu]
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    needs: [run_check_runners, setup]
+    container:
+      image: huggingface/transformers-pytorch-deepspeed-nightly-gpu
+      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Update clone
+        working-directory: /workspace/transformers
+        run: git fetch && git checkout ${{ github.sha }}
+
+      - name: Remove cached torch extensions
+        run: rm -rf /github/home/.cache/torch_extensions/
+
+      # To avoid unknown test failures
+      - name: Pre build DeepSpeed *again*
+        working-directory: /workspace
+        run: |
+          python3 -m pip uninstall -y deepspeed
+          rm -rf DeepSpeed
+          git clone https://github.com/microsoft/DeepSpeed && cd DeepSpeed && rm -rf build
+          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 DS_BUILD_AIO=1 DS_BUILD_UTILS=1 python3 -m pip install . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Environment
+        working-directory: /workspace/transformers
+        run: |
+          python utils/print_env.py
+
+      - name: Run all tests on GPU
+        working-directory: /workspace/transformers
+        run: |
+          python -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu tests/deepspeed tests/extended
+
+      - name: Failure short reports
+        if: ${{ failure() }}
+        continue-on-error: true
+        run: cat /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu/failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: ${{ matrix.machine_type }}_run_tests_torch_cuda_extensions_gpu_test_reports
+          path: /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu
+
+  send_results:
+    name: Send results to webhook
+    runs-on: ubuntu-latest
+    if: always()
+    needs: [run_check_runners, setup, run_tests_single_gpu, run_tests_multi_gpu, run_all_tests_torch_cuda_extensions_gpu]
+    steps:
+      - name: Preliminary job status
+        shell: bash
+        # For the meaning of these environment variables, see the job `Setup`
+        run: |
+          echo "Runner status: ${{ needs.run_check_runners.result }}"
+          echo "Setup status: ${{ needs.setup.result }}"
+
+      - uses: actions/checkout@v2
+      - uses: actions/download-artifact@v2
+      - name: Send message to Slack
+        env:
+          CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
+          CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID }}
+          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
+          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
+          CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_PAST_FUTURE }}
+          CI_EVENT: nightly-build
+          SETUP_STATUS: ${{ needs.setup.result }}
+          RUNNER_STATUS: ${{ needs.run_check_runners.result }}
+        # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
+        # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
+        run: |
+          pip install slack_sdk
+          python utils/notification_service.py "${{ needs.setup.outputs.matrix }}"

.github/workflows/self-past-caller.yml

@@ -1,40 +1,136 @@
-name: Self-hosted runner (past-ci)
-
+name: Self-hosted runner (past-ci-caller)
 
 on:
-  workflow_call:
-    inputs:
-      framework:
-        required: true
-        type: string
-      version:
-        required: true
-        type: string
-      # Use this to control the commit to test against
-      sha:
-        default: 'main'
-        required: false
-        type: string
+  push:
+    branches:
+      - run-past-ci*
 
 jobs:
-  model-ci:
-    name: Model CI
-    uses: ./.github/workflows/self-scheduled.yml
+  run_past_ci_pytorch_1-11:
+    name: PyTorch 1.11
+    if: always()
+    uses: ./.github/workflows/self-past.yml
     with:
-      job: run_models_gpu
-      slack_report_channel: "#transformers-ci-past-future"
-      runner: past-ci
-      docker: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
-      ci_event: Past CI - ${{ inputs.framework }}-${{ inputs.version }}
+      framework: pytorch
+      version: "1.11"
     secrets: inherit
 
-  deepspeed-ci:
-    name: DeepSpeed CI
-    uses: ./.github/workflows/self-scheduled.yml
+  run_past_ci_pytorch_1-10:
+    name: PyTorch 1.10
+    if: always()
+    needs: [run_past_ci_pytorch_1-11]
+    uses: ./.github/workflows/self-past.yml
     with:
-      job: run_torch_cuda_extensions_gpu
-      slack_report_channel: "#transformers-ci-past-future"
-      runner: past-ci
-      docker: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
-      ci_event: Past CI - ${{ inputs.framework }}-${{ inputs.version }}
+      framework: pytorch
+      version: "1.10"
     secrets: inherit
+
+  run_past_ci_pytorch_1-9:
+    name: PyTorch 1.9
+    if: always()
+    needs: [run_past_ci_pytorch_1-10]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: pytorch
+      version: "1.9"
+    secrets: inherit
+
+  run_past_ci_pytorch_1-8:
+    name: PyTorch 1.8
+    if: always()
+    needs: [run_past_ci_pytorch_1-9]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: pytorch
+      version: "1.8"
+    secrets: inherit
+
+  run_past_ci_pytorch_1-7:
+    name: PyTorch 1.7
+    if: always()
+    needs: [run_past_ci_pytorch_1-8]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: pytorch
+      version: "1.7"
+    secrets: inherit
+
+  run_past_ci_pytorch_1-6:
+    name: PyTorch 1.6
+    if: always()
+    needs: [run_past_ci_pytorch_1-7]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: pytorch
+      version: "1.6"
+    secrets: inherit
+
+  run_past_ci_pytorch_1-5:
+    name: PyTorch 1.5
+    if: always()
+    needs: [run_past_ci_pytorch_1-6]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: pytorch
+      version: "1.5"
+    secrets: inherit
+
+  run_past_ci_pytorch_1-4:
+    name: PyTorch 1.4
+    if: always()
+    needs: [run_past_ci_pytorch_1-5]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: pytorch
+      version: "1.4"
+    secrets: inherit
+
+  run_past_ci_tensorflow_2-8:
+    name: TensorFlow 2.8
+    if: always()
+    needs: [run_past_ci_pytorch_1-4]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: tensorflow
+      version: "2.8"
+    secrets: inherit
+
+  run_past_ci_tensorflow_2-7:
+    name: TensorFlow 2.7
+    if: always()
+    needs: [run_past_ci_tensorflow_2-8]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: tensorflow
+      version: "2.7"
+    secrets: inherit
+
+  run_past_ci_tensorflow_2-6:
+    name: TensorFlow 2.6
+    if: always()
+    needs: [run_past_ci_tensorflow_2-7]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: tensorflow
+      version: "2.6"
+    secrets: inherit
+
+  run_past_ci_tensorflow_2-5:
+    name: TensorFlow 2.5
+    if: always()
+    needs: [run_past_ci_tensorflow_2-6]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: tensorflow
+      version: "2.5"
+    secrets: inherit
+
+  run_past_ci_tensorflow_2-4:
+    name: TensorFlow 2.4
+    if: always()
+    needs: [run_past_ci_tensorflow_2-5]
+    uses: ./.github/workflows/self-past.yml
+    with:
+      framework: tensorflow
+      version: "2.4"
+    secrets: inherit

.github/workflows/self-past.yml

@@ -0,0 +1,216 @@
+name: Self-hosted runner (past)
+
+# Note that each job's dependencies go into a corresponding docker file.
+#
+# For example for `run_all_tests_torch_cuda_extensions_gpu` the docker image is
+# `huggingface/transformers-pytorch-deepspeed-latest-gpu`, which can be found at
+# `docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile`
+
+on:
+  workflow_call:
+    inputs:
+      framework:
+        required: true
+        type: string
+      version:
+        required: true
+        type: string
+
+env:
+  HF_HOME: /mnt/cache
+  TRANSFORMERS_IS_CI: yes
+  OMP_NUM_THREADS: 8
+  MKL_NUM_THREADS: 8
+  RUN_SLOW: yes
+  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
+  TF_FORCE_GPU_ALLOW_GROWTH: true
+  RUN_PT_TF_CROSS_TESTS: 1
+
+jobs:
+  setup:
+    name: Setup
+    runs-on: ubuntu-latest
+    outputs:
+      matrix: ${{ steps.set-matrix.outputs.matrix }}
+    steps:
+      - name: Checkout transformers
+        uses: actions/checkout@v2
+        with:
+          fetch-depth: 2
+
+      - name: Cleanup
+        run: |
+          rm -rf tests/__pycache__
+          rm -rf tests/models/__pycache__
+          rm -rf reports
+
+      - id: set-matrix
+        name: Identify models to test
+        run: |
+          cd tests
+          echo "::set-output name=matrix::$(python3 -c 'import os; tests = os.getcwd(); model_tests = os.listdir(os.path.join(tests, "models")); d1 = sorted(list(filter(os.path.isdir, os.listdir(tests)))); d2 = sorted(list(filter(os.path.isdir, [f"models/{x}" for x in model_tests]))); d1.remove("models"); d = d2 + d1; print(d)')"
+
+  run_check_runners:
+    name: Check Runners
+    needs: setup
+    strategy:
+      matrix:
+        machine_type: [single-gpu, multi-gpu]
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
+    container:
+      image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+  run_tests_single_gpu:
+    name: Model tests
+    strategy:
+      fail-fast: false
+      matrix:
+        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
+        machine_type: [single-gpu]
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
+    container:
+      image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    needs: [setup, run_check_runners]
+    steps:
+      - name: Update clone
+        working-directory: /transformers
+        run: git fetch && git checkout ${{ github.sha }}
+
+      - name: Echo folder ${{ matrix.folders }}
+        shell: bash
+        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
+        # set the artifact folder names (because the character `/` is not allowed).
+        run: |
+          echo "${{ matrix.folders }}"
+          matrix_folders=${{ matrix.folders }}
+          matrix_folders=${matrix_folders/'models/'/'models_'}
+          echo "$matrix_folders"
+          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Environment
+        working-directory: /transformers
+        run: |
+          python3 utils/print_env.py
+
+      - name: Run all tests on GPU
+        working-directory: /transformers
+        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
+
+      - name: Failure short reports
+        if: ${{ failure() }}
+        continue-on-error: true
+        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
+          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
+
+  run_tests_multi_gpu:
+    name: Model tests
+    strategy:
+      fail-fast: false
+      matrix:
+        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
+        machine_type: [multi-gpu]
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker-past-ci') }}
+    container:
+      image: huggingface/transformers-${{ inputs.framework }}-past-${{ inputs.version }}-gpu
+      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    needs: [setup, run_check_runners]
+    steps:
+      - name: Update clone
+        working-directory: /transformers
+        run: git fetch && git checkout ${{ github.sha }}
+
+      - name: Echo folder ${{ matrix.folders }}
+        shell: bash
+        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
+        # set the artifact folder names (because the character `/` is not allowed).
+        run: |
+          echo "${{ matrix.folders }}"
+          matrix_folders=${{ matrix.folders }}
+          matrix_folders=${matrix_folders/'models/'/'models_'}
+          echo "$matrix_folders"
+          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Environment
+        working-directory: /transformers
+        run: |
+          python3 utils/print_env.py
+
+      - name: Run all tests on GPU
+        working-directory: /transformers
+        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
+
+      - name: Failure short reports
+        if: ${{ failure() }}
+        continue-on-error: true
+        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
+          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
+
+  send_results:
+    name: Send results to webhook
+    runs-on: ubuntu-latest
+    if: always()
+    needs: [setup, run_check_runners, run_tests_single_gpu, run_tests_multi_gpu]
+    steps:
+      - name: Preliminary job status
+        shell: bash
+        # For the meaning of these environment variables, see the job `Setup`
+        run: |
+          echo "Runner status: ${{ needs.run_check_runners.result }}"
+          echo "Setup status: ${{ needs.setup.result }}"
+
+      - uses: actions/checkout@v2
+      - uses: actions/download-artifact@v2
+
+      # Create a directory to store test failure tables in the next step
+      - name: Create directory
+        run: mkdir test_failure_tables
+
+      - name: Send message to Slack
+        env:
+          CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
+          CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID }}
+          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
+          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
+          CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_PAST_FUTURE }}
+          CI_EVENT: Past CI - ${{ inputs.framework }}-${{ inputs.version }}
+          SETUP_STATUS: ${{ needs.setup.result }}
+          RUNNER_STATUS: ${{ needs.run_check_runners.result }}
+        # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
+        # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
+        run: |
+          pip install slack_sdk
+          python utils/notification_service.py "${{ needs.setup.outputs.matrix }}"
+
+      # Upload complete failure tables, as they might be big and only truncated versions could be sent to Slack.
+      - name: Failure table artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: test_failure_tables_${{ inputs.framework }}-${{ inputs.version }}
+          path: test_failure_tables

.github/workflows/self-pr-slow-ci.yml

@@ -1,135 +0,0 @@
-name: PR slow CI
-
-on:
-  pull_request:
-    paths:
-      - "src/transformers/models/*/modeling_*.py"
-      - "tests/**/test_*.py"
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  HF_HOME: /mnt/cache
-  TRANSFORMERS_IS_CI: yes
-  OMP_NUM_THREADS: 8
-  MKL_NUM_THREADS: 8
-  RUN_SLOW: yes
-  # For gated repositories, we still need to agree to share information on the Hub repo. page in order to get access.
-  # This token is created under the bot `hf-transformers-bot`.
-  HF_HUB_READ_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}
-  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
-  TF_FORCE_GPU_ALLOW_GROWTH: true
-  RUN_PT_TF_CROSS_TESTS: 1
-  CUDA_VISIBLE_DEVICES: 0,1
-
-jobs:
-  find_models_to_run:
-      runs-on: ubuntu-22.04
-      name: Find models to run slow tests
-      # Triggered only if the required label `run-slow` is added
-      if: ${{ contains(github.event.pull_request.labels.*.name, 'run-slow') }}
-      outputs:
-        models: ${{ steps.models_to_run.outputs.models }}
-      steps:
-        - uses: actions/checkout@v4
-          with:
-            fetch-depth: "0"
-            ref: ${{ github.event.pull_request.head.sha }}
-
-        - name: Get commit message
-          run: |
-            echo "commit_message=$(git show -s --format=%s)" >> $GITHUB_ENV
-
-        - name: Get models to run slow tests
-          run: |
-            echo "${{ env.commit_message }}"
-            python -m pip install GitPython
-            python utils/pr_slow_ci_models.py --commit_message "${{ env.commit_message }}" | tee output.txt
-            echo "models=$(tail -n 1 output.txt)" >> $GITHUB_ENV
-
-        - name: Models to run slow tests
-          id: models_to_run
-          run: |
-            echo "${{ env.models }}"
-            echo "models=${{ env.models }}" >> $GITHUB_OUTPUT
-
-  run_models_gpu:
-      name: Run all tests for the model
-      # Triggered only `find_models_to_run` is triggered (label `run-slow` is added) which gives the models to run
-      # (either a new model PR or via a commit message)
-      if: ${{ needs.find_models_to_run.outputs.models != '[]' }}
-      needs: find_models_to_run
-      strategy:
-        fail-fast: false
-        matrix:
-          folders: ${{ fromJson(needs.find_models_to_run.outputs.models) }}
-          machine_type: [single-gpu, multi-gpu]
-      runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, ci]
-      container:
-        image: huggingface/transformers-all-latest-gpu
-        options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-      steps:
-      - name: Echo input and matrix info
-        shell: bash
-        run: |
-          echo "${{ matrix.folders }}"
-
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git fetch origin pull/${{ github.event.pull_request.number }}/head:pull/${{ github.event.pull_request.number }}/merge && git checkout pull/${{ github.event.pull_request.number }}/merge
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: |
-          export CUDA_VISIBLE_DEVICES="$(python3 utils/set_cuda_devices_for_ci.py --test_folder ${{ matrix.folders }})"
-          echo $CUDA_VISIBLE_DEVICES
-          python3 -m pytest -v -rsfE --make-reports=${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports tests/${{ matrix.folders }}
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports/failures_short.txt
-
-      - name: Make sure report directory exists
-        shell: bash
-        run: |
-          mkdir -p /transformers/reports/${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports
-          echo "hello" > /transformers/reports/${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports/hello.txt
-          echo "${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports"
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports

.github/workflows/self-push-amd-mi210-caller.yml

@@ -1,25 +0,0 @@
-name: Self-hosted runner (AMD mi210 CI caller)
-
-on:
-  workflow_run:
-    workflows: ["Self-hosted runner (push-caller)"]
-    branches: ["main"]
-    types: [completed]
-  push:
-    branches:
-      - run_amd_push_ci_caller*
-    paths:
-      - "src/**"
-      - "tests/**"
-      - ".github/**"
-      - "templates/**"
-      - "utils/**"
-
-jobs:
-  run_amd_ci:
-    name: AMD mi210
-    if: (cancelled() != true) && ((github.event_name == 'workflow_run') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_amd_push_ci_caller')))
-    uses: ./.github/workflows/self-push-amd.yml
-    with:
-      gpu_flavor: mi210
-    secrets: inherit

.github/workflows/self-push-amd-mi250-caller.yml

@@ -1,25 +0,0 @@
-name: Self-hosted runner (AMD mi250 CI caller)
-
-on:
-  workflow_run:
-    workflows: ["Self-hosted runner (push-caller)"]
-    branches: ["main"]
-    types: [completed]
-  push:
-    branches:
-      - run_amd_push_ci_caller*
-    paths:
-      - "src/**"
-      - "tests/**"
-      - ".github/**"
-      - "templates/**"
-      - "utils/**"
-
-jobs:
-  run_amd_ci:
-    name: AMD mi250
-    if: (cancelled() != true) && ((github.event_name == 'workflow_run') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_amd_push_ci_caller')))
-    uses: ./.github/workflows/self-push-amd.yml
-    with:
-      gpu_flavor: mi250
-    secrets: inherit

.github/workflows/self-push-amd-mi300-caller.yml

@@ -1,25 +0,0 @@
-name: Self-hosted runner (AMD mi300 CI caller)
-
-on:
-  workflow_run:
-    workflows: ["Self-hosted runner (push-caller)"]
-    branches: ["main"]
-    types: [completed]
-  push:
-    branches:
-      - run_amd_push_ci_caller*
-    paths:
-      - "src/**"
-      - "tests/**"
-      - ".github/**"
-      - "templates/**"
-      - "utils/**"
-
-jobs:
-  run_amd_ci:
-    name: AMD mi300
-    if: (cancelled() != true) && ((github.event_name == 'workflow_run') || ((github.event_name == 'push') && (startsWith(github.ref_name, 'run_amd_push_ci_caller') || startsWith(github.ref_name, 'mi300-ci'))))
-    uses: ./.github/workflows/self-push-amd.yml
-    with:
-      gpu_flavor: mi300
-    secrets: inherit

.github/workflows/self-push-amd.yml

@@ -1,329 +0,0 @@
-name: Self-hosted runner AMD GPU (push)
-
-on:
-  workflow_call:
-    inputs:
-      gpu_flavor:
-        required: true
-        type: string
-
-env:
-  HF_HOME: /mnt/cache
-  TRANSFORMERS_IS_CI: yes
-  OMP_NUM_THREADS: 8
-  MKL_NUM_THREADS: 8
-  PYTEST_TIMEOUT: 60
-  TF_FORCE_GPU_ALLOW_GROWTH: true
-  RUN_PT_TF_CROSS_TESTS: 1
-  HF_HUB_READ_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}
-
-jobs:
-  check_runner_status:
-    name: Check Runner Status
-    runs-on: ubuntu-22.04
-    steps:
-      - name: Checkout transformers
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 2
-
-      - name: Check Runner Status
-        run: python utils/check_self_hosted_runner.py --target_runners amd-mi210-single-gpu-ci-runner-docker --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-
-  check_runners:
-    name: Check Runners
-    needs: check_runner_status
-    strategy:
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: [self-hosted, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
-    container:
-      image: huggingface/transformers-pytorch-amd-gpu-push-ci  # <--- We test only for PyTorch for now
-      options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: ROCM-SMI
-        run: |
-          rocm-smi
-      - name: ROCM-INFO
-        run: |
-          rocminfo  | grep "Agent" -A 14
-      - name: Show ROCR environment
-        run: |
-          echo "ROCR: $ROCR_VISIBLE_DEVICES"
-
-  setup_gpu:
-    name: Setup
-    needs: check_runners
-    strategy:
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: [self-hosted, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
-    container:
-      image: huggingface/transformers-pytorch-amd-gpu-push-ci  # <--- We test only for PyTorch for now
-      options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    outputs:
-      matrix: ${{ steps.set-matrix.outputs.matrix }}
-      test_map: ${{ steps.set-matrix.outputs.test_map }}
-    steps:
-      # Necessary to get the correct branch name and commit SHA for `workflow_run` event
-      # We also take into account the `push` event (we might want to test some changes in a branch)
-      - name: Prepare custom environment variables
-        shell: bash
-        # `CI_BRANCH_PUSH`: The branch name from the push event
-        # `CI_BRANCH_WORKFLOW_RUN`: The name of the branch on which this workflow is triggered by `workflow_run` event
-        # `CI_BRANCH`: The non-empty branch name from the above two (one and only one of them is empty)
-        # `CI_SHA_PUSH`: The commit SHA from the push event
-        # `CI_SHA_WORKFLOW_RUN`: The commit SHA that triggers this workflow by `workflow_run` event
-        # `CI_SHA`: The non-empty commit SHA from the above two (one and only one of them is empty)
-        run: |
-          CI_BRANCH_PUSH=${{ github.event.ref }}
-          CI_BRANCH_PUSH=${CI_BRANCH_PUSH/'refs/heads/'/''}
-          CI_BRANCH_WORKFLOW_RUN=${{ github.event.workflow_run.head_branch }}
-          CI_SHA_PUSH=${{ github.event.head_commit.id }}
-          CI_SHA_WORKFLOW_RUN=${{ github.event.workflow_run.head_sha }}
-          echo $CI_BRANCH_PUSH
-          echo $CI_BRANCH_WORKFLOW_RUN
-          echo $CI_SHA_PUSH
-          echo $CI_SHA_WORKFLOW_RUN
-          [[ ! -z "$CI_BRANCH_PUSH" ]] && echo "CI_BRANCH=$CI_BRANCH_PUSH" >> $GITHUB_ENV || echo "CI_BRANCH=$CI_BRANCH_WORKFLOW_RUN" >> $GITHUB_ENV
-          [[ ! -z "$CI_SHA_PUSH" ]] && echo "CI_SHA=$CI_SHA_PUSH" >> $GITHUB_ENV || echo "CI_SHA=$CI_SHA_WORKFLOW_RUN" >> $GITHUB_ENV
-
-      - name: print environment variables
-        run: |
-          echo "env.CI_BRANCH = ${{ env.CI_BRANCH }}"
-          echo "env.CI_SHA = ${{ env.CI_SHA }}"
-
-      - name: Update clone using environment variables
-        working-directory: /transformers
-        run: |
-          echo "original branch = $(git branch --show-current)"
-          git fetch && git checkout ${{ env.CI_BRANCH }}
-          echo "updated branch = $(git branch --show-current)"
-          git checkout ${{ env.CI_SHA }}
-          echo "log = $(git log -n 1)"
-
-      - name: Cleanup
-        working-directory: /transformers
-        run: |
-          rm -rf tests/__pycache__
-          rm -rf tests/models/__pycache__
-          rm -rf reports
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Fetch the tests to run
-        working-directory: /transformers
-        # TODO: add `git-python` in the docker images
-        run: |
-          pip install --upgrade git-python
-          python3 utils/tests_fetcher.py --diff_with_last_commit | tee test_preparation.txt
-
-      - name: Report fetched tests
-        uses: actions/upload-artifact@v4
-        with:
-          name: test_fetched
-          path: /transformers/test_preparation.txt
-
-      - id: set-matrix
-        name: Organize tests into models
-        working-directory: /transformers
-        # The `keys` is used as GitHub actions matrix for jobs, i.e. `models/bert`, `tokenization`, `pipeline`, etc.
-        # The `test_map` is used to get the actual identified test files under each key.
-        # If no test to run (so no `test_map.json` file), create a dummy map (empty matrix will fail)
-        run: |
-          if [ -f test_map.json ]; then
-              keys=$(python3 -c 'import json; fp = open("test_map.json"); test_map = json.load(fp); fp.close(); d = list(test_map.keys()); print(d)')
-              test_map=$(python3 -c 'import json; fp = open("test_map.json"); test_map = json.load(fp); fp.close(); print(test_map)')
-          else
-              keys=$(python3 -c 'keys = ["dummy"]; print(keys)')
-              test_map=$(python3 -c 'test_map = {"dummy": []}; print(test_map)')
-          fi
-          echo $keys
-          echo $test_map
-          echo "matrix=$keys" >> $GITHUB_OUTPUT
-          echo "test_map=$test_map" >> $GITHUB_OUTPUT
-
-  run_models_gpu:
-    name: Model tests
-    needs: setup_gpu
-    # `dummy` means there is no test to run
-    if: contains(fromJson(needs.setup_gpu.outputs.matrix), 'dummy') != true
-    strategy:
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup_gpu.outputs.matrix) }}
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: [self-hosted, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
-    container:
-      image: huggingface/transformers-pytorch-amd-gpu-push-ci  # <--- We test only for PyTorch for now
-      options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      # Necessary to get the correct branch name and commit SHA for `workflow_run` event
-      # We also take into account the `push` event (we might want to test some changes in a branch)
-      - name: Prepare custom environment variables
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          CI_BRANCH_PUSH=${{ github.event.ref }}
-          CI_BRANCH_PUSH=${CI_BRANCH_PUSH/'refs/heads/'/''}
-          CI_BRANCH_WORKFLOW_RUN=${{ github.event.workflow_run.head_branch }}
-          CI_SHA_PUSH=${{ github.event.head_commit.id }}
-          CI_SHA_WORKFLOW_RUN=${{ github.event.workflow_run.head_sha }}
-          echo $CI_BRANCH_PUSH
-          echo $CI_BRANCH_WORKFLOW_RUN
-          echo $CI_SHA_PUSH
-          echo $CI_SHA_WORKFLOW_RUN
-          [[ ! -z "$CI_BRANCH_PUSH" ]] && echo "CI_BRANCH=$CI_BRANCH_PUSH" >> $GITHUB_ENV || echo "CI_BRANCH=$CI_BRANCH_WORKFLOW_RUN" >> $GITHUB_ENV
-          [[ ! -z "$CI_SHA_PUSH" ]] && echo "CI_SHA=$CI_SHA_PUSH" >> $GITHUB_ENV || echo "CI_SHA=$CI_SHA_WORKFLOW_RUN" >> $GITHUB_ENV
-
-      - name: print environment variables
-        run: |
-          echo "env.CI_BRANCH = ${{ env.CI_BRANCH }}"
-          echo "env.CI_SHA = ${{ env.CI_SHA }}"
-
-      - name: Update clone using environment variables
-        working-directory: /transformers
-        run: |
-          echo "original branch = $(git branch --show-current)"
-          git fetch && git checkout ${{ env.CI_BRANCH }}
-          echo "updated branch = $(git branch --show-current)"
-          git checkout ${{ env.CI_SHA }}
-          echo "log = $(git log -n 1)"
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          echo "${{ fromJson(needs.setup_gpu.outputs.test_map)[matrix.folders] }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: ROCM-SMI
-        run: |
-          rocm-smi
-      - name: ROCM-INFO
-        run: |
-          rocminfo  | grep "Agent" -A 14
-      - name: Show ROCR environment
-        run: |
-          echo "ROCR: $ROCR_VISIBLE_DEVICES"
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all non-slow selected tests on GPU
-        working-directory: /transformers
-        run: |
-          python3 -m pytest -n 2 --dist=loadfile -v --make-reports=${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports ${{ fromJson(needs.setup_gpu.outputs.test_map)[matrix.folders] }} -m "not not_device_test"
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports
-
-  send_results:
-    name: Send results to webhook
-    runs-on: ubuntu-22.04
-    if: always()
-    needs: [
-        check_runner_status,
-        check_runners,
-        setup_gpu,
-        run_models_gpu,
-#        run_tests_torch_cuda_extensions_single_gpu,
-#        run_tests_torch_cuda_extensions_multi_gpu
-    ]
-    steps:
-      - name: Preliminary job status
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          echo "Runner availability: ${{ needs.check_runner_status.result }}"
-          echo "Setup status: ${{ needs.setup_gpu.result }}"
-          echo "Runner status: ${{ needs.check_runners.result }}"
-
-      # Necessary to get the correct branch name and commit SHA for `workflow_run` event
-      # We also take into account the `push` event (we might want to test some changes in a branch)
-      - name: Prepare custom environment variables
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          CI_BRANCH_PUSH=${{ github.event.ref }}
-          CI_BRANCH_PUSH=${CI_BRANCH_PUSH/'refs/heads/'/''}
-          CI_BRANCH_WORKFLOW_RUN=${{ github.event.workflow_run.head_branch }}
-          CI_SHA_PUSH=${{ github.event.head_commit.id }}
-          CI_SHA_WORKFLOW_RUN=${{ github.event.workflow_run.head_sha }}
-          echo $CI_BRANCH_PUSH
-          echo $CI_BRANCH_WORKFLOW_RUN
-          echo $CI_SHA_PUSH
-          echo $CI_SHA_WORKFLOW_RUN
-          [[ ! -z "$CI_BRANCH_PUSH" ]] && echo "CI_BRANCH=$CI_BRANCH_PUSH" >> $GITHUB_ENV || echo "CI_BRANCH=$CI_BRANCH_WORKFLOW_RUN" >> $GITHUB_ENV
-          [[ ! -z "$CI_SHA_PUSH" ]] && echo "CI_SHA=$CI_SHA_PUSH" >> $GITHUB_ENV || echo "CI_SHA=$CI_SHA_WORKFLOW_RUN" >> $GITHUB_ENV
-
-      - name: print environment variables
-        run: |
-          echo "env.CI_BRANCH = ${{ env.CI_BRANCH }}"
-          echo "env.CI_SHA = ${{ env.CI_SHA }}"
-
-      - uses: actions/checkout@v4
-        # To avoid failure when multiple commits are merged into `main` in a short period of time.
-        # Checking out to an old commit beyond the fetch depth will get an error `fatal: reference is not a tree: ...
-        # (Only required for `workflow_run` event, where we get the latest HEAD on `main` instead of the event commit)
-        with:
-          fetch-depth: 20
-
-      - name: Update clone using environment variables
-        run: |
-          echo "original branch = $(git branch --show-current)"
-          git fetch && git checkout ${{ env.CI_BRANCH }}
-          echo "updated branch = $(git branch --show-current)"
-          git checkout ${{ env.CI_SHA }}
-          echo "log = $(git log -n 1)"
-
-      - uses: actions/download-artifact@v4
-      - name: Send message to Slack
-        env:
-          CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
-          CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID }}
-          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
-          CI_SLACK_CHANNEL_ID_AMD: ${{ secrets.CI_SLACK_CHANNEL_ID_AMD }}
-          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
-          CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_AMD }}
-          ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-          CI_EVENT: Push CI (AMD) - ${{ inputs.gpu_flavor }}
-          CI_TITLE_PUSH: ${{ github.event.head_commit.message }}
-          CI_TITLE_WORKFLOW_RUN: ${{ github.event.workflow_run.head_commit.message }}
-          CI_SHA: ${{ env.CI_SHA }}
-          RUNNER_STATUS: ${{ needs.check_runner_status.result }}
-          RUNNER_ENV_STATUS: ${{ needs.check_runners.result }}
-          SETUP_STATUS: ${{ needs.setup_gpu.result }}
-
-        # We pass `needs.setup_gpu.outputs.matrix` as the argument. A processing in `notification_service.py` to change
-        # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
-        run: |
-          pip install slack_sdk
-          pip show slack_sdk
-          python utils/notification_service.py "${{ needs.setup_gpu.outputs.matrix }}"

.github/workflows/self-push-caller.yml

@@ -14,25 +14,25 @@ on:
 
 jobs:
   check-for-setup:
-      runs-on: ubuntu-22.04
+      runs-on: ubuntu-latest
       name: Check if setup was changed
       outputs:
         changed: ${{ steps.was_changed.outputs.changed }}
       steps:
-        - uses: actions/checkout@v4
+        - uses: actions/checkout@v3
           with: 
             fetch-depth: "2"
         
         - name: Get changed files
           id: changed-files
-          uses: tj-actions/changed-files@v41
+          uses: tj-actions/changed-files@v22.2
         
         - name: Was setup changed 
           id: was_changed
           run: |
             for file in ${{ steps.changed-files.outputs.all_changed_files }}; do
               if [ `basename "${file}"` = "setup.py" ]; then
-                echo "changed=1" >> $GITHUB_OUTPUT
+                echo ::set-output name=changed::"1"
               fi
             done
 
@@ -40,13 +40,11 @@ jobs:
     needs: check-for-setup
     if: (github.event_name == 'push') && (needs.check-for-setup.outputs.changed == '1')
     uses: ./.github/workflows/build-docker-images.yml
-    with:
-      image_postfix: "-push-ci"
     secrets: inherit
 
   run_push_ci:
     name: Trigger Push CI
-    runs-on: ubuntu-22.04
+    runs-on: ubuntu-latest
     if: ${{ always() }}
     needs: build-docker-containers
     steps:

.github/workflows/self-push.yml

@@ -25,18 +25,11 @@ env:
   PYTEST_TIMEOUT: 60
   TF_FORCE_GPU_ALLOW_GROWTH: true
   RUN_PT_TF_CROSS_TESTS: 1
-  CUDA_VISIBLE_DEVICES: 0,1
 
 jobs:
   setup:
     name: Setup
-    strategy:
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, push-ci]
-    container:
-      image: huggingface/transformers-all-latest-gpu-push-ci
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    runs-on: ubuntu-latest
     outputs:
       matrix: ${{ steps.set-matrix.outputs.matrix }}
       test_map: ${{ steps.set-matrix.outputs.test_map }}
@@ -69,8 +62,12 @@ jobs:
           echo "env.CI_BRANCH = ${{ env.CI_BRANCH }}"
           echo "env.CI_SHA = ${{ env.CI_SHA }}"
 
+      - name: Checkout transformers
+        uses: actions/checkout@v2
+        with:
+          fetch-depth: 2
+
       - name: Update clone using environment variables
-        working-directory: /transformers
         run: |
           echo "original branch = $(git branch --show-current)"
           git fetch && git checkout ${{ env.CI_BRANCH }}
@@ -79,32 +76,25 @@ jobs:
           echo "log = $(git log -n 1)"
 
       - name: Cleanup
-        working-directory: /transformers
         run: |
           rm -rf tests/__pycache__
           rm -rf tests/models/__pycache__
           rm -rf reports
 
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
       - name: Fetch the tests to run
-        working-directory: /transformers
         # TODO: add `git-python` in the docker images
         run: |
           pip install --upgrade git-python
-          python3 utils/tests_fetcher.py --diff_with_last_commit | tee test_preparation.txt
+          python utils/tests_fetcher.py --diff_with_last_commit | tee test_preparation.txt
 
       - name: Report fetched tests
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v2
         with:
           name: test_fetched
-          path: /transformers/test_preparation.txt
+          path: test_preparation.txt
 
       - id: set-matrix
         name: Organize tests into models
-        working-directory: /transformers
         # The `keys` is used as GitHub actions matrix for jobs, i.e. `models/bert`, `tokenization`, `pipeline`, etc.
         # The `test_map` is used to get the actual identified test files under each key.
         # If no test to run (so no `test_map.json` file), create a dummy map (empty matrix will fail)
@@ -118,12 +108,27 @@ jobs:
           fi
           echo $keys
           echo $test_map
-          echo "matrix=$keys" >> $GITHUB_OUTPUT
-          echo "test_map=$test_map" >> $GITHUB_OUTPUT
+          echo "::set-output name=matrix::$keys"
+          echo "::set-output name=test_map::$test_map"
+
+  run_check_runners:
+    name: Check Runners
+    needs: setup
+    strategy:
+      matrix:
+        machine_type: [single-gpu, multi-gpu]
+    runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
+    container:
+      image: huggingface/transformers-all-latest-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
 
   run_tests_single_gpu:
     name: Model tests
-    needs: setup
+    needs: [setup, run_check_runners]
     # `dummy` means there is no test to run
     if: contains(fromJson(needs.setup.outputs.matrix), 'dummy') != true
     strategy:
@@ -131,9 +136,9 @@ jobs:
       matrix:
         folders: ${{ fromJson(needs.setup.outputs.matrix) }}
         machine_type: [single-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, push-ci]
+    runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
     container:
-      image: huggingface/transformers-all-latest-gpu-push-ci
+      image: huggingface/transformers-all-latest-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
       # Necessary to get the correct branch name and commit SHA for `workflow_run` event
@@ -168,10 +173,6 @@ jobs:
           git checkout ${{ env.CI_SHA }}
           echo "log = $(git log -n 1)"
 
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
       - name: Echo folder ${{ matrix.folders }}
         shell: bash
         # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
@@ -193,10 +194,6 @@ jobs:
         run: |
           python3 utils/print_env.py
 
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
       - name: Run all non-slow selected tests on GPU
         working-directory: /transformers
         run: |
@@ -207,16 +204,16 @@ jobs:
         continue-on-error: true
         run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
 
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports"
+      - name: Test suite reports artifacts
         if: ${{ always() }}
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v2
         with:
           name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
           path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
 
   run_tests_multi_gpu:
     name: Model tests
-    needs: setup
+    needs: [setup, run_check_runners]
     # `dummy` means there is no test to run
     if: contains(fromJson(needs.setup.outputs.matrix), 'dummy') != true
     strategy:
@@ -224,9 +221,9 @@ jobs:
       matrix:
         folders: ${{ fromJson(needs.setup.outputs.matrix) }}
         machine_type: [multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, push-ci]
+    runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
     container:
-      image: huggingface/transformers-all-latest-gpu-push-ci
+      image: huggingface/transformers-all-latest-gpu
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
       # Necessary to get the correct branch name and commit SHA for `workflow_run` event
@@ -261,10 +258,6 @@ jobs:
           git checkout ${{ env.CI_SHA }}
           echo "log = $(git log -n 1)"
 
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
       - name: Echo folder ${{ matrix.folders }}
         shell: bash
         # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
@@ -286,10 +279,6 @@ jobs:
         run: |
           python3 utils/print_env.py
 
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
       - name: Run all non-slow selected tests on GPU
         env:
           MKL_SERVICE_FORCE_INTEL: 1
@@ -302,24 +291,24 @@ jobs:
         continue-on-error: true
         run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
 
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports"
+      - name: Test suite reports artifacts
         if: ${{ always() }}
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v2
         with:
           name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
           path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
 
   run_tests_torch_cuda_extensions_single_gpu:
     name: Torch CUDA extension tests
-    needs: setup
+    needs: [setup, run_check_runners]
     if: contains(fromJson(needs.setup.outputs.matrix), 'deepspeed') || contains(fromJson(needs.setup.outputs.matrix), 'extended')
     strategy:
       fail-fast: false
       matrix:
         machine_type: [single-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, push-ci]
+    runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
     container:
-      image: huggingface/transformers-pytorch-deepspeed-latest-gpu-push-ci
+      image: huggingface/transformers-pytorch-deepspeed-latest-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
       # Necessary to get the correct branch name and commit SHA for `workflow_run` event
@@ -354,10 +343,6 @@ jobs:
           git checkout ${{ env.CI_SHA }}
           echo "log = $(git log -n 1)"
 
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /workspace/transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
       - name: Remove cached torch extensions
         run: rm -rf /github/home/.cache/torch_extensions/
 
@@ -366,7 +351,7 @@ jobs:
         working-directory: /workspace
         run: |
           python3 -m pip uninstall -y deepspeed
-          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 python3 -m pip install deepspeed --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
+          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 DS_BUILD_AIO=1 DS_BUILD_UTILS=1 python3 -m pip install deepspeed --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
 
       - name: NVIDIA-SMI
         run: |
@@ -377,39 +362,35 @@ jobs:
         run: |
           python utils/print_env.py
 
-      - name: Show installed libraries and their versions
-        working-directory: /workspace/transformers
-        run: pip freeze
-
       - name: Run all non-slow selected tests on GPU
         working-directory: /workspace/transformers
         # TODO: Here we pass all tests in the 2 folders for simplicity. It's better to pass only the identified tests.
         run: |
-          python -m pytest -n 1 --dist=loadfile -v --make-reports=${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports tests/deepspeed tests/extended
+          python -m pytest -n 1 --dist=loadfile -v --make-reports=${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu tests/deepspeed tests/extended
 
       - name: Failure short reports
         if: ${{ failure() }}
         continue-on-error: true
-        run: cat /workspace/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports/failures_short.txt
+        run: cat /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu/failures_short.txt
 
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports"
+      - name: Test suite reports artifacts
         if: ${{ always() }}
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v2
         with:
-          name: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
-          path: /workspace/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
+          name: ${{ matrix.machine_type }}_run_tests_torch_cuda_extensions_gpu_test_reports
+          path: /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu
 
   run_tests_torch_cuda_extensions_multi_gpu:
     name: Torch CUDA extension tests
-    needs: setup
+    needs: [setup, run_check_runners]
     if: contains(fromJson(needs.setup.outputs.matrix), 'deepspeed') || contains(fromJson(needs.setup.outputs.matrix), 'extended')
     strategy:
       fail-fast: false
       matrix:
         machine_type: [multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, push-ci]
+    runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
     container:
-      image: huggingface/transformers-pytorch-deepspeed-latest-gpu-push-ci
+      image: huggingface/transformers-pytorch-deepspeed-latest-gpu
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
       # Necessary to get the correct branch name and commit SHA for `workflow_run` event
@@ -444,10 +425,6 @@ jobs:
           git checkout ${{ env.CI_SHA }}
           echo "log = $(git log -n 1)"
 
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /workspace/transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
       - name: Remove cached torch extensions
         run: rm -rf /github/home/.cache/torch_extensions/
 
@@ -456,7 +433,7 @@ jobs:
         working-directory: /workspace
         run: |
           python3 -m pip uninstall -y deepspeed
-          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 python3 -m pip install deepspeed --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
+          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 DS_BUILD_AIO=1 DS_BUILD_UTILS=1 python3 -m pip install deepspeed --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
 
       - name: NVIDIA-SMI
         run: |
@@ -467,34 +444,31 @@ jobs:
         run: |
           python utils/print_env.py
 
-      - name: Show installed libraries and their versions
-        working-directory: /workspace/transformers
-        run: pip freeze
-
       - name: Run all non-slow selected tests on GPU
         working-directory: /workspace/transformers
         # TODO: Here we pass all tests in the 2 folders for simplicity. It's better to pass only the identified tests.
         run: |
-          python -m pytest -n 1 --dist=loadfile -v --make-reports=${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports tests/deepspeed tests/extended
+          python -m pytest -n 1 --dist=loadfile -v --make-reports=${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu tests/deepspeed tests/extended
 
       - name: Failure short reports
         if: ${{ failure() }}
         continue-on-error: true
-        run: cat /workspace/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports/failures_short.txt
+        run: cat /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu/failures_short.txt
 
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports"
+      - name: Test suite reports artifacts
         if: ${{ always() }}
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v2
         with:
-          name: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
-          path: /workspace/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
+          name: ${{ matrix.machine_type }}_run_tests_torch_cuda_extensions_gpu_test_reports
+          path: /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu
 
   send_results:
     name: Send results to webhook
-    runs-on: ubuntu-22.04
+    runs-on: ubuntu-latest
     if: always()
     needs: [
         setup,
+        run_check_runners,
         run_tests_single_gpu,
         run_tests_multi_gpu,
         run_tests_torch_cuda_extensions_single_gpu,
@@ -506,6 +480,7 @@ jobs:
         # For the meaning of these environment variables, see the job `Setup`
         run: |
           echo "Setup status: ${{ needs.setup.result }}"
+          echo "Runner status: ${{ needs.run_check_runners.result }}"
 
       # Necessary to get the correct branch name and commit SHA for `workflow_run` event
       # We also take into account the `push` event (we might want to test some changes in a branch)
@@ -530,12 +505,7 @@ jobs:
           echo "env.CI_BRANCH = ${{ env.CI_BRANCH }}"
           echo "env.CI_SHA = ${{ env.CI_SHA }}"
 
-      - uses: actions/checkout@v4
-        # To avoid failure when multiple commits are merged into `main` in a short period of time.
-        # Checking out to an old commit beyond the fetch depth will get an error `fatal: reference is not a tree: ...
-        # (Only required for `workflow_run` event, where we get the latest HEAD on `main` instead of the event commit)
-        with:
-          fetch-depth: 20
+      - uses: actions/checkout@v2
 
       - name: Update clone using environment variables
         run: |
@@ -545,7 +515,7 @@ jobs:
           git checkout ${{ env.CI_SHA }}
           echo "log = $(git log -n 1)"
 
-      - uses: actions/download-artifact@v4
+      - uses: actions/download-artifact@v2
       - name: Send message to Slack
         env:
           CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
@@ -553,16 +523,15 @@ jobs:
           CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
           CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
           CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID }}
-          ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
           CI_EVENT: push
           CI_TITLE_PUSH: ${{ github.event.head_commit.message }}
           CI_TITLE_WORKFLOW_RUN: ${{ github.event.workflow_run.head_commit.message }}
           CI_SHA: ${{ env.CI_SHA }}
           SETUP_STATUS: ${{ needs.setup.result }}
+          RUNNER_STATUS: ${{ needs.run_check_runners.result }}
 
         # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
         # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
         run: |
           pip install slack_sdk
-          pip show slack_sdk
           python utils/notification_service.py "${{ needs.setup.outputs.matrix }}"

.github/workflows/self-scheduled-amd-caller.yml

@@ -1,14 +0,0 @@
-name: Self-hosted runner (AMD scheduled CI caller)
-
-on:
-  schedule:
-    - cron: "17 2 * * *"
-
-jobs:
-  run_scheduled_amd_ci:
-    name: Trigger Scheduled AMD CI
-    runs-on: ubuntu-22.04
-    if: ${{ always() }}
-    steps:
-      - name: Trigger scheduled AMD CI via workflow_run
-        run: echo "Trigger scheduled AMD CI via workflow_run"

.github/workflows/self-scheduled-amd-mi210-caller.yml

@@ -1,20 +0,0 @@
-name: Self-hosted runner (AMD mi210 scheduled CI caller)
-
-on:
-  workflow_run:
-    workflows: ["Self-hosted runner (AMD scheduled CI caller)"]
-    branches: ["main"]
-    types: [completed]
-  push:
-    branches:
-      - run_amd_scheduled_ci_caller*
-
-jobs:
-  run_amd_ci:
-    name: AMD mi210
-    if: (cancelled() != true) && ((github.event_name == 'workflow_run') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_amd_scheduled_ci_caller')))
-    uses: ./.github/workflows/self-scheduled-amd.yml
-    with:
-      gpu_flavor: mi210
-      slack_report_channel: "#transformers-ci-daily-amd"
-    secrets: inherit

.github/workflows/self-scheduled-amd-mi250-caller.yml

@@ -1,20 +0,0 @@
-name: Self-hosted runner (AMD mi250 scheduled CI caller)
-
-on:
-  workflow_run:
-    workflows: ["Self-hosted runner (AMD scheduled CI caller)"]
-    branches: ["main"]
-    types: [completed]
-  push:
-    branches:
-      - run_amd_scheduled_ci_caller*
-
-jobs:
-  run_amd_ci:
-    name: AMD mi250
-    if: (cancelled() != true) && ((github.event_name == 'workflow_run') || ((github.event_name == 'push') && startsWith(github.ref_name, 'run_amd_scheduled_ci_caller')))
-    uses: ./.github/workflows/self-scheduled-amd.yml
-    with:
-      gpu_flavor: mi250
-      slack_report_channel: "#transformers-ci-daily-amd"
-    secrets: inherit

.github/workflows/self-scheduled-amd-mi300-caller.yml

@@ -1,21 +0,0 @@
-name: Self-hosted runner (AMD mi300 scheduled CI caller)
-
-on:
-  workflow_run:
-    workflows: ["Self-hosted runner (AMD scheduled CI caller)"]
-    branches: ["main"]
-    types: [completed]
-  push:
-    branches:
-      - run_amd_scheduled_ci_caller*
-
-jobs:
-  run_amd_ci:
-    name: AMD mi300
-    needs: build-docker-containers
-    if: (cancelled() != true) && ((github.event_name == 'workflow_run') || ((github.event_name == 'push') && (startsWith(github.ref_name, 'run_amd_push_ci_caller') || startsWith(github.ref_name, 'mi300-ci'))))
-    uses: ./.github/workflows/self-scheduled-amd.yml
-    with:
-      gpu_flavor: mi300
-      slack_report_channel: "#transformers-ci-daily-amd"
-    secrets: inherit

.github/workflows/self-scheduled-amd.yml

@@ -1,519 +0,0 @@
-name: Self-hosted runner (scheduled-amd)
-
-# Note: For the AMD CI, we rely on a caller workflow and on the workflow_call event to trigger the
-# CI in order to run it on both MI210 and MI250, without having to use matrix here which pushes
-# us towards the limit of allowed jobs on GitHub Actions.
-on:
-  workflow_call:
-    inputs:
-      gpu_flavor:
-        required: true
-        type: string
-
-env:
-  HF_HOME: /mnt/cache
-  TRANSFORMERS_IS_CI: yes
-  OMP_NUM_THREADS: 8
-  MKL_NUM_THREADS: 8
-  RUN_SLOW: yes
-  HF_HUB_READ_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}
-  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
-
-
-# Important note: each job (run_tests_single_gpu, run_tests_multi_gpu, run_examples_gpu, run_pipelines_torch_gpu) requires all the previous jobs before running.
-# This is done so that we avoid parallelizing the scheduled tests, to leave available
-# runners for the push CI that is running on the same machine.
-jobs:
-  check_runner_status:
-    name: Check Runner Status
-    runs-on: ubuntu-22.04
-    steps:
-      - name: Checkout transformers
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 2
-
-      - name: Check Runner Status
-        run: python utils/check_self_hosted_runner.py --target_runners hf-amd-mi210-ci-1gpu-1,hf-amd-mi250-ci-1gpu-1,hf-amd-mi300-ci-1gpu-1 --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-
-  check_runners:
-    name: Check Runners
-    needs: check_runner_status
-    strategy:
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: [self-hosted, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
-    container:
-      image: huggingface/transformers-pytorch-amd-gpu
-      options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: ROCM-SMI
-        run: |
-          rocm-smi
-      - name: ROCM-INFO
-        run: |
-          rocminfo  | grep "Agent" -A 14
-      - name: Show ROCR environment
-        run: |
-          echo "ROCR: $ROCR_VISIBLE_DEVICES"
-
-  setup:
-    name: Setup
-    needs: check_runners
-    strategy:
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: [self-hosted, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
-    container:
-      image: huggingface/transformers-pytorch-amd-gpu
-      options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    outputs:
-      matrix: ${{ steps.set-matrix.outputs.matrix }}
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: |
-          git fetch && git checkout ${{ github.sha }}
-
-      - name: Cleanup
-        working-directory: /transformers
-        run: |
-          rm -rf tests/__pycache__
-          rm -rf tests/models/__pycache__
-          rm -rf reports
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - id: set-matrix
-        name: Identify models to test
-        working-directory: /transformers/tests
-        run: |
-          echo "matrix=$(python3 -c 'import os; tests = os.getcwd(); model_tests = os.listdir(os.path.join(tests, "models")); d1 = sorted(list(filter(os.path.isdir, os.listdir(tests)))); d2 = sorted(list(filter(os.path.isdir, [f"models/{x}" for x in model_tests]))); d1.remove("models"); d = d2 + d1; print(d)')" >> $GITHUB_OUTPUT
-
-      - name: ROCM-SMI
-        run: |
-          rocm-smi
-
-      - name: ROCM-INFO
-        run: |
-          rocminfo  | grep "Agent" -A 14
-      - name: Show ROCR environment
-        run: |
-          echo "ROCR: $ROCR_VISIBLE_DEVICES"
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-  run_models_gpu_single_gpu:
-    name: Single GPU tests
-    strategy:
-      max-parallel: 1  # For now, not to parallelize. Can change later if it works well.
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [single-gpu]
-    runs-on: [self-hosted, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
-    container:
-      image: huggingface/transformers-pytorch-amd-gpu
-      options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
-    steps:
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: ROCM-SMI
-        run: |
-          rocm-smi
-      - name: ROCM-INFO
-        run: |
-          rocminfo  | grep "Agent" -A 14
-      - name: Show ROCR environment
-        run: |
-          echo "ROCR: $ROCR_VISIBLE_DEVICES"
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports tests/${{ matrix.folders }} -m "not not_device_test"
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports
-
-  run_models_gpu_multi_gpu:
-    name: Multi GPU tests
-    strategy:
-      max-parallel: 1
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
-        machine_type: [multi-gpu]
-    runs-on: [self-hosted, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
-    container:
-      image: huggingface/transformers-pytorch-amd-gpu
-      options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
-    steps:
-      - name: Echo folder ${{ matrix.folders }}
-        shell: bash
-        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
-        # set the artifact folder names (because the character `/` is not allowed).
-        run: |
-          echo "${{ matrix.folders }}"
-          matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'models/'/'models_'}
-          echo "$matrix_folders"
-          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
-
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: ROCM-SMI
-        run: |
-          rocm-smi
-      - name: ROCM-INFO
-        run: |
-          rocminfo  | grep "Agent" -A 14
-      - name: Show ROCR environment
-        run: |
-          echo "ROCR: $ROCR_VISIBLE_DEVICES"
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports tests/${{ matrix.folders }} -m "not not_device_test"
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports
-
-  run_examples_gpu:
-    name: Examples tests
-    strategy:
-      fail-fast: false
-      matrix:
-        machine_type: [single-gpu]
-    runs-on: [self-hosted, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
-    container:
-      image: huggingface/transformers-pytorch-amd-gpu
-      options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: ROCM-SMI
-        run: |
-          rocm-smi
-      - name: ROCM-INFO
-        run: |
-          rocminfo  | grep "Agent" -A 14
-      - name: Show ROCR environment
-        run: |
-          echo "ROCR: $ROCR_VISIBLE_DEVICES"
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run examples tests on GPU
-        working-directory: /transformers
-        run: |
-          pip install -r examples/pytorch/_tests_requirements.txt
-          python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_run_examples_gpu_test_reports examples/pytorch -m "not not_device_test"
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_run_examples_gpu_test_reports/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_examples_gpu_test_reports"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_examples_gpu_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_run_examples_gpu_test_reports
-
-  run_pipelines_torch_gpu:
-    name: PyTorch pipelines tests
-    strategy:
-      fail-fast: false
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: [self-hosted, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
-    container:
-      image: huggingface/transformers-pytorch-amd-gpu
-      options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    needs: setup
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: ROCM-SMI
-        run: |
-          rocm-smi
-      - name: ROCM-INFO
-        run: |
-          rocminfo  | grep "Agent" -A 14
-      - name: Show ROCR environment
-        run: |
-          echo "ROCR: $ROCR_VISIBLE_DEVICES"
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all pipeline tests on GPU
-        working-directory: /transformers
-        run: |
-          python3 -m pytest -n 1 -v --dist=loadfile --make-reports=${{ matrix.machine_type }}_run_pipelines_torch_gpu_test_reports tests/pipelines -m "not not_device_test"
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_run_pipelines_torch_gpu_test_reports/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_pipelines_torch_gpu_test_reports"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_pipelines_torch_gpu_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_run_pipelines_torch_gpu_test_reports
-
-  run_torch_cuda_extensions_gpu:
-    name: Torch ROCm deepspeed tests
-    strategy:
-      fail-fast: false
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-
-    runs-on: [self-hosted, amd-gpu, '${{ matrix.machine_type }}', '${{ inputs.gpu_flavor }}']
-    needs: setup
-    container:
-      image: huggingface/transformers-pytorch-deepspeed-amd-gpu
-      options: --device /dev/kfd --device /dev/dri --env ROCR_VISIBLE_DEVICES --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: ROCM-SMI
-        run: |
-          rocm-smi
-      - name: ROCM-INFO
-        run: |
-          rocminfo  | grep "Agent" -A 14
-
-      - name: Show ROCR environment
-        run: |
-          echo "ROCR: $ROCR_VISIBLE_DEVICES"
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: /transformers
-        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports tests/deepspeed tests/extended -m "not not_device_test"
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
-
-  run_extract_warnings:
-    name: Extract warnings in CI artifacts
-    runs-on: ubuntu-22.04
-    if: always()
-    needs: [
-      check_runner_status,
-      check_runners,
-      setup,
-      run_models_gpu_single_gpu,
-      run_models_gpu_multi_gpu,
-      run_examples_gpu,
-      run_pipelines_torch_gpu,
-      run_torch_cuda_extensions_gpu
-    ]
-    steps:
-      - name: Checkout transformers
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 2
-
-      - name: Install transformers
-        run: pip install transformers
-
-      - name: Show installed libraries and their versions
-        run: pip freeze
-
-      - name: Create output directory
-        run: mkdir warnings_in_ci
-
-      - uses: actions/download-artifact@v4
-        with:
-          path: warnings_in_ci
-
-      - name: Show artifacts
-        run: echo "$(python3 -c 'import os; d = os.listdir(); print(d)')"
-        working-directory: warnings_in_ci
-
-      - name: Extract warnings in CI artifacts
-        run: |
-          python3 utils/extract_warnings.py --workflow_run_id ${{ github.run_id }} --output_dir warnings_in_ci --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }} --from_gh
-          echo "$(python3 -c 'import os; import json; fp = open("warnings_in_ci/selected_warnings.json"); d = json.load(fp); d = "\n".join(d) ;print(d)')"
-
-      - name: Upload artifact
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: warnings_in_ci
-          path: warnings_in_ci/selected_warnings.json
-
-  send_results:
-    name: Send results to webhook
-    runs-on: ubuntu-22.04
-    if: always()
-    needs: [
-      check_runner_status,
-      check_runners,
-      setup,
-      run_models_gpu_single_gpu,
-      run_models_gpu_multi_gpu,
-      run_examples_gpu,
-      run_pipelines_torch_gpu,
-      run_torch_cuda_extensions_gpu,
-      run_extract_warnings
-    ]
-    steps:
-      - name: Preliminary job status
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          echo "Runner availability: ${{ needs.check_runner_status.result }}"
-          echo "Runner status: ${{ needs.check_runners.result }}"
-          echo "Setup status: ${{ needs.setup.result }}"
-
-      - uses: actions/checkout@v4
-      - uses: actions/download-artifact@v4
-      - name: Send message to Slack
-        env:
-          CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
-          CI_SLACK_CHANNEL_ID_DAILY_AMD: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY_AMD }}
-          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
-          CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY_AMD }}
-          ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-          CI_EVENT: Scheduled CI (AMD) - ${{ inputs.gpu_flavor }}
-          CI_SHA: ${{ github.sha }}
-          CI_WORKFLOW_REF: ${{ github.workflow_ref }}
-          RUNNER_STATUS: ${{ needs.check_runner_status.result }}
-          RUNNER_ENV_STATUS: ${{ needs.check_runners.result }}
-          SETUP_STATUS: ${{ needs.setup.result }}
-        # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
-        # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
-        run: |
-          sudo apt-get install -y curl
-          pip install slack_sdk
-          pip show slack_sdk
-          python utils/notification_service.py "${{ needs.setup.outputs.matrix }}"
-
-      # Upload complete failure tables, as they might be big and only truncated versions could be sent to Slack.
-      - name: Failure table artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: test_failure_tables
-          path: test_failure_tables

.github/workflows/self-scheduled-caller.yml

@@ -1,78 +0,0 @@
-name: Self-hosted runner (scheduled)
-
-
-on:
-  repository_dispatch:
-  schedule:
-    - cron: "17 2 * * *"
-  push:
-    branches:
-      - run_scheduled_ci*
-
-jobs:
-  model-ci:
-    name: Model CI
-    uses: ./.github/workflows/self-scheduled.yml
-    with:
-      job: run_models_gpu
-      slack_report_channel: "#transformers-ci-daily-models"
-      runner: daily-ci
-      docker: huggingface/transformers-all-latest-gpu
-      ci_event: Daily CI
-    secrets: inherit
-
-  torch-pipeline:
-    name: Torch pipeline CI
-    uses: ./.github/workflows/self-scheduled.yml
-    with:
-      job: run_pipelines_torch_gpu
-      slack_report_channel: "#transformers-ci-daily-pipeline-torch"
-      runner: daily-ci
-      docker: huggingface/transformers-pytorch-gpu
-      ci_event: Daily CI
-    secrets: inherit
-
-  tf-pipeline:
-    name: TF pipeline CI
-    uses: ./.github/workflows/self-scheduled.yml
-    with:
-      job: run_pipelines_tf_gpu
-      slack_report_channel: "#transformers-ci-daily-pipeline-tf"
-      runner: daily-ci
-      docker: huggingface/transformers-tensorflow-gpu
-      ci_event: Daily CI
-    secrets: inherit
-
-  example-ci:
-    name: Example CI
-    uses: ./.github/workflows/self-scheduled.yml
-    with:
-      job: run_examples_gpu
-      slack_report_channel: "#transformers-ci-daily-examples"
-      runner: daily-ci
-      docker: huggingface/transformers-all-latest-gpu
-      ci_event: Daily CI
-    secrets: inherit
-
-  deepspeed-ci:
-    name: DeepSpeed CI
-    uses: ./.github/workflows/self-scheduled.yml
-    with:
-      job: run_torch_cuda_extensions_gpu
-      slack_report_channel: "#transformers-ci-daily-deepspeed"
-      runner: daily-ci
-      docker: huggingface/transformers-pytorch-deepspeed-latest-gpu
-      ci_event: Daily CI
-      working-directory-prefix: /workspace
-    secrets: inherit
-
-  quantization-ci:
-    name: Quantization CI
-    uses: ./.github/workflows/self-scheduled.yml
-    with:
-      job: run_quantization_torch_gpu
-      slack_report_channel: "#transformers-ci-daily-quantization"
-      runner: daily-ci
-      docker: huggingface/transformers-quantization-latest-gpu
-      ci_event: Daily CI
-    secrets: inherit

.github/workflows/self-scheduled.yml

@@ -2,32 +2,14 @@ name: Self-hosted runner (scheduled)
 
 # Note that each job's dependencies go into a corresponding docker file.
 #
-# For example for `run_torch_cuda_extensions_gpu` the docker image is
+# For example for `run_all_tests_torch_cuda_extensions_gpu` the docker image is
 # `huggingface/transformers-pytorch-deepspeed-latest-gpu`, which can be found at
 # `docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile`
 
 on:
-  workflow_call:
-    inputs:
-      job:
-        required: true
-        type: string
-      slack_report_channel:
-        required: true
-        type: string
-      runner:
-        required: true
-        type: string
-      docker:
-        required: true
-        type: string
-      ci_event:
-        required: true
-        type: string
-      working-directory-prefix:
-        default: ''
-        required: false
-        type: string
+  repository_dispatch:
+  schedule:
+    - cron: "0 2 * * *"
 
 env:
   HF_HOME: /mnt/cache
@@ -35,30 +17,37 @@ env:
   OMP_NUM_THREADS: 8
   MKL_NUM_THREADS: 8
   RUN_SLOW: yes
-  # For gated repositories, we still need to agree to share information on the Hub repo. page in order to get access.
-  # This token is created under the bot `hf-transformers-bot`.
-  HF_HUB_READ_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}
   SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
   TF_FORCE_GPU_ALLOW_GROWTH: true
   RUN_PT_TF_CROSS_TESTS: 1
-  CUDA_VISIBLE_DEVICES: 0,1
-  NUM_SLICES: 2
 
 jobs:
-  setup:
-    if: contains(fromJSON('["run_models_gpu", "run_quantization_torch_gpu"]'), inputs.job)
-    name: Setup
+  run_check_runners:
+    name: Check Runners
     strategy:
       matrix:
         machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    container:
+      image: huggingface/transformers-all-latest-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+  setup:
+    name: Setup
+    needs: run_check_runners
+    strategy:
+      matrix:
+        machine_type: [single-gpu, multi-gpu]
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
     container:
       image: huggingface/transformers-all-latest-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     outputs:
-      folder_slices: ${{ steps.set-matrix.outputs.folder_slices }}
-      slice_ids: ${{ steps.set-matrix.outputs.slice_ids }}
-      quantization_matrix: ${{ steps.set-matrix-quantization.outputs.quantization_matrix }}
+      matrix: ${{ steps.set-matrix.outputs.matrix }}
     steps:
       - name: Update clone
         working-directory: /transformers
@@ -72,298 +61,37 @@ jobs:
           rm -rf tests/models/__pycache__
           rm -rf reports
 
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
       - id: set-matrix
-        if: ${{ inputs.job == 'run_models_gpu' }}
         name: Identify models to test
         working-directory: /transformers/tests
         run: |
-          echo "folder_slices=$(python3 ../utils/split_model_tests.py --num_splits ${{ env.NUM_SLICES }})" >> $GITHUB_OUTPUT
-          echo "slice_ids=$(python3 -c 'd = list(range(${{ env.NUM_SLICES }})); print(d)')" >> $GITHUB_OUTPUT
-      
-      - id: set-matrix-quantization
-        if: ${{ inputs.job == 'run_quantization_torch_gpu' }}
-        name: Identify quantization method to test
-        working-directory: /transformers/tests
-        run: |
-          echo "quantization_matrix=$(python3 -c 'import os; tests = os.getcwd(); quantization_tests = os.listdir(os.path.join(tests, "quantization")); d = sorted(list(filter(os.path.isdir, [f"quantization/{x}" for x in quantization_tests]))) ;  print(d)')" >> $GITHUB_OUTPUT
+          echo "::set-output name=matrix::$(python3 -c 'import os; tests = os.getcwd(); model_tests = os.listdir(os.path.join(tests, "models")); d1 = sorted(list(filter(os.path.isdir, os.listdir(tests)))); d2 = sorted(list(filter(os.path.isdir, [f"models/{x}" for x in model_tests]))); d1.remove("models"); d = d2 + d1; print(d)')"
 
       - name: NVIDIA-SMI
         run: |
           nvidia-smi
 
-  run_models_gpu:
-    if: ${{ inputs.job == 'run_models_gpu' }}
-    name: " "
-    needs: setup
-    strategy:
-      fail-fast: false
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-        slice_id: ${{ fromJSON(needs.setup.outputs.slice_ids) }}
-    uses: ./.github/workflows/model_jobs.yml
-    with:
-      folder_slices: ${{ needs.setup.outputs.folder_slices }}
-      machine_type: ${{ matrix.machine_type }}
-      slice_id: ${{ matrix.slice_id }}
-      runner: ${{ inputs.runner }}
-      docker: ${{ inputs.docker }}
-    secrets: inherit
-
-  run_pipelines_torch_gpu:
-    if: ${{ inputs.job == 'run_pipelines_torch_gpu' }}
-    name: PyTorch pipelines
-    strategy:
-      fail-fast: false
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
-    container:
-      image: huggingface/transformers-pytorch-gpu
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all pipeline tests on GPU
-        working-directory: /transformers
-        run: |
-          python3 -m pytest -n 1 -v --dist=loadfile --make-reports=${{ matrix.machine_type }}_run_pipelines_torch_gpu_test_reports tests/pipelines
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_run_pipelines_torch_gpu_test_reports/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_pipelines_torch_gpu_test_reports"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_pipelines_torch_gpu_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_run_pipelines_torch_gpu_test_reports
-
-  run_pipelines_tf_gpu:
-    if: ${{ inputs.job == 'run_pipelines_tf_gpu' }}
-    name: TensorFlow pipelines
-    strategy:
-      fail-fast: false
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
-    container:
-      image: huggingface/transformers-tensorflow-gpu
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: |
-          git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run all pipeline tests on GPU
-        working-directory: /transformers
-        run: |
-          python3 -m pytest -n 1 -v --dist=loadfile --make-reports=${{ matrix.machine_type }}_run_pipelines_tf_gpu_test_reports tests/pipelines
-
-      - name: Failure short reports
-        if: ${{ always() }}
-        run: |
-          cat /transformers/reports/${{ matrix.machine_type }}_run_pipelines_tf_gpu_test_reports/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_pipelines_tf_gpu_test_reports"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_pipelines_tf_gpu_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_run_pipelines_tf_gpu_test_reports
-
-  run_examples_gpu:
-    if: ${{ inputs.job == 'run_examples_gpu' }}
-    name: Examples directory
+  run_tests_single_gpu:
+    name: Model tests
     strategy:
       fail-fast: false
       matrix:
+        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
         machine_type: [single-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
     container:
       image: huggingface/transformers-all-latest-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: /transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-
-      - name: Run examples tests on GPU
-        working-directory: /transformers
-        run: |
-          pip install -r examples/pytorch/_tests_requirements.txt
-          python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_run_examples_gpu_test_reports examples/pytorch
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_run_examples_gpu_test_reports/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_examples_gpu_test_reports"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_examples_gpu_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_run_examples_gpu_test_reports
-
-  run_torch_cuda_extensions_gpu:
-    if: ${{ inputs.job == 'run_torch_cuda_extensions_gpu' }}
-    name: Torch CUDA extension tests
-    strategy:
-      fail-fast: false
-      matrix:
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
-    container:
-      image: ${{ inputs.docker }}
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-    steps:
-      - name: Update clone
-        working-directory: ${{ inputs.working-directory-prefix }}/transformers
-        run: git fetch && git checkout ${{ github.sha }}
-
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
-        working-directory: ${{ inputs.working-directory-prefix }}/transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
-
-      - name: Update / Install some packages (for Past CI)
-        if: ${{ contains(inputs.docker, '-past-') && contains(inputs.docker, '-pytorch-') }}
-        working-directory: ${{ inputs.working-directory-prefix }}/transformers
-        run: |
-          python3 -m pip install -U datasets
-          python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
-
-      - name: Remove cached torch extensions
-        run: rm -rf /github/home/.cache/torch_extensions/
-
-      # To avoid unknown test failures
-      - name: Pre build DeepSpeed *again* (for daily CI)
-        if: ${{ contains(inputs.ci_event, 'Daily CI') }}
-        working-directory: ${{ inputs.working-directory-prefix }}/
-        run: |
-          python3 -m pip uninstall -y deepspeed
-          DS_DISABLE_NINJA=1 DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 python3 -m pip install deepspeed --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
-
-      # To avoid unknown test failures
-      - name: Pre build DeepSpeed *again* (for nightly & Past CI)
-        if: ${{ contains(inputs.ci_event, 'Nightly CI') || contains(inputs.ci_event, 'Past CI') }}
-        working-directory: ${{ inputs.working-directory-prefix }}/
-        run: |
-          python3 -m pip uninstall -y deepspeed
-          rm -rf DeepSpeed
-          git clone https://github.com/microsoft/DeepSpeed && cd DeepSpeed && rm -rf build
-          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 python3 -m pip install . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
-
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-
-      - name: Environment
-        working-directory: ${{ inputs.working-directory-prefix }}/transformers
-        run: |
-          python3 utils/print_env.py
-
-      - name: Show installed libraries and their versions
-        working-directory: ${{ inputs.working-directory-prefix }}/transformers
-        run: pip freeze
-
-      - name: Run all tests on GPU
-        working-directory: ${{ inputs.working-directory-prefix }}/transformers
-        run: |
-          python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports tests/deepspeed tests/extended
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat ${{ inputs.working-directory-prefix }}/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports/failures_short.txt
-
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports"
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
-          path: ${{ inputs.working-directory-prefix }}/transformers/reports/${{ matrix.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
-
-  run_quantization_torch_gpu:
-    if: ${{ inputs.job == 'run_quantization_torch_gpu' }}
-    name: " "
-    needs: setup
-    strategy:
-      max-parallel: 4
-      fail-fast: false
-      matrix:
-        folders: ${{ fromJson(needs.setup.outputs.quantization_matrix) }}
-        machine_type: [single-gpu, multi-gpu]
-    runs-on: ['${{ matrix.machine_type }}', nvidia-gpu, t4, '${{ inputs.runner }}']
-    container:
-      image: huggingface/transformers-quantization-latest-gpu
-      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    needs: [run_check_runners, setup]
     steps:
       - name: Echo folder ${{ matrix.folders }}
         shell: bash
+        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
+        # set the artifact folder names (because the character `/` is not allowed).
         run: |
           echo "${{ matrix.folders }}"
           matrix_folders=${{ matrix.folders }}
-          matrix_folders=${matrix_folders/'quantization/'/'quantization_'}
+          matrix_folders=${matrix_folders/'models/'/'models_'}
           echo "$matrix_folders"
           echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
 
@@ -371,9 +99,58 @@ jobs:
         working-directory: /transformers
         run: git fetch && git checkout ${{ github.sha }}
 
-      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Environment
         working-directory: /transformers
-        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
+        run: |
+          python3 utils/print_env.py
+
+      - name: Run all tests on GPU
+        working-directory: /transformers
+        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
+
+      - name: Failure short reports
+        if: ${{ failure() }}
+        continue-on-error: true
+        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
+          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
+
+  run_tests_multi_gpu:
+    name: Model tests
+    strategy:
+      fail-fast: false
+      matrix:
+        folders: ${{ fromJson(needs.setup.outputs.matrix) }}
+        machine_type: [multi-gpu]
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    container:
+      image: huggingface/transformers-all-latest-gpu
+      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    needs: [run_check_runners, setup]
+    steps:
+      - name: Echo folder ${{ matrix.folders }}
+        shell: bash
+        # For folders like `models/bert`, set an env. var. (`matrix_folders`) to `models_bert`, which will be used to
+        # set the artifact folder names (because the character `/` is not allowed).
+        run: |
+          echo "${{ matrix.folders }}"
+          matrix_folders=${{ matrix.folders }}
+          matrix_folders=${matrix_folders/'models/'/'models_'}
+          echo "$matrix_folders"
+          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV
+
+      - name: Update clone
+        working-directory: /transformers
+        run: git fetch && git checkout ${{ github.sha }}
 
       - name: NVIDIA-SMI
         run: |
@@ -384,90 +161,238 @@ jobs:
         run: |
           python3 utils/print_env.py
 
-      - name: Show installed libraries and their versions
+      - name: Run all tests on GPU
         working-directory: /transformers
-        run: pip freeze
-
-      - name: Run quantization tests on GPU
-        working-directory: /transformers
-        run: |
-          python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_run_quantization_torch_gpu_${{ matrix.folders }}_test_reports tests/${{ matrix.folders }}
+        run: python3 -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }} tests/${{ matrix.folders }}
 
       - name: Failure short reports
         if: ${{ failure() }}
         continue-on-error: true
-        run: cat /transformers/reports/${{ matrix.machine_type }}_run_quantization_torch_gpu_${{ matrix.folders }}_test_reports/failures_short.txt
+        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}/failures_short.txt
 
-      - name: "Test suite reports artifacts: ${{ matrix.machine_type }}_run_quantization_torch_gpu_${{ env.matrix_folders }}_test_reports"
+      - name: Test suite reports artifacts
         if: ${{ always() }}
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v2
         with:
-          name: ${{ matrix.machine_type }}_run_quantization_torch_gpu_${{ env.matrix_folders }}_test_reports
-          path: /transformers/reports/${{ matrix.machine_type }}_run_quantization_torch_gpu_${{ matrix.folders }}_test_reports
+          name: ${{ matrix.machine_type }}_run_all_tests_gpu_${{ env.matrix_folders }}_test_reports
+          path: /transformers/reports/${{ matrix.machine_type }}_tests_gpu_${{ matrix.folders }}
 
-  run_extract_warnings:
-    # Let's only do this for the job `run_models_gpu` to simplify the (already complex) logic.
-    if: ${{ always() && inputs.job == 'run_models_gpu' }}
-    name: Extract warnings in CI artifacts
-    runs-on: ubuntu-22.04
-    needs: [setup, run_models_gpu]
+  run_examples_gpu:
+    name: Examples directory
+    runs-on: [self-hosted, single-gpu-docker]
+    container:
+      image: huggingface/transformers-all-latest-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    needs: [run_check_runners, setup]
     steps:
-      - name: Checkout transformers
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 2
+      - name: Update clone
+        working-directory: /transformers
+        run: git fetch && git checkout ${{ github.sha }}
 
-      - name: Install transformers
-        run: pip install transformers
-
-      - name: Show installed libraries and their versions
-        run: pip freeze
-
-      - name: Create output directory
-        run: mkdir warnings_in_ci
-
-      - uses: actions/download-artifact@v4
-        with:
-          path: warnings_in_ci
-
-      - name: Show artifacts
-        run: echo "$(python3 -c 'import os; d = os.listdir(); print(d)')"
-        working-directory: warnings_in_ci
-
-      - name: Extract warnings in CI artifacts
+      - name: NVIDIA-SMI
         run: |
-          python3 utils/extract_warnings.py --workflow_run_id ${{ github.run_id }} --output_dir warnings_in_ci --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }} --from_gh
-          echo "$(python3 -c 'import os; import json; fp = open("warnings_in_ci/selected_warnings.json"); d = json.load(fp); d = "\n".join(d) ;print(d)')"
+          nvidia-smi
 
-      - name: Upload artifact
+      - name: Environment
+        working-directory: /transformers
+        run: |
+          python3 utils/print_env.py
+
+      - name: Run examples tests on GPU
+        working-directory: /transformers
+        run: |
+          pip install -r examples/pytorch/_tests_requirements.txt
+          python3 -m pytest -v --make-reports=single-gpu_examples_gpu examples/pytorch
+
+      - name: Failure short reports
+        if: ${{ failure() }}
+        continue-on-error: true
+        run: cat /transformers/reports/single-gpu_examples_gpu/failures_short.txt
+
+      - name: Test suite reports artifacts
         if: ${{ always() }}
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v2
         with:
-          name: warnings_in_ci
-          path: warnings_in_ci/selected_warnings.json
+          name: single-gpu_run_examples_gpu
+          path: /transformers/reports/single-gpu_examples_gpu
+
+  run_pipelines_torch_gpu:
+    name: PyTorch pipelines
+    strategy:
+      fail-fast: false
+      matrix:
+        machine_type: [single-gpu, multi-gpu]
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    container:
+      image: huggingface/transformers-pytorch-gpu
+      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    needs: [run_check_runners, setup]
+    steps:
+      - name: Update clone
+        working-directory: /transformers
+        run: git fetch && git checkout ${{ github.sha }}
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Environment
+        working-directory: /transformers
+        run: |
+          python3 utils/print_env.py
+
+      - name: Run all pipeline tests on GPU
+        working-directory: /transformers
+        env:
+          RUN_PIPELINE_TESTS: yes
+        run: |
+          python3 -m pytest -n 1 -v --dist=loadfile -m is_pipeline_test --make-reports=${{ matrix.machine_type }}_tests_torch_pipeline_gpu tests
+
+      - name: Failure short reports
+        if: ${{ failure() }}
+        continue-on-error: true
+        run: cat /transformers/reports/${{ matrix.machine_type }}_tests_torch_pipeline_gpu/failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: ${{ matrix.machine_type }}_run_tests_torch_pipeline_gpu
+          path: /transformers/reports/${{ matrix.machine_type }}_tests_torch_pipeline_gpu
+
+  run_pipelines_tf_gpu:
+    name: TensorFlow pipelines
+    strategy:
+      fail-fast: false
+      matrix:
+        machine_type: [single-gpu, multi-gpu]
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    container:
+      image: huggingface/transformers-tensorflow-gpu
+      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    needs: [run_check_runners, setup]
+    steps:
+      - name: Update clone
+        working-directory: /transformers
+        run: |
+          git fetch && git checkout ${{ github.sha }}
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Environment
+        working-directory: /transformers
+        run: |
+          python3 utils/print_env.py
+
+      - name: Run all pipeline tests on GPU
+        working-directory: /transformers
+        env:
+          RUN_PIPELINE_TESTS: yes
+        run: |
+          python3 -m pytest -n 1 -v --dist=loadfile -m is_pipeline_test --make-reports=${{ matrix.machine_type }}_tests_tf_pipeline_gpu tests
+
+      - name: Failure short reports
+        if: ${{ always() }}
+        run: |
+          cat /transformers/reports/${{ matrix.machine_type }}_tests_tf_pipeline_gpu/failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: ${{ matrix.machine_type }}_run_tests_tf_pipeline_gpu
+          path: /transformers/reports/${{ matrix.machine_type }}_tests_tf_pipeline_gpu
+
+  run_all_tests_torch_cuda_extensions_gpu:
+    name: Torch CUDA extension tests
+    strategy:
+      fail-fast: false
+      matrix:
+        machine_type: [single-gpu, multi-gpu]
+    runs-on: ${{ format('{0}-{1}', matrix.machine_type, 'docker') }}
+    needs: [run_check_runners, setup]
+    container:
+      image: huggingface/transformers-pytorch-deepspeed-latest-gpu
+      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Update clone
+        working-directory: /workspace/transformers
+        run: git fetch && git checkout ${{ github.sha }}
+
+      - name: Remove cached torch extensions
+        run: rm -rf /github/home/.cache/torch_extensions/
+
+      # To avoid unknown test failures
+      - name: Pre build DeepSpeed *again*
+        working-directory: /workspace
+        run: |
+          python3 -m pip uninstall -y deepspeed
+          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 DS_BUILD_AIO=1 DS_BUILD_UTILS=1 python3 -m pip install deepspeed --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
+
+      - name: NVIDIA-SMI
+        run: |
+          nvidia-smi
+
+      - name: Environment
+        working-directory: /workspace/transformers
+        run: |
+          python utils/print_env.py
+
+      - name: Run all tests on GPU
+        working-directory: /workspace/transformers
+        run: |
+          python -m pytest -v --make-reports=${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu tests/deepspeed tests/extended
+
+      - name: Failure short reports
+        if: ${{ failure() }}
+        continue-on-error: true
+        run: cat /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu/failures_short.txt
+
+      - name: Test suite reports artifacts
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v2
+        with:
+          name: ${{ matrix.machine_type }}_run_tests_torch_cuda_extensions_gpu_test_reports
+          path: /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu
 
   send_results:
-    name: Slack Report
+    name: Send results to webhook
+    runs-on: ubuntu-latest
+    if: always()
     needs: [
+      run_check_runners,
       setup,
-      run_models_gpu,
-      run_pipelines_torch_gpu,
-      run_pipelines_tf_gpu,
+      run_tests_single_gpu,
+      run_tests_multi_gpu,
       run_examples_gpu,
-      run_torch_cuda_extensions_gpu,
-      run_quantization_torch_gpu,
-      run_extract_warnings
+      run_pipelines_tf_gpu,
+      run_pipelines_torch_gpu,
+      run_all_tests_torch_cuda_extensions_gpu
     ]
-    if: ${{ always() }}
-    uses: ./.github/workflows/slack-report.yml
-    with:
-      job: ${{ inputs.job }}
-      # This would be `skipped` if `setup` is skipped.
-      setup_status: ${{ needs.setup.result }}
-      slack_report_channel: ${{ inputs.slack_report_channel }}
-      # This would be an empty string if `setup` is skipped.
-      folder_slices: ${{ needs.setup.outputs.folder_slices }}
-      quantization_matrix: ${{ needs.setup.outputs.quantization_matrix }}
-      ci_event: ${{ inputs.ci_event }}
+    steps:
+      - name: Preliminary job status
+        shell: bash
+        # For the meaning of these environment variables, see the job `Setup`
+        run: |
+          echo "Runner status: ${{ needs.run_check_runners.result }}"
+          echo "Setup status: ${{ needs.setup.result }}"
 
-    secrets: inherit
+      - uses: actions/checkout@v2
+      - uses: actions/download-artifact@v2
+      - name: Send message to Slack
+        env:
+          CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
+          CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID }}
+          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
+          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
+          CI_SLACK_REPORT_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
+          CI_EVENT: scheduled
+          SETUP_STATUS: ${{ needs.setup.result }}
+          RUNNER_STATUS: ${{ needs.run_check_runners.result }}
+        # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
+        # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
+        run: |
+          pip install slack_sdk
+          python utils/notification_service.py "${{ needs.setup.outputs.matrix }}"

.github/workflows/slack-report.yml

@@ -1,101 +0,0 @@
-name: CI slack report
-
-on:
-  workflow_call:
-    inputs:
-      job:
-        required: true
-        type: string
-      slack_report_channel:
-        required: true
-        type: string
-      setup_status:
-        required: true
-        type: string
-      folder_slices:
-        required: true
-        type: string
-      quantization_matrix:
-        required: true
-        type: string
-      ci_event:
-        required: true
-        type: string
-
-env:
-  TRANSFORMERS_CI_RESULTS_UPLOAD_TOKEN: ${{ secrets.TRANSFORMERS_CI_RESULTS_UPLOAD_TOKEN }}
-
-jobs:
-  send_results:
-    name: Send results to webhook
-    runs-on: ubuntu-22.04
-    if: always()
-    steps:
-      - name: Preliminary job status
-        shell: bash
-        # For the meaning of these environment variables, see the job `Setup`
-        run: |
-          echo "Setup status: ${{ inputs.setup_status }}"
-
-      - uses: actions/checkout@v4
-      - uses: actions/download-artifact@v4
-      - name: Send message to Slack
-        if: ${{ inputs.job != 'run_quantization_torch_gpu' }}
-        env:
-          CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
-          CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID }}
-          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY }}
-          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
-          SLACK_REPORT_CHANNEL: ${{ inputs.slack_report_channel }}
-          ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-          CI_EVENT: ${{ inputs.ci_event }}
-          CI_SHA: ${{ github.sha }}
-          CI_WORKFLOW_REF: ${{ github.workflow_ref }}
-          CI_TEST_JOB: ${{ inputs.job }}
-          SETUP_STATUS: ${{ inputs.setup_status }}
-        # We pass `needs.setup.outputs.matrix` as the argument. A processing in `notification_service.py` to change
-        # `models/bert` to `models_bert` is required, as the artifact names use `_` instead of `/`.
-        # For a job that doesn't depend on (i.e. `needs`) `setup`, the value for `inputs.folder_slices` would be an
-        # empty string, and the called script still get one argument (which is the emtpy string).
-        run: |
-          sudo apt-get install -y curl
-          pip install huggingface_hub
-          pip install slack_sdk
-          pip show slack_sdk
-          python utils/notification_service.py "${{ inputs.folder_slices }}"
-
-      # Upload complete failure tables, as they might be big and only truncated versions could be sent to Slack.
-      - name: Failure table artifacts
-        uses: actions/upload-artifact@v4
-        with:
-          name: ci_results_${{ inputs.job }}
-          path: ci_results_${{ inputs.job }}
-      
-      - uses: actions/checkout@v4
-      - uses: actions/download-artifact@v4
-      - name: Send message to Slack for quantization workflow
-        if: ${{ inputs.job == 'run_quantization_torch_gpu' }}
-        env:
-          CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
-          ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
-          SLACK_REPORT_CHANNEL: ${{ inputs.slack_report_channel }}
-          CI_EVENT: ${{ inputs.ci_event }}
-          CI_SHA: ${{ github.sha }}
-          CI_TEST_JOB: ${{ inputs.job }}
-          SETUP_STATUS: ${{ inputs.setup_status }}
-        # We pass `needs.setup.outputs.quantization_matrix` as the argument. A processing in `notification_service_quantization.py` to change
-        # `quantization/bnb` to `quantization_bnb` is required, as the artifact names use `_` instead of `/`.
-        run: |
-          sudo apt-get install -y curl
-          pip install huggingface_hub
-          pip install slack_sdk
-          pip show slack_sdk
-          python utils/notification_service_quantization.py "${{ inputs.quantization_matrix }}" 
-
-      # Upload complete failure tables, as they might be big and only truncated versions could be sent to Slack.
-      - name: Failure table artifacts
-        if: ${{ inputs.job == 'run_quantization_torch_gpu' }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: ci_results_${{ inputs.job }}
-          path: ci_results_${{ inputs.job }}

.github/workflows/ssh-runner.yml

@@ -1,63 +0,0 @@
-name: SSH into our runners
-
-on:
-  workflow_dispatch:
-    inputs:
-      runner_type:
-        description: 'Type of runner to test (a10 or t4)'
-        required: true 
-      docker_image:
-        description: 'Name of the Docker image'
-        required: true
-      num_gpus:
-        description: 'Type of the number of gpus to use (`single` or `multi`)'
-        required: true
-
-env:
-  HF_HUB_READ_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}
-  HF_HOME: /mnt/cache 
-  TRANSFORMERS_IS_CI: yes 
-  OMP_NUM_THREADS: 8 
-  MKL_NUM_THREADS: 8 
-  RUN_SLOW: yes # For gated repositories, we still need to agree to share information on the Hub repo. page in order to get access. # This token is created under the bot `hf-transformers-bot`. 
-  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }} 
-  TF_FORCE_GPU_ALLOW_GROWTH: true 
-  CUDA_VISIBLE_DEVICES: 0,1
-  RUN_PT_TF_CROSS_TESTS: 1
-
-jobs:
-  ssh_runner:
-    name: "SSH"
-    runs-on: ["${{ github.event.inputs.num_gpus }}-gpu", nvidia-gpu, "${{ github.event.inputs.runner_type }}", ci]
-    container:
-      image: ${{ github.event.inputs.docker_image }}
-      options: --gpus all --privileged --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
-
-    steps:
-      - name: Update clone
-        working-directory: /transformers
-        run: |
-          git fetch && git checkout ${{ github.sha }}
-
-      - name: Cleanup
-        working-directory: /transformers
-        run: |
-          rm -rf tests/__pycache__
-          rm -rf tests/models/__pycache__
-          rm -rf reports
-
-      - name: Show installed libraries and their versions
-        working-directory: /transformers
-        run: pip freeze
-      
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-      
-      - name: Tailscale # In order to be able to SSH when a test fails
-        uses: huggingface/tailscale-action@main
-        with:
-          authkey: ${{ secrets.TAILSCALE_SSH_AUTHKEY }}
-          slackChannel: ${{ secrets.SLACK_CIFEEDBACK_CHANNEL }}
-          slackToken: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
-          waitForSSH: true

.github/workflows/stale.yml

@@ -2,26 +2,26 @@ name: Stale Bot
 
 on:
   schedule:
-    - cron: "0 8 * * *"
+    - cron: "0 15 * * *"
 
 jobs:
   close_stale_issues:
     name: Close Stale Issues
     if: github.repository == 'huggingface/transformers'
-    runs-on: ubuntu-22.04
+    runs-on: ubuntu-latest
     env:
       GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
     steps:
-    - uses: actions/checkout@v4
+    - uses: actions/checkout@v2
 
     - name: Setup Python
-      uses: actions/setup-python@v4
+      uses: actions/setup-python@v1
       with:
-        python-version: 3.8
+        python-version: 3.7
 
     - name: Install requirements
       run: |
         pip install PyGithub
     - name: Close stale issues
       run: |
-        python scripts/stale.py
+        python scripts/stale.py

.github/workflows/trufflehog.yml

@@ -1,18 +0,0 @@
-on:
-  push:
-
-name: Secret Leaks
-
-permissions:
-  contents: read
-
-jobs:
-  trufflehog:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-      - name: Secret Scanning
-        uses: trufflesecurity/trufflehog@main

.github/workflows/update_metdata.yml

@@ -4,24 +4,37 @@ on:
   push:
     branches:
       - main
-      - update_transformers_metadata*
+      - update_transformers_metadata
 
 jobs:
   build_and_package:
-    runs-on: ubuntu-22.04
+    runs-on: ubuntu-latest
     defaults:
       run:
         shell: bash -l {0}
 
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v2
+
+      - name: Load cached virtual environment
+        uses: actions/cache@v2
+        id: cache
+        with:
+          path: ~/venv/
+          key: v3-metadata-${{ hashFiles('setup.py') }}
+
+      - name: Create virtual environment on cache miss
+        if: steps.cache.outputs.cache-hit != 'true'
+        run: |
+          python -m venv ~/venv && . ~/venv/bin/activate
+          pip install --upgrade pip
 
       - name: Setup environment
         run: |
-          pip install --upgrade pip
-          pip install datasets pandas==2.0.3
-          pip install .[torch,tf,flax]
+          . ~/venv/bin/activate
+          pip install git+https://github.com/huggingface/transformers#egg=transformers[dev]
 
       - name: Update metadata
         run: |
-          python utils/update_metadata.py --token ${{ secrets.LYSANDRE_HF_TOKEN }} --commit_sha ${{ github.sha }}
+          . ~/venv/bin/activate
+          python utils/update_metadata.py --token ${{ secrets.SYLVAIN_HF_TOKEN }} --commit_sha ${{ github.sha }}

.github/workflows/upload_pr_documentation.yml

@@ -1,16 +0,0 @@
-name: Upload PR Documentation
-
-on:
-  workflow_run:
-    workflows: ["Build PR Documentation"]
-    types:
-      - completed
-
-jobs:
-  build:
-    uses: huggingface/doc-builder/.github/workflows/upload_pr_documentation.yml@main
-    with:
-      package_name: transformers
-    secrets:
-      hf_token: ${{ secrets.HF_DOC_BUILD_PUSH }}
-      comment_bot_token: ${{ secrets.COMMENT_BOT_TOKEN }}

.gitignore

@@ -163,7 +163,4 @@ tags
 *.lock
 
 # DS_Store (MacOS)
-.DS_Store
-
-# ruff
-.ruff_cache
+.DS_Store

CODE_OF_CONDUCT.md

@@ -7,8 +7,8 @@ We as members, contributors, and leaders pledge to make participation in our
 community a harassment-free experience for everyone, regardless of age, body
 size, visible or invisible disability, ethnicity, sex characteristics, gender
 identity and expression, level of experience, education, socio-economic status,
-nationality, personal appearance, race, caste, color, religion, or sexual
-identity and orientation.
+nationality, personal appearance, race, religion, or sexual identity
+and orientation.
 
 We pledge to act and interact in ways that contribute to an open, welcoming,
 diverse, inclusive, and healthy community.
@@ -23,17 +23,17 @@ community include:
 * Giving and gracefully accepting constructive feedback
 * Accepting responsibility and apologizing to those affected by our mistakes,
   and learning from the experience
-* Focusing on what is best not just for us as individuals, but for the overall
-  community
+* Focusing on what is best not just for us as individuals, but for the
+  overall community
 
 Examples of unacceptable behavior include:
 
-* The use of sexualized language or imagery, and sexual attention or advances of
-  any kind
+* The use of sexualized language or imagery, and sexual attention or
+  advances of any kind
 * Trolling, insulting or derogatory comments, and personal or political attacks
 * Public or private harassment
-* Publishing others' private information, such as a physical or email address,
-  without their explicit permission
+* Publishing others' private information, such as a physical or email
+  address, without their explicit permission
 * Other conduct which could reasonably be considered inappropriate in a
   professional setting
 
@@ -83,15 +83,15 @@ behavior was inappropriate. A public apology may be requested.
 
 ### 2. Warning
 
-**Community Impact**: A violation through a single incident or series of
-actions.
+**Community Impact**: A violation through a single incident or series
+of actions.
 
 **Consequence**: A warning with consequences for continued behavior. No
 interaction with the people involved, including unsolicited interaction with
 those enforcing the Code of Conduct, for a specified period of time. This
 includes avoiding interactions in community spaces as well as external channels
-like social media. Violating these terms may lead to a temporary or permanent
-ban.
+like social media. Violating these terms may lead to a temporary or
+permanent ban.
 
 ### 3. Temporary Ban
 
@@ -107,27 +107,23 @@ Violating these terms may lead to a permanent ban.
 ### 4. Permanent Ban
 
 **Community Impact**: Demonstrating a pattern of violation of community
-standards, including sustained inappropriate behavior, harassment of an
+standards, including sustained inappropriate behavior,  harassment of an
 individual, or aggression toward or disparagement of classes of individuals.
 
-**Consequence**: A permanent ban from any sort of public interaction within the
-community.
+**Consequence**: A permanent ban from any sort of public interaction within
+the community.
 
 ## Attribution
 
 This Code of Conduct is adapted from the [Contributor Covenant][homepage],
-version 2.1, available at
-[https://www.contributor-covenant.org/version/2/1/code_of_conduct.html][v2.1].
+version 2.0, available at
+https://www.contributor-covenant.org/version/2/0/code_of_conduct.html.
 
-Community Impact Guidelines were inspired by
-[Mozilla's code of conduct enforcement ladder][Mozilla CoC].
-
-For answers to common questions about this code of conduct, see the FAQ at
-[https://www.contributor-covenant.org/faq][FAQ]. Translations are available at
-[https://www.contributor-covenant.org/translations][translations].
+Community Impact Guidelines were inspired by [Mozilla's code of conduct
+enforcement ladder](https://github.com/mozilla/diversity).
 
 [homepage]: https://www.contributor-covenant.org
-[v2.1]: https://www.contributor-covenant.org/version/2/1/code_of_conduct.html
-[Mozilla CoC]: https://github.com/mozilla/diversity
-[FAQ]: https://www.contributor-covenant.org/faq
-[translations]: https://www.contributor-covenant.org/translations
+
+For answers to common questions about this code of conduct, see the FAQ at
+https://www.contributor-covenant.org/faq. Translations are available at
+https://www.contributor-covenant.org/translations.

CONTRIBUTING.md

@@ -14,339 +14,349 @@ See the License for the specific language governing permissions and
 limitations under the License.
 -->
 
-# Contribute to 🤗 Transformers
+# How to contribute to transformers?
 
 Everyone is welcome to contribute, and we value everybody's contribution. Code
-contributions are not the only way to help the community. Answering questions, helping
-others, and improving the documentation are also immensely valuable.
+is thus not the only way to help the community. Answering questions, helping
+others, reaching out and improving the documentations are immensely valuable to
+the community.
 
-It also helps us if you spread the word! Reference the library in blog posts
-about the awesome projects it made possible, shout out on Twitter every time it has
-helped you, or simply ⭐️ the repository to say thank you.
+It also helps us if you spread the word: reference the library from blog posts
+on the awesome projects it made possible, shout out on Twitter every time it has
+helped you, or simply star the repo to say "thank you".
 
-However you choose to contribute, please be mindful and respect our
+Whichever way you choose to contribute, please be mindful to respect our
 [code of conduct](https://github.com/huggingface/transformers/blob/main/CODE_OF_CONDUCT.md).
 
-**This guide was heavily inspired by the awesome [scikit-learn guide to contributing](https://github.com/scikit-learn/scikit-learn/blob/main/CONTRIBUTING.md).**
+## You can contribute in so many ways!
 
-## Ways to contribute
+There are 4 ways you can contribute to transformers:
+* Fixing outstanding issues with the existing code;
+* Implementing new models;
+* Contributing to the examples or to the documentation;
+* Submitting issues related to bugs or desired new features.
 
-There are several ways you can contribute to 🤗 Transformers:
-
-* Fix outstanding issues with the existing code.
-* Submit issues related to bugs or desired new features.
-* Implement new models.
-* Contribute to the examples or to the documentation.
-
-If you don't know where to start, there is a special [Good First
+In particular there is a special [Good First
 Issue](https://github.com/huggingface/transformers/contribute) listing. It will give you a list of
-open issues that are beginner-friendly and help you start contributing to open-source. The best way to do that is to open a Pull Request and link it to the issue that you'd like to work on. We try to give priority to opened PRs as we can easily track the progress of the fix, and if the contributor does not have time anymore, someone else can take the PR over.
+open Issues that are open to anybody to work on. Just comment in the issue that you'd like to work
+on it. In that same listing you will also find some Issues with `Good Second Issue` label. These are
+typically slightly more complicated than the Issues with just `Good First Issue` label. But if you
+feel you know what you're doing, go for it.
 
-For something slightly more challenging, you can also take a look at the [Good Second Issue](https://github.com/huggingface/transformers/labels/Good%20Second%20Issue) list. In general though, if you feel like you know what you're doing, go for it and we'll help you get there! 🚀
+*All are equally valuable to the community.*
 
-> All contributions are equally valuable to the community. 🥰
+## Submitting a new issue or feature request
 
-## Fixing outstanding issues
-
-If you notice an issue with the existing code and have a fix in mind, feel free to [start contributing](#create-a-pull-request) and open a Pull Request!
-
-## Submitting a bug-related issue or feature request
-
-Do your best to follow these guidelines when submitting a bug-related issue or a feature
+Do your best to follow these guidelines when submitting an issue or a feature
 request. It will make it easier for us to come back to you quickly and with good
 feedback.
 
 ### Did you find a bug?
 
-The 🤗 Transformers library is robust and reliable thanks to users who report the problems they encounter.
+The 🤗 Transformers library is robust and reliable thanks to the users who notify us of
+the problems they encounter. So thank you for reporting an issue.
 
-Before you report an issue, we would really appreciate it if you could **make sure the bug was not
-already reported** (use the search bar on GitHub under Issues). Your issue should also be related to bugs in the library itself, and not your code. If you're unsure whether the bug is in your code or the library, please ask in the [forum](https://discuss.huggingface.co/) or on our [discord](https://discord.com/invite/hugging-face-879548962464493619) first. This helps us respond quicker to fixing issues related to the library versus general questions.
+First, we would really appreciate it if you could **make sure the bug was not
+already reported** (use the search bar on Github under Issues).
 
-> [!TIP]
-> We have a [docs bot](https://huggingface.co/spaces/huggingchat/hf-docs-chat), and we highly encourage you to ask all your questions there. There is always a chance your bug can be fixed with a simple flag 👾🔫
+Did not find it? :( So we can act quickly on it, please follow these steps:
 
-Once you've confirmed the bug hasn't already been reported, please include the following information in your issue so we can quickly resolve it:
-
-* Your **OS type and version** and **Python**, **PyTorch** and
-  **TensorFlow** versions when applicable.
+* Include your **OS type and version**, the versions of **Python**, **PyTorch** and
+  **Tensorflow** when applicable;
 * A short, self-contained, code snippet that allows us to reproduce the bug in
-  less than 30s.
-* The *full* traceback if an exception is raised.
-* Attach any other additional information, like screenshots, you think may help.
+  less than 30s;
+* Provide the *full* traceback if an exception is raised.
 
-To get the OS and software versions automatically, run the following command:
+To get the OS and software versions automatically, you can run the following command:
 
 ```bash
 transformers-cli env
 ```
 
-You can also run the same command from the root of the repository:
+or from the root of the repository the following command:
 
 ```bash
 python src/transformers/commands/transformers_cli.py env
 ```
 
-### Do you want a new feature?
 
-If there is a new feature you'd like to see in 🤗 Transformers, please open an issue and describe:
+### Do you want to implement a new model?
 
-1. What is the *motivation* behind this feature? Is it related to a problem or frustration with the library? Is it a feature related to something you need for a project? Is it something you worked on and think it could benefit the community?
+Awesome! Please provide the following information:
 
-   Whatever it is, we'd love to hear about it!
-
-2. Describe your requested feature in as much detail as possible. The more you can tell us about it, the better we'll be able to help you.
-3. Provide a *code snippet* that demonstrates the features usage.
-4. If the feature is related to a paper, please include a link.
-
-If your issue is well written we're already 80% of the way there by the time you create it.
-
-We have added [templates](https://github.com/huggingface/transformers/tree/main/templates) to help you get started with your issue.
-
-## Do you want to implement a new model?
-
-New models are constantly released and if you want to implement a new model, please provide the following information:
-
-* A short description of the model and a link to the paper.
-* Link to the implementation if it is open-sourced.
+* Short description of the model and link to the paper;
+* Link to the implementation if it is open-source;
 * Link to the model weights if they are available.
 
-If you are willing to contribute the model yourself, let us know so we can help you add it to 🤗 Transformers!
+If you are willing to contribute the model yourself, let us know so we can best
+guide you.
 
-We have a technical guide for [how to add a model to 🤗 Transformers](https://huggingface.co/docs/transformers/add_new_model).
+We have added a **detailed guide and templates** to guide you in the process of adding a new model. You can find them
+in the [`templates`](https://github.com/huggingface/transformers/tree/main/templates) folder.
 
-## Do you want to add documentation?
+### Do you want a new feature (that is not a model)?
 
-We're always looking for improvements to the documentation that make it more clear and accurate. Please let us know how the documentation can be improved such as typos and any content that is missing, unclear or inaccurate. We'll be happy to make the changes or help you make a contribution if you're interested!
+A world-class feature request addresses the following points:
 
-For more details about how to generate, build, and write the documentation, take a look at the documentation [README](https://github.com/huggingface/transformers/tree/main/docs).
+1. Motivation first:
+  * Is it related to a problem/frustration with the library? If so, please explain
+    why. Providing a code snippet that demonstrates the problem is best.
+  * Is it related to something you would need for a project? We'd love to hear
+    about it!
+  * Is it something you worked on and think could benefit the community?
+    Awesome! Tell us what problem it solved for you.
+2. Write a *full paragraph* describing the feature;
+3. Provide a **code snippet** that demonstrates its future use;
+4. In case this is related to a paper, please attach a link;
+5. Attach any additional information (drawings, screenshots, etc.) you think may help.
 
-## Create a Pull Request
+If your issue is well written we're already 80% of the way there by the time you
+post it.
 
-Before writing any code, we strongly advise you to search through the existing PRs or
-issues to make sure nobody is already working on the same thing. If you are
+We have added **templates** to guide you in the process of adding a new example script for training or testing the
+models in the library. You can find them in the [`templates`](https://github.com/huggingface/transformers/tree/main/templates)
+folder.
+
+## Start contributing! (Pull Requests)
+
+Before writing code, we strongly advise you to search through the existing PRs or
+issues to make sure that nobody is already working on the same thing. If you are
 unsure, it is always a good idea to open an issue to get some feedback.
 
-You will need basic `git` proficiency to contribute to
-🤗 Transformers. While `git` is not the easiest tool to use, it has the greatest
-manual. Type `git --help` in a shell and enjoy! If you prefer books, [Pro
+You will need basic `git` proficiency to be able to contribute to
+🤗 Transformers. `git` is not the easiest tool to use but it has the greatest
+manual. Type `git --help` in a shell and enjoy. If you prefer books, [Pro
 Git](https://git-scm.com/book/en/v2) is a very good reference.
 
-You'll need **[Python 3.8](https://github.com/huggingface/transformers/blob/main/setup.py#L449)** or above to contribute to 🤗 Transformers. Follow the steps below to start contributing:
+Follow these steps to start contributing ([supported Python versions](https://github.com/huggingface/transformers/blob/main/setup.py#L426)):
 
 1. Fork the [repository](https://github.com/huggingface/transformers) by
-   clicking on the **[Fork](https://github.com/huggingface/transformers/fork)** button on the repository's page. This creates a copy of the code
+   clicking on the 'Fork' button on the repository's page. This creates a copy of the code
    under your GitHub user account.
 
 2. Clone your fork to your local disk, and add the base repository as a remote:
 
    ```bash
-   git clone git@github.com:<your Github handle>/transformers.git
-   cd transformers
-   git remote add upstream https://github.com/huggingface/transformers.git
+   $ git clone git@github.com:<your Github handle>/transformers.git
+   $ cd transformers
+   $ git remote add upstream https://github.com/huggingface/transformers.git
    ```
 
 3. Create a new branch to hold your development changes:
 
    ```bash
-   git checkout -b a-descriptive-name-for-my-changes
+   $ git checkout -b a-descriptive-name-for-my-changes
    ```
 
-   🚨 **Do not** work on the `main` branch!
+   **Do not** work on the `main` branch.
 
 4. Set up a development environment by running the following command in a virtual environment:
 
    ```bash
-   pip install -e ".[dev]"
+   $ pip install -e ".[dev]"
    ```
 
-   If 🤗 Transformers was already installed in the virtual environment, remove
+   (If transformers was already installed in the virtual environment, remove
    it with `pip uninstall transformers` before reinstalling it in editable
-   mode with the `-e` flag.
+   mode with the `-e` flag.)
 
-   Depending on your OS, and since the number of optional dependencies of Transformers is growing, you might get a
-   failure with this command. If that's the case make sure to install the Deep Learning framework you are working with
-   (PyTorch, TensorFlow and/or Flax) then do:
+   To run the full test suite, you might need the additional dependency on `datasets` which requires a separate source
+   install:
 
    ```bash
-   pip install -e ".[quality]"
+   $ git clone https://github.com/huggingface/datasets
+   $ cd datasets
+   $ pip install -e .
    ```
 
-   which should be enough for most use cases.
+   If you have already cloned that repo, you might need to `git pull` to get the most recent changes in the `datasets`
+   library.
+   
+   Depending on your OS, you might need to install some external libraries, as well, if the `pip` installation fails.
+   
+   For macOS, you will likely need [MeCab](https://taku910.github.io/mecab/), which can be installed from Homebrew:
+   
+   ```bash
+   brew install mecab
+   ```
 
-5. Develop the features in your branch.
+5. Develop the features on your branch.
 
-   As you work on your code, you should make sure the test suite
-   passes. Run the tests impacted by your changes like this:
+   As you work on the features, you should make sure that the test suite
+   passes. You should run the tests impacted by your changes like this:
 
    ```bash
-   pytest tests/<TEST_TO_RUN>.py
+   $ pytest tests/<TEST_TO_RUN>.py
+   ```
+
+   You can also run the full suite with the following command, but it takes
+   a beefy machine to produce a result in a decent amount of time now that
+   Transformers has grown a lot. Here is the command for it:
+
+   ```bash
+   $ make test
    ```
 
    For more information about tests, check out the
-   [Testing](https://huggingface.co/docs/transformers/testing) guide.
+   [dedicated documentation](https://huggingface.co/docs/transformers/testing)
 
-   🤗 Transformers relies on `black` and `ruff` to format its source code
+   🤗 Transformers relies on `black` and `isort` to format its source code
    consistently. After you make changes, apply automatic style corrections and code verifications
    that can't be automated in one go with:
 
    ```bash
-   make fixup
+   $ make fixup
    ```
 
    This target is also optimized to only work with files modified by the PR you're working on.
 
-   If you prefer to run the checks one after the other, the following command applies the
+   If you prefer to run the checks one after the other, the following command apply the
    style corrections:
 
    ```bash
-   make style
+   $ make style
    ```
 
-   🤗 Transformers also uses `ruff` and a few custom scripts to check for coding mistakes. Quality
-   controls are run by the CI, but you can run the same checks with:
+   🤗 Transformers also uses `flake8` and a few custom scripts to check for coding mistakes. Quality
+   control runs in CI, however you can also run the same checks with:
 
    ```bash
-   make quality
+   $ make quality
    ```
 
-   Finally, we have a lot of scripts to make sure we don't forget to update
-   some files when adding a new model. You can run these scripts with:
+   Finally we have a lot of scripts that check we didn't forget to update
+   some files when adding a new model, that you can run with
 
    ```bash
-   make repo-consistency
+   $ make repo-consistency
    ```
 
-   To learn more about those checks and how to fix any issues with them, check out the
-   [Checks on a Pull Request](https://huggingface.co/docs/transformers/pr_checks) guide.
-
-   If you're modifying documents under the `docs/source` directory, make sure the documentation can still be built. This check will also run in the CI when you open a pull request. To run a local check
-   make sure you install the documentation builder:
+   To learn more about those checks and how to fix any issue with them, check out the
+   [documentation](https://huggingface.co/docs/transformers/pr_checks)
 
+   If you're modifying documents under `docs/source`, make sure to validate that
+   they can still be built. This check also runs in CI. To run a local check
+   make sure you have installed the documentation builder requirements. First you will need to clone the
+   repository containing our tools to build the documentation:
+   
    ```bash
-   pip install ".[docs]"
+   $ pip install git+https://github.com/huggingface/doc-builder
    ```
 
-   Run the following command from the root of the repository:
+   Then, make sure you have all the dependencies to be able to build the doc with:
+   
+   ```bash
+   $ pip install ".[docs]"
+   ```
+
+   Finally run the following command from the root of the repository:
 
    ```bash
-   doc-builder build transformers docs/source/en --build_dir ~/tmp/test-build
+   $ doc-builder build transformers docs/source/ --build_dir ~/tmp/test-build
    ```
 
    This will build the documentation in the `~/tmp/test-build` folder where you can inspect the generated
-   Markdown files with your favorite editor. You can also preview the docs on GitHub when you open a pull request.
+   Markdown files with your favorite editor. You won't be able to see the final rendering on the website
+   before your PR is merged, we are actively working on adding a tool for this.
 
-   Once you're happy with your changes, add the changed files with `git add` and
-   record your changes locally with `git commit`:
+   Once you're happy with your changes, add changed files using `git add` and
+   make a commit with `git commit` to record your changes locally:
 
    ```bash
-   git add modified_file.py
-   git commit
+   $ git add modified_file.py
+   $ git commit
    ```
 
-   Please remember to write [good commit
-   messages](https://chris.beams.io/posts/git-commit/) to clearly communicate the changes you made!
+   Please write [good commit
+   messages](https://chris.beams.io/posts/git-commit/).
 
-   To keep your copy of the code up to date with the original
-   repository, rebase your branch on `upstream/branch` *before* you open a pull request or if requested by a maintainer:
+   It is a good idea to sync your copy of the code with the original
+   repository regularly. This way you can quickly account for changes:
 
    ```bash
-   git fetch upstream
-   git rebase upstream/main
+   $ git fetch upstream
+   $ git rebase upstream/main
    ```
 
-   Push your changes to your branch:
+   Push the changes to your account using:
 
    ```bash
-   git push -u origin a-descriptive-name-for-my-changes
+   $ git push -u origin a-descriptive-name-for-my-changes
    ```
 
-   If you've already opened a pull request, you'll need to force push with the `--force` flag. Otherwise, if the pull request hasn't been opened yet, you can just push your changes normally.
+6. Once you are satisfied (**and the checklist below is happy too**), go to the
+   webpage of your fork on GitHub. Click on 'Pull request' to send your changes
+   to the project maintainers for review.
 
-6. Now you can go to your fork of the repository on GitHub and click on **Pull Request** to open a pull request. Make sure you tick off all the boxes on our [checklist](#pull-request-checklist) below. When you're ready, you can send your changes to the project maintainers for review.
-
-7. It's ok if maintainers request changes, it happens to our core contributors
-   too! So everyone can see the changes in the pull request, work in your local
+7. It's ok if maintainers ask you for changes. It happens to core contributors
+   too! So everyone can see the changes in the Pull request, work in your local
    branch and push the changes to your fork. They will automatically appear in
    the pull request.
 
-### Pull request checklist
 
-☐ The pull request title should summarize your contribution.<br>
-☐ If your pull request addresses an issue, please mention the issue number in the pull
-request description to make sure they are linked (and people viewing the issue know you
-are working on it).<br>
-☐ To indicate a work in progress please prefix the title with `[WIP]`. These are
-useful to avoid duplicated work, and to differentiate it from PRs ready to be merged.<br>
-☐ Make sure existing tests pass.<br>
-☐ If adding a new feature, also add tests for it.<br>
-   - If you are adding a new model, make sure you use
-     `ModelTester.all_model_classes = (MyModel, MyModelWithLMHead,...)` to trigger the common tests.
+### Checklist
+
+1. The title of your pull request should be a summary of its contribution;
+2. If your pull request addresses an issue, please mention the issue number in
+   the pull request description to make sure they are linked (and people
+   consulting the issue know you are working on it);
+3. To indicate a work in progress please prefix the title with `[WIP]`. These
+   are useful to avoid duplicated work, and to differentiate it from PRs ready
+   to be merged;
+4. Make sure existing tests pass;
+5. Add high-coverage tests. No quality testing = no merge.
+   - If you are adding a new model, make sure that you use
+     `ModelTester.all_model_classes = (MyModel, MyModelWithLMHead,...)`, which triggers the common tests.
    - If you are adding new `@slow` tests, make sure they pass using
-     `RUN_SLOW=1 python -m pytest tests/models/my_new_model/test_my_new_model.py`.
-   - If you are adding a new tokenizer, write tests and make sure
-     `RUN_SLOW=1 python -m pytest tests/models/{your_model_name}/test_tokenization_{your_model_name}.py` passes.
-   - CircleCI does not run the slow tests, but GitHub Actions does every night!<br>
+     `RUN_SLOW=1 python -m pytest tests/test_my_new_model.py`.
+   - If you are adding a new tokenizer, write tests, and make sure
+     `RUN_SLOW=1 python -m pytest tests/test_tokenization_{your_model_name}.py` passes.
+   CircleCI does not run the slow tests, but github actions does every night!
+6. All public methods must have informative docstrings that work nicely with sphinx. See `modeling_bert.py` for an
+   example.
+7. Due to the rapidly growing repository, it is important to make sure that no files that would significantly weigh down the repository are added. This includes images, videos and other non-text files. We prefer to leverage a hf.co hosted `dataset` like
+   the ones hosted on [`hf-internal-testing`](https://huggingface.co/hf-internal-testing) in which to place these files and reference 
+   them by URL. We recommend putting them in the following dataset: [huggingface/documentation-images](https://huggingface.co/datasets/huggingface/documentation-images).
+   If an external contribution, feel free to add the images to your PR and ask a Hugging Face member to migrate your images
+   to this dataset.
 
-☐ All public methods must have informative docstrings (see
-[`modeling_bert.py`](https://github.com/huggingface/transformers/blob/main/src/transformers/models/bert/modeling_bert.py)
-for an example).<br>
-☐ Due to the rapidly growing repository, don't add any images, videos and other
-non-text files that'll significantly weigh down the repository. Instead, use a Hub
-repository such as [`hf-internal-testing`](https://huggingface.co/hf-internal-testing)
-to host these files and reference them by URL. We recommend placing documentation
-related images in the following repository:
-[huggingface/documentation-images](https://huggingface.co/datasets/huggingface/documentation-images).
-You can open a PR on this dataset repository and ask a Hugging Face member to merge it.
-
-For more information about the checks run on a pull request, take a look at our [Checks on a Pull Request](https://huggingface.co/docs/transformers/pr_checks) guide.
+See more about the checks run on a pull request in our [PR guide](pr_checks)
 
 ### Tests
 
 An extensive test suite is included to test the library behavior and several examples. Library tests can be found in
-the [tests](https://github.com/huggingface/transformers/tree/main/tests) folder and examples tests in the
-[examples](https://github.com/huggingface/transformers/tree/main/examples) folder.
+the [tests folder](https://github.com/huggingface/transformers/tree/main/tests) and examples tests in the
+[examples folder](https://github.com/huggingface/transformers/tree/main/examples).
 
 We like `pytest` and `pytest-xdist` because it's faster. From the root of the
-repository, specify a *path to a subfolder or a test file* to run the test:
+repository, here's how to run tests with `pytest` for the library:
 
 ```bash
-python -m pytest -n auto --dist=loadfile -s -v ./tests/models/my_new_model
+$ python -m pytest -n auto --dist=loadfile -s -v ./tests/
 ```
 
-Similarly, for the `examples` directory, specify a *path to a subfolder or test file* to run the test. For example, the following command tests the text classification subfolder in the PyTorch `examples` directory:
+and for the examples:
 
 ```bash
-pip install -r examples/xxx/requirements.txt  # only needed the first time
-python -m pytest -n auto --dist=loadfile -s -v ./examples/pytorch/text-classification
+$ pip install -r examples/xxx/requirements.txt  # only needed the first time
+$ python -m pytest -n auto --dist=loadfile -s -v ./examples/
 ```
+In fact, that's how `make test` and `make test-examples` are implemented (sans the `pip install` line)!
 
-In fact, this is actually how our `make test` and `make test-examples` commands are implemented (not including the `pip install`)!
-
-You can also specify a smaller set of tests in order to test only the feature
+You can specify a smaller set of tests in order to test only the feature
 you're working on.
 
-By default, slow tests are skipped but you can set the `RUN_SLOW` environment variable to
-`yes` to run them. This will download many gigabytes of models so make sure you
-have enough disk space, a good internet connection or a lot of patience!
-
-<Tip warning={true}>
-
-Remember to specify a *path to a subfolder or a test file* to run the test. Otherwise, you'll run all the tests in the `tests` or `examples` folder, which will take a very long time!
-
-</Tip>
+By default, slow tests are skipped. Set the `RUN_SLOW` environment variable to
+`yes` to run them. This will download many gigabytes of models — make sure you
+have enough disk space and a good Internet connection, or a lot of patience!
 
 ```bash
-RUN_SLOW=yes python -m pytest -n auto --dist=loadfile -s -v ./tests/models/my_new_model
-RUN_SLOW=yes python -m pytest -n auto --dist=loadfile -s -v ./examples/pytorch/text-classification
+$ RUN_SLOW=yes python -m pytest -n auto --dist=loadfile -s -v ./tests/
+$ RUN_SLOW=yes python -m pytest -n auto --dist=loadfile -s -v ./examples/
 ```
 
-Like the slow tests, there are other environment variables available which are not enabled by default during testing:
-- `RUN_CUSTOM_TOKENIZERS`: Enables tests for custom tokenizers.
-- `RUN_PT_FLAX_CROSS_TESTS`: Enables tests for PyTorch + Flax integration.
-- `RUN_PT_TF_CROSS_TESTS`: Enables tests for TensorFlow + PyTorch integration.
-
-More environment variables and additional information can be found in the [testing_utils.py](https://github.com/huggingface/transformers/blob/main/src/transformers/testing_utils.py).
+Likewise, set the `RUN_CUSTOM_TOKENIZERS` environment variable to `yes` to run
+tests for custom tokenizers, which don't run by default either.
 
 🤗 Transformers uses `pytest` as a test runner only. It doesn't use any
 `pytest`-specific features in the test suite itself.
@@ -355,43 +365,43 @@ This means `unittest` is fully supported. Here's how to run tests with
 `unittest`:
 
 ```bash
-python -m unittest discover -s tests -t . -v
-python -m unittest discover -s examples -t examples -v
+$ python -m unittest discover -s tests -t . -v
+$ python -m unittest discover -s examples -t examples -v
 ```
 
+
 ### Style guide
 
-For documentation strings, 🤗 Transformers follows the [Google Python Style Guide](https://google.github.io/styleguide/pyguide.html).
+For documentation strings, 🤗 Transformers follows the [google style](https://google.github.io/styleguide/pyguide.html).
 Check our [documentation writing guide](https://github.com/huggingface/transformers/tree/main/docs#writing-documentation---specification)
 for more information.
 
+**This guide was heavily inspired by the awesome [scikit-learn guide to contributing](https://github.com/scikit-learn/scikit-learn/blob/main/CONTRIBUTING.md).**
+
 ### Develop on Windows
 
-On Windows (unless you're working in [Windows Subsystem for Linux](https://learn.microsoft.com/en-us/windows/wsl/) or WSL), you need to configure git to transform Windows `CRLF` line endings to Linux `LF` line endings:
+On windows, you need to configure git to transform Windows `CRLF` line endings to Linux `LF` line endings:
 
-```bash
-git config core.autocrlf input
-```
+`git config core.autocrlf input`
 
-One way to run the `make` command on Windows is with MSYS2:
+One way one can run the make command on Window is to pass by MSYS2:
 
-1. [Download MSYS2](https://www.msys2.org/), and we assume it's installed in `C:\msys64`.
-2. Open the command line `C:\msys64\msys2.exe` (it should be available from the **Start** menu).
-3. Run in the shell: `pacman -Syu` and install `make` with `pacman -S make`.
+1. [Download MSYS2](https://www.msys2.org/), we assume to have it installed in C:\msys64
+2. Open the command line C:\msys64\msys2.exe (it should be available from the start menu)
+3. Run in the shell: `pacman -Syu` and install make with `pacman -S make`
 4. Add `C:\msys64\usr\bin` to your PATH environment variable.
 
-You can now use `make` from any terminal (PowerShell, cmd.exe, etc.)! 🎉
+You can now use `make` from any terminal (Powershell, cmd.exe, etc) 🎉
 
-### Sync a forked repository with upstream main (the Hugging Face repository)
+### Syncing forked main with upstream (HuggingFace) main
 
-When updating the main branch of a forked repository, please follow these steps to avoid pinging the upstream repository which adds reference notes to each upstream PR, and sends unnecessary notifications to the developers involved in these PRs.
-
-1. When possible, avoid syncing with the upstream using a branch and PR on the forked repository. Instead, merge directly into the forked main.
+To avoid pinging the upstream repository which adds reference notes to each upstream PR and sends unnecessary notifications to the developers involved in these PRs,
+when syncing the main branch of a forked repository, please, follow these steps:
+1. When possible, avoid syncing with the upstream using a branch and PR on the forked repository. Instead merge directly into the forked main.
 2. If a PR is absolutely necessary, use the following steps after checking out your branch:
-
-   ```bash
-   git checkout -b your-branch-for-syncing
-   git pull --squash --no-commit upstream main
-   git commit -m '<your message without GitHub references>'
-   git push --set-upstream origin your-branch-for-syncing
-   ```
+```
+$ git checkout -b your-branch-for-syncing
+$ git pull --squash --no-commit upstream main
+$ git commit -m '<your message without GitHub references>'
+$ git push --set-upstream origin your-branch-for-syncing
+```

ISSUES.md

@@ -18,7 +18,7 @@ limitations under the License.
 
 This is an Open Source Project so please be mindful that like in any other project of this kind there is no obligation to answer all requests for help.
 
-However, we want to encourage you to ask for help whenever you think it's needed! We are happy about every question we get because it allows us to better understand your needs, possible misunderstandings, and most importantly a way for you to help us make this library better. That being said, this document's main purpose is to provide guidelines at how you can formulate your requests to increase your chances to be understood and to get support.
+However, we want to encourage you to ask for help whenever you think it's needed! We are happy about every  question we get because it allows us to better understand your needs, possible misunderstandings, and most importantly a way for you to help us make this library better. That being said, this document's main purpose is to provide guidelines at how you can formulate your requests to increase your chances to be understood and to get support.
 
 There are two main venues to receive support: [the forums](https://discuss.huggingface.co/) and [the GitHub issues](https://github.com/huggingface/transformers/issues).
 
@@ -152,13 +152,13 @@ You are not required to read the following guidelines before opening an issue. H
 
    ```bash
     cd examples/seq2seq
-    torchrun --nproc_per_node=2 ./finetune_trainer.py \
+    python -m torch.distributed.launch --nproc_per_node=2 ./finetune_trainer.py \
     --model_name_or_path sshleifer/distill-mbart-en-ro-12-4 --data_dir wmt_en_ro \
     --output_dir output_dir --overwrite_output_dir \
     --do_train --n_train 500 --num_train_epochs 1 \
     --per_device_train_batch_size 1  --freeze_embeds \
     --src_lang en_XX --tgt_lang ro_RO --task translation \
-    --fp16
+    --fp16 --sharded_ddp
    ```
 
    If you don't break it up, one has to scroll horizontally which often makes it quite difficult to quickly see what's happening.

MANIFEST.in

@@ -0,0 +1 @@
+include LICENSE

Makefile

@@ -1,18 +1,17 @@
-.PHONY: deps_table_update modified_only_fixup extra_style_checks quality style fixup fix-copies test test-examples benchmark
+.PHONY: deps_table_update modified_only_fixup extra_style_checks quality style fixup fix-copies test test-examples
 
 # make sure to test the local checkout in scripts and not the pre-installed one (don't use quotes!)
 export PYTHONPATH = src
 
 check_dirs := examples tests src utils
 
-exclude_folders :=  ""
-
 modified_only_fixup:
 	$(eval modified_py_files := $(shell python utils/get_modified_files.py $(check_dirs)))
 	@if test -n "$(modified_py_files)"; then \
 		echo "Checking/fixing $(modified_py_files)"; \
-		ruff check $(modified_py_files) --fix --exclude $(exclude_folders); \
-		ruff format $(modified_py_files) --exclude $(exclude_folders);\
+		black --preview $(modified_py_files); \
+		isort $(modified_py_files); \
+		flake8 $(modified_py_files); \
 	else \
 		echo "No library .py files were modified"; \
 	fi
@@ -41,36 +40,32 @@ repo-consistency:
 	python utils/check_repo.py
 	python utils/check_inits.py
 	python utils/check_config_docstrings.py
-	python utils/check_config_attributes.py
-	python utils/check_doctest_list.py
-	python utils/update_metadata.py --check-only
-	python utils/check_docstrings.py
-	python utils/check_support_list.py
+	python utils/tests_fetcher.py --sanity_check
 
 # this target runs checks on all files
 
 quality:
-	@python -c "from transformers import *" || (echo '🚨 import failed, this means you introduced unprotected imports! 🚨'; exit 1)
-	ruff check $(check_dirs) setup.py conftest.py
-	ruff format --check $(check_dirs) setup.py conftest.py
+	black --check --preview $(check_dirs)
+	isort --check-only $(check_dirs)
 	python utils/custom_init_isort.py --check_only
 	python utils/sort_auto_mappings.py --check_only
+	flake8 $(check_dirs)
+	doc-builder style src/transformers docs/source --max_len 119 --check_only --path_to_docs docs/source
 	python utils/check_doc_toc.py
-	python utils/check_docstrings.py --check_all
-
 
 # Format source code automatically and check is there are any problems left that need manual fixing
 
 extra_style_checks:
 	python utils/custom_init_isort.py
 	python utils/sort_auto_mappings.py
+	doc-builder style src/transformers docs/source --max_len 119 --path_to_docs docs/source
 	python utils/check_doc_toc.py --fix_and_overwrite
 
 # this target runs checks on all files and potentially modifies some of them
 
 style:
-	ruff check $(check_dirs) setup.py conftest.py --fix --exclude $(exclude_folders)
-	ruff format $(check_dirs) setup.py conftest.py --exclude $(exclude_folders)
+	black --preview $(check_dirs)
+	isort $(check_dirs)
 	${MAKE} autogenerate_code
 	${MAKE} extra_style_checks
 
@@ -84,8 +79,6 @@ fix-copies:
 	python utils/check_copies.py --fix_and_overwrite
 	python utils/check_table.py --fix_and_overwrite
 	python utils/check_dummies.py --fix_and_overwrite
-	python utils/check_doctest_list.py --fix_and_overwrite
-	python utils/check_docstrings.py --fix_and_overwrite
 
 # Run tests for the library
 
@@ -97,11 +90,6 @@ test:
 test-examples:
 	python -m pytest -n auto --dist=loadfile -s -v ./examples/pytorch/
 
-# Run benchmark
-
-benchmark:
-	python3 benchmark/benchmark.py --config-dir benchmark/config --config-name generation --commit=diff backend.model=google/gemma-2b backend.cache_implementation=null,static backend.torch_compile=false,true --multirun
-
 # Run tests for SageMaker DLC release
 
 test-sagemaker: # install sagemaker dependencies in advance with pip install .[sagemaker]
@@ -121,10 +109,3 @@ post-release:
 
 post-patch:
 	python utils/release.py --post_release --patch
-
-build-release:
-	rm -rf dist
-	rm -rf build
-	python setup.py bdist_wheel
-	python setup.py sdist
-	python utils/check_build.py

README.md

@@ -15,40 +15,36 @@ limitations under the License.
 -->
 
 <p align="center">
-  <picture>
-    <source media="(prefers-color-scheme: dark)" srcset="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/transformers-logo-dark.svg">
-    <source media="(prefers-color-scheme: light)" srcset="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/transformers-logo-light.svg">
-    <img alt="Hugging Face Transformers Library" src="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/transformers-logo-light.svg" width="352" height="59" style="max-width: 100%;">
-  </picture>
-  <br/>
-  <br/>
-</p>
-
+    <br>
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_logo_name.png" width="400"/>
+    <br>
+<p>
 <p align="center">
-    <a href="https://circleci.com/gh/huggingface/transformers"><img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main"></a>
-    <a href="https://github.com/huggingface/transformers/blob/main/LICENSE"><img alt="GitHub" src="https://img.shields.io/github/license/huggingface/transformers.svg?color=blue"></a>
-    <a href="https://huggingface.co/docs/transformers/index"><img alt="Documentation" src="https://img.shields.io/website/http/huggingface.co/docs/transformers/index.svg?down_color=red&down_message=offline&up_message=online"></a>
-    <a href="https://github.com/huggingface/transformers/releases"><img alt="GitHub release" src="https://img.shields.io/github/release/huggingface/transformers.svg"></a>
-    <a href="https://github.com/huggingface/transformers/blob/main/CODE_OF_CONDUCT.md"><img alt="Contributor Covenant" src="https://img.shields.io/badge/Contributor%20Covenant-v2.0%20adopted-ff69b4.svg"></a>
+    <a href="https://circleci.com/gh/huggingface/transformers">
+        <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
+    </a>
+    <a href="https://github.com/huggingface/transformers/blob/main/LICENSE">
+        <img alt="GitHub" src="https://img.shields.io/github/license/huggingface/transformers.svg?color=blue">
+    </a>
+    <a href="https://huggingface.co/docs/transformers/index">
+        <img alt="Documentation" src="https://img.shields.io/website/http/huggingface.co/docs/transformers/index.svg?down_color=red&down_message=offline&up_message=online">
+    </a>
+    <a href="https://github.com/huggingface/transformers/releases">
+        <img alt="GitHub release" src="https://img.shields.io/github/release/huggingface/transformers.svg">
+    </a>
+    <a href="https://github.com/huggingface/transformers/blob/main/CODE_OF_CONDUCT.md">
+        <img alt="Contributor Covenant" src="https://img.shields.io/badge/Contributor%20Covenant-v2.0%20adopted-ff69b4.svg">
+    </a>
     <a href="https://zenodo.org/badge/latestdoi/155220641"><img src="https://zenodo.org/badge/155220641.svg" alt="DOI"></a>
 </p>
 
 <h4 align="center">
     <p>
         <b>English</b> |
-        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_zh-hans.md">简体中文</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_zh-hant.md">繁體中文</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_ko.md">한국어</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_es.md">Español</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_ja.md">日本語</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_hd.md">हिन्दी</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_ru.md">Русский</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_pt-br.md">Рortuguês</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_te.md">తెలుగు</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_fr.md">Français</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_de.md">Deutsch</a> |
-        <a href="https://github.com/huggingface/transformers/blob/main/i18n/README_vi.md">Tiếng Việt</a> |
-    </p>
+        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hans.md">简体中文</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hant.md">繁體中文</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_ko.md">한국어</a>
+    <p>
 </h4>
 
 <h3 align="center">
@@ -59,13 +55,13 @@ limitations under the License.
     <a href="https://hf.co/course"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/course_banner.png"></a>
 </h3>
 
-🤗 Transformers provides thousands of pretrained models to perform tasks on different modalities such as text, vision, and audio.
+🤗 Transformers provides thousands of pretrained models to perform tasks on different modalities such as text, vision, and audio. 
 
 These models can be applied on:
 
-* 📝 Text, for tasks like text classification, information extraction, question answering, summarization, translation, and text generation, in over 100 languages.
-* 🖼️ Images, for tasks like image classification, object detection, and segmentation.
-* 🗣️ Audio, for tasks like speech recognition and audio classification.
+* 📝 Text, for tasks like text classification, information extraction, question answering, summarization, translation, text generation, in over 100 languages. 
+* 🖼️ Images, for tasks like image classification, object detection, and segmentation. 
+* 🗣️ Audio, for tasks like speech recognition and audio classification. 
 
 Transformer models can also perform tasks on **several modalities combined**, such as table question answering, optical character recognition, information extraction from scanned documents, video classification, and visual question answering.
 
@@ -79,52 +75,29 @@ You can test most of our models directly on their pages from the [model hub](htt
 
 Here are a few examples:
 
-In Natural Language Processing:
-- [Masked word completion with BERT](https://huggingface.co/google-bert/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
-- [Named Entity Recognition with Electra](https://huggingface.co/dbmdz/electra-large-discriminator-finetuned-conll03-english?text=My+name+is+Sarah+and+I+live+in+London+city)
-- [Text generation with Mistral](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2)
-- [Natural Language Inference with RoBERTa](https://huggingface.co/FacebookAI/roberta-large-mnli?text=The+dog+was+lost.+Nobody+lost+any+animal)
+ In Natural Language Processing:
+- [Masked word completion with BERT](https://huggingface.co/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
+- [Name Entity Recognition with Electra](https://huggingface.co/dbmdz/electra-large-discriminator-finetuned-conll03-english?text=My+name+is+Sarah+and+I+live+in+London+city)
+- [Text generation with GPT-2](https://huggingface.co/gpt2?text=A+long+time+ago%2C+)
+- [Natural Language Inference with RoBERTa](https://huggingface.co/roberta-large-mnli?text=The+dog+was+lost.+Nobody+lost+any+animal)
 - [Summarization with BART](https://huggingface.co/facebook/bart-large-cnn?text=The+tower+is+324+metres+%281%2C063+ft%29+tall%2C+about+the+same+height+as+an+81-storey+building%2C+and+the+tallest+structure+in+Paris.+Its+base+is+square%2C+measuring+125+metres+%28410+ft%29+on+each+side.+During+its+construction%2C+the+Eiffel+Tower+surpassed+the+Washington+Monument+to+become+the+tallest+man-made+structure+in+the+world%2C+a+title+it+held+for+41+years+until+the+Chrysler+Building+in+New+York+City+was+finished+in+1930.+It+was+the+first+structure+to+reach+a+height+of+300+metres.+Due+to+the+addition+of+a+broadcasting+aerial+at+the+top+of+the+tower+in+1957%2C+it+is+now+taller+than+the+Chrysler+Building+by+5.2+metres+%2817+ft%29.+Excluding+transmitters%2C+the+Eiffel+Tower+is+the+second+tallest+free-standing+structure+in+France+after+the+Millau+Viaduct)
-- [Question answering with DistilBERT](https://huggingface.co/distilbert/distilbert-base-uncased-distilled-squad?text=Which+name+is+also+used+to+describe+the+Amazon+rainforest+in+English%3F&context=The+Amazon+rainforest+%28Portuguese%3A+Floresta+Amaz%C3%B4nica+or+Amaz%C3%B4nia%3B+Spanish%3A+Selva+Amaz%C3%B3nica%2C+Amazon%C3%ADa+or+usually+Amazonia%3B+French%3A+For%C3%AAt+amazonienne%3B+Dutch%3A+Amazoneregenwoud%29%2C+also+known+in+English+as+Amazonia+or+the+Amazon+Jungle%2C+is+a+moist+broadleaf+forest+that+covers+most+of+the+Amazon+basin+of+South+America.+This+basin+encompasses+7%2C000%2C000+square+kilometres+%282%2C700%2C000+sq+mi%29%2C+of+which+5%2C500%2C000+square+kilometres+%282%2C100%2C000+sq+mi%29+are+covered+by+the+rainforest.+This+region+includes+territory+belonging+to+nine+nations.+The+majority+of+the+forest+is+contained+within+Brazil%2C+with+60%25+of+the+rainforest%2C+followed+by+Peru+with+13%25%2C+Colombia+with+10%25%2C+and+with+minor+amounts+in+Venezuela%2C+Ecuador%2C+Bolivia%2C+Guyana%2C+Suriname+and+French+Guiana.+States+or+departments+in+four+nations+contain+%22Amazonas%22+in+their+names.+The+Amazon+represents+over+half+of+the+planet%27s+remaining+rainforests%2C+and+comprises+the+largest+and+most+biodiverse+tract+of+tropical+rainforest+in+the+world%2C+with+an+estimated+390+billion+individual+trees+divided+into+16%2C000+species)
-- [Translation with T5](https://huggingface.co/google-t5/t5-base?text=My+name+is+Wolfgang+and+I+live+in+Berlin)
+- [Question answering with DistilBERT](https://huggingface.co/distilbert-base-uncased-distilled-squad?text=Which+name+is+also+used+to+describe+the+Amazon+rainforest+in+English%3F&context=The+Amazon+rainforest+%28Portuguese%3A+Floresta+Amaz%C3%B4nica+or+Amaz%C3%B4nia%3B+Spanish%3A+Selva+Amaz%C3%B3nica%2C+Amazon%C3%ADa+or+usually+Amazonia%3B+French%3A+For%C3%AAt+amazonienne%3B+Dutch%3A+Amazoneregenwoud%29%2C+also+known+in+English+as+Amazonia+or+the+Amazon+Jungle%2C+is+a+moist+broadleaf+forest+that+covers+most+of+the+Amazon+basin+of+South+America.+This+basin+encompasses+7%2C000%2C000+square+kilometres+%282%2C700%2C000+sq+mi%29%2C+of+which+5%2C500%2C000+square+kilometres+%282%2C100%2C000+sq+mi%29+are+covered+by+the+rainforest.+This+region+includes+territory+belonging+to+nine+nations.+The+majority+of+the+forest+is+contained+within+Brazil%2C+with+60%25+of+the+rainforest%2C+followed+by+Peru+with+13%25%2C+Colombia+with+10%25%2C+and+with+minor+amounts+in+Venezuela%2C+Ecuador%2C+Bolivia%2C+Guyana%2C+Suriname+and+French+Guiana.+States+or+departments+in+four+nations+contain+%22Amazonas%22+in+their+names.+The+Amazon+represents+over+half+of+the+planet%27s+remaining+rainforests%2C+and+comprises+the+largest+and+most+biodiverse+tract+of+tropical+rainforest+in+the+world%2C+with+an+estimated+390+billion+individual+trees+divided+into+16%2C000+species)
+- [Translation with T5](https://huggingface.co/t5-base?text=My+name+is+Wolfgang+and+I+live+in+Berlin)
 
 In Computer Vision:
 - [Image classification with ViT](https://huggingface.co/google/vit-base-patch16-224)
 - [Object Detection with DETR](https://huggingface.co/facebook/detr-resnet-50)
 - [Semantic Segmentation with SegFormer](https://huggingface.co/nvidia/segformer-b0-finetuned-ade-512-512)
-- [Panoptic Segmentation with Mask2Former](https://huggingface.co/facebook/mask2former-swin-large-coco-panoptic)
-- [Depth Estimation with Depth Anything](https://huggingface.co/docs/transformers/main/model_doc/depth_anything)
-- [Video Classification with VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)
-- [Universal Segmentation with OneFormer](https://huggingface.co/shi-labs/oneformer_ade20k_dinat_large)
+- [Panoptic Segmentation with DETR](https://huggingface.co/facebook/detr-resnet-50-panoptic)
 
 In Audio:
-- [Automatic Speech Recognition with Whisper](https://huggingface.co/openai/whisper-large-v3)
+- [Automatic Speech Recognition with Wav2Vec2](https://huggingface.co/facebook/wav2vec2-base-960h)
 - [Keyword Spotting with Wav2Vec2](https://huggingface.co/superb/wav2vec2-base-superb-ks)
-- [Audio Classification with Audio Spectrogram Transformer](https://huggingface.co/MIT/ast-finetuned-audioset-10-10-0.4593)
 
 In Multimodal tasks:
-- [Table Question Answering with TAPAS](https://huggingface.co/google/tapas-base-finetuned-wtq)
 - [Visual Question Answering with ViLT](https://huggingface.co/dandelin/vilt-b32-finetuned-vqa)
-- [Image captioning with LLaVa](https://huggingface.co/llava-hf/llava-1.5-7b-hf)
-- [Zero-shot Image Classification with SigLIP](https://huggingface.co/google/siglip-so400m-patch14-384)
-- [Document Question Answering with LayoutLM](https://huggingface.co/impira/layoutlm-document-qa)
-- [Zero-shot Video Classification with X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)
-- [Zero-shot Object Detection with OWLv2](https://huggingface.co/docs/transformers/en/model_doc/owlv2)
-- [Zero-shot Image Segmentation with CLIPSeg](https://huggingface.co/docs/transformers/model_doc/clipseg)
-- [Automatic Mask Generation with SAM](https://huggingface.co/docs/transformers/model_doc/sam)
 
-
-## 100 projects using Transformers
-
-Transformers is more than a toolkit to use pretrained models: it's a community of projects built around it and the
-Hugging Face Hub. We want Transformers to enable developers, researchers, students, professors, engineers, and anyone
-else to build their dream projects.
-
-In order to celebrate the 100,000 stars of transformers, we have decided to put the spotlight on the
-community, and we have created the [awesome-transformers](./awesome-transformers.md) page which lists 100
-incredible projects built in the vicinity of transformers.
-
-If you own or use a project that you believe should be part of the list, please open a PR to add it!
+**[Write With Transformer](https://transformer.huggingface.co)**, built by the Hugging Face team, is the official demo of this repo’s text generation capabilities.
 
 ## If you are looking for custom support from the Hugging Face team
 
@@ -145,7 +118,7 @@ To immediately use a model on a given input (text, image, audio, ...), we provid
 [{'label': 'POSITIVE', 'score': 0.9996980428695679}]
 ```
 
-The second line of code downloads and caches the pretrained model used by the pipeline, while the third evaluates it on the given text. Here, the answer is "positive" with a confidence of 99.97%.
+The second line of code downloads and caches the pretrained model used by the pipeline, while the third evaluates it on the given text. Here the answer is "positive" with a confidence of 99.97%.
 
 Many tasks have a pre-trained `pipeline` ready to go, in NLP but also in computer vision and speech. For example, we can easily extract detected objects in an image:
 
@@ -160,7 +133,7 @@ Many tasks have a pre-trained `pipeline` ready to go, in NLP but also in compute
 >>> image = Image.open(image_data)
 
 # Allocate a pipeline for object detection
->>> object_detector = pipeline('object-detection')
+>>> object_detector = pipeline('object_detection')
 >>> object_detector(image)
 [{'score': 0.9982201457023621,
   'label': 'remote',
@@ -179,7 +152,7 @@ Many tasks have a pre-trained `pipeline` ready to go, in NLP but also in compute
   'box': {'xmin': 345, 'ymin': 23, 'xmax': 640, 'ymax': 368}}]
 ```
 
-Here, we get a list of objects detected in the image, with a box surrounding the object and a confidence score. Here is the original image on the left, with the predictions displayed on the right:
+Here we get a list of objects detected in the image, with a box surrounding the object and a confidence score. Here is the original image on the right, with the predictions displayed on the left:
 
 <h3 align="center">
     <a><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample.png" width="400"></a>
@@ -192,8 +165,8 @@ In addition to `pipeline`, to download and use any of the pretrained models on y
 ```python
 >>> from transformers import AutoTokenizer, AutoModel
 
->>> tokenizer = AutoTokenizer.from_pretrained("google-bert/bert-base-uncased")
->>> model = AutoModel.from_pretrained("google-bert/bert-base-uncased")
+>>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
+>>> model = AutoModel.from_pretrained("bert-base-uncased")
 
 >>> inputs = tokenizer("Hello world!", return_tensors="pt")
 >>> outputs = model(**inputs)
@@ -203,14 +176,14 @@ And here is the equivalent code for TensorFlow:
 ```python
 >>> from transformers import AutoTokenizer, TFAutoModel
 
->>> tokenizer = AutoTokenizer.from_pretrained("google-bert/bert-base-uncased")
->>> model = TFAutoModel.from_pretrained("google-bert/bert-base-uncased")
+>>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
+>>> model = TFAutoModel.from_pretrained("bert-base-uncased")
 
 >>> inputs = tokenizer("Hello world!", return_tensors="tf")
 >>> outputs = model(**inputs)
 ```
 
-The tokenizer is responsible for all the preprocessing the pretrained model expects and can be called directly on a single string (as in the above examples) or a list. It will output a dictionary that you can use in downstream code or simply directly pass to your model using the ** argument unpacking operator.
+The tokenizer is responsible for all the preprocessing the pretrained model expects, and can be called directly on a single string (as in the above examples) or a list. It will output a dictionary that you can use in downstream code or simply directly pass to your model using the ** argument unpacking operator.
 
 The model itself is a regular [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) or a [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) (depending on your backend) which you can use as usual. [This tutorial](https://huggingface.co/docs/transformers/training) explains how to integrate such a model into a classic PyTorch or TensorFlow training loop, or how to use our `Trainer` API to quickly fine-tune on a new dataset.
 
@@ -225,12 +198,12 @@ The model itself is a regular [Pytorch `nn.Module`](https://pytorch.org/docs/sta
 1. Lower compute costs, smaller carbon footprint:
     - Researchers can share trained models instead of always retraining.
     - Practitioners can reduce compute time and production costs.
-    - Dozens of architectures with over 400,000 pretrained models across all modalities.
+    - Dozens of architectures with over 60,000 pretrained models across all modalities.
 
 1. Choose the right framework for every part of a model's lifetime:
     - Train state-of-the-art models in 3 lines of code.
     - Move a single model between TF2.0/PyTorch/JAX frameworks at will.
-    - Seamlessly pick the right framework for training, evaluation, and production.
+    - Seamlessly pick the right framework for training, evaluation and production.
 
 1. Easily customize a model or an example to your needs:
     - We provide examples for each architecture to reproduce the results published by its original authors.
@@ -241,20 +214,20 @@ The model itself is a regular [Pytorch `nn.Module`](https://pytorch.org/docs/sta
 
 - This library is not a modular toolbox of building blocks for neural nets. The code in the model files is not refactored with additional abstractions on purpose, so that researchers can quickly iterate on each of the models without diving into additional abstractions/files.
 - The training API is not intended to work on any model but is optimized to work with the models provided by the library. For generic machine learning loops, you should use another library (possibly, [Accelerate](https://huggingface.co/docs/accelerate)).
-- While we strive to present as many use cases as possible, the scripts in our [examples folder](https://github.com/huggingface/transformers/tree/main/examples) are just that: examples. It is expected that they won't work out-of-the-box on your specific problem and that you will be required to change a few lines of code to adapt them to your needs.
+- While we strive to present as many use cases as possible, the scripts in our [examples folder](https://github.com/huggingface/transformers/tree/main/examples) are just that: examples. It is expected that they won't work out-of-the box on your specific problem and that you will be required to change a few lines of code to adapt them to your needs.
 
 ## Installation
 
 ### With pip
 
-This repository is tested on Python 3.8+, Flax 0.4.1+, PyTorch 1.11+, and TensorFlow 2.6+.
+This repository is tested on Python 3.6+, Flax 0.3.2+, PyTorch 1.3.1+ and TensorFlow 2.3+.
 
 You should install 🤗 Transformers in a [virtual environment](https://docs.python.org/3/library/venv.html). If you're unfamiliar with Python virtual environments, check out the [user guide](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/).
 
 First, create a virtual environment with the version of Python you're going to use and activate it.
 
-Then, you will need to install at least one of Flax, PyTorch, or TensorFlow.
-Please refer to [TensorFlow installation page](https://www.tensorflow.org/install/), [PyTorch installation page](https://pytorch.org/get-started/locally/#start-locally) and/or [Flax](https://github.com/google/flax#quick-install) and [Jax](https://github.com/google/jax#installation) installation pages regarding the specific installation command for your platform.
+Then, you will need to install at least one of Flax, PyTorch or TensorFlow.
+Please refer to [TensorFlow installation page](https://www.tensorflow.org/install/), [PyTorch installation page](https://pytorch.org/get-started/locally/#start-locally) and/or [Flax](https://github.com/google/flax#quick-install) and [Jax](https://github.com/google/jax#installation) installation pages regarding the specific install command for your platform.
 
 When one of those backends has been installed, 🤗 Transformers can be installed using pip as follows:
 
@@ -266,29 +239,169 @@ If you'd like to play with the examples or need the bleeding edge of the code an
 
 ### With conda
 
+Since Transformers version v4.0.0, we now have a conda channel: `huggingface`.
+
 🤗 Transformers can be installed using conda as follows:
 
 ```shell script
-conda install conda-forge::transformers
+conda install -c huggingface transformers
 ```
 
-> **_NOTE:_** Installing `transformers` from the `huggingface` channel is deprecated.
-
 Follow the installation pages of Flax, PyTorch or TensorFlow to see how to install them with conda.
 
 > **_NOTE:_**  On Windows, you may be prompted to activate Developer Mode in order to benefit from caching. If this is not an option for you, please let us know in [this issue](https://github.com/huggingface/huggingface_hub/issues/1062).
 
 ## Model architectures
 
-**[All the model checkpoints](https://huggingface.co/models)** provided by 🤗 Transformers are seamlessly integrated from the huggingface.co [model hub](https://huggingface.co/models), where they are uploaded directly by [users](https://huggingface.co/users) and [organizations](https://huggingface.co/organizations).
+**[All the model checkpoints](https://huggingface.co/models)** provided by 🤗 Transformers are seamlessly integrated from the huggingface.co [model hub](https://huggingface.co) where they are uploaded directly by [users](https://huggingface.co/users) and [organizations](https://huggingface.co/organizations).
 
 Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
 
-🤗 Transformers currently provides the following architectures: see [here](https://huggingface.co/docs/transformers/model_summary) for a high-level summary of each them.
+🤗 Transformers currently provides the following architectures (see [here](https://huggingface.co/docs/transformers/model_summary) for a high-level summary of each them):
+
+1. **[ALBERT](https://huggingface.co/docs/transformers/model_doc/albert)** (from Google Research and the Toyota Technological Institute at Chicago) released with the paper [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://arxiv.org/abs/1909.11942), by Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut.
+1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/abs/1910.13461) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
+1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (from École polytechnique) released with the paper [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) by Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis.
+1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (from VinAI Research) released with the paper [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701) by Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen.
+1. **[BEiT](https://huggingface.co/docs/transformers/model_doc/beit)** (from Microsoft) released with the paper [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) by Hangbo Bao, Li Dong, Furu Wei.
+1. **[BERT](https://huggingface.co/docs/transformers/model_doc/bert)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
+1. **[BERT For Sequence Generation](https://huggingface.co/docs/transformers/model_doc/bert-generation)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (from VinAI Research) released with the paper [BERTweet: A pre-trained language model for English Tweets](https://aclanthology.org/2020.emnlp-demos.2/) by Dat Quoc Nguyen, Thanh Vu and Anh Tuan Nguyen.
+1. **[BigBird-Pegasus](https://huggingface.co/docs/transformers/model_doc/bigbird_pegasus)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
+1. **[BigBird-RoBERTa](https://huggingface.co/docs/transformers/model_doc/big_bird)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
+1. **[Blenderbot](https://huggingface.co/docs/transformers/model_doc/blenderbot)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
+1. **[BlenderbotSmall](https://huggingface.co/docs/transformers/model_doc/blenderbot-small)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
+1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigSicence Workshop](https://bigscience.huggingface.co/).
+1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
+1. **[ByT5](https://huggingface.co/docs/transformers/model_doc/byt5)** (from Google Research) released with the paper [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) by Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel.
+1. **[CamemBERT](https://huggingface.co/docs/transformers/model_doc/camembert)** (from Inria/Facebook/Sorbonne) released with the paper [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
+1. **[CANINE](https://huggingface.co/docs/transformers/model_doc/canine)** (from Google Research) released with the paper [CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation](https://arxiv.org/abs/2103.06874) by Jonathan H. Clark, Dan Garrette, Iulia Turc, John Wieting.
+1. **[CLIP](https://huggingface.co/docs/transformers/model_doc/clip)** (from OpenAI) released with the paper [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) by Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever.
+1. **[CodeGen](https://huggingface.co/docs/transformers/model_doc/codegen)** (from Salesforce) released with the paper [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) by Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong.
+1. **[ConvBERT](https://huggingface.co/docs/transformers/model_doc/convbert)** (from YituTech) released with the paper [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
+1. **[ConvNeXT](https://huggingface.co/docs/transformers/model_doc/convnext)** (from Facebook AI) released with the paper [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) by Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, Saining Xie.
+1. **[CPM](https://huggingface.co/docs/transformers/model_doc/cpm)** (from Tsinghua University) released with the paper [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun.
+1. **[CTRL](https://huggingface.co/docs/transformers/model_doc/ctrl)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
+1. **[CvT](https://huggingface.co/docs/transformers/model_doc/cvt)** (from Microsoft) released with the paper [CvT: Introducing Convolutions to Vision Transformers](https://arxiv.org/abs/2103.15808) by Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, Lei Zhang.
+1. **[Data2Vec](https://huggingface.co/docs/transformers/model_doc/data2vec)** (from Facebook) released with the paper [Data2Vec:  A General Framework for Self-supervised Learning in Speech, Vision and Language](https://arxiv.org/abs/2202.03555) by Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu, Michael Auli.
+1. **[DeBERTa](https://huggingface.co/docs/transformers/model_doc/deberta)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
+1. **[DeBERTa-v2](https://huggingface.co/docs/transformers/model_doc/deberta-v2)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
+1. **[Decision Transformer](https://huggingface.co/docs/transformers/model_doc/decision_transformer)** (from Berkeley/Facebook/Google) released with the paper [Decision Transformer: Reinforcement Learning via Sequence Modeling](https://arxiv.org/abs/2106.01345) by Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Michael Laskin, Pieter Abbeel, Aravind Srinivas, Igor Mordatch.
+1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
+1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (from Facebook) released with the paper [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko.
+1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (from Microsoft Research) released with the paper [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.
+1. **[DistilBERT](https://huggingface.co/docs/transformers/model_doc/distilbert)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation) and a German version of DistilBERT.
+1. **[DiT](https://huggingface.co/docs/transformers/model_doc/dit)** (from Microsoft Research) released with the paper [DiT: Self-supervised Pre-training for Document Image Transformer](https://arxiv.org/abs/2203.02378) by Junlong Li, Yiheng Xu, Tengchao Lv, Lei Cui, Cha Zhang, Furu Wei.
+1. **[Donut](https://huggingface.co/docs/transformers/model_doc/donut)** (from NAVER), released together with the paper [OCR-free Document Understanding Transformer](https://arxiv.org/abs/2111.15664) by Geewook Kim, Teakgyu Hong, Moonbin Yim, Jeongyeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, Seunghyun Park.
+1. **[DPR](https://huggingface.co/docs/transformers/model_doc/dpr)** (from Facebook) released with the paper [Dense Passage Retrieval for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) by Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
+1. **[DPT](https://huggingface.co/docs/transformers/master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.
+1. **[ELECTRA](https://huggingface.co/docs/transformers/model_doc/electra)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
+1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (from Google Research) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+1. **[ERNIE](https://huggingface.co/docs/transformers/model_doc/ernie)** (from Baidu) released with the paper [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu.
+1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
+1. **[FLAVA](https://huggingface.co/docs/transformers/model_doc/flava)** (from Facebook AI) released with the paper [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) by Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela.
+1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
+1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
+1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
+1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
+1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
+1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach
+1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
+1. **[GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj)** (from EleutherAI) released in the repository [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) by Ben Wang and Aran Komatsuzaki.
+1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (from UCSD, NVIDIA) released with the paper [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) by Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang.
+1. **[Hubert](https://huggingface.co/docs/transformers/model_doc/hubert)** (from Facebook) released with the paper [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://arxiv.org/abs/2106.07447) by Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed.
+1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (from Berkeley) released with the paper [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer.
+1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (from OpenAI) released with the paper [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) by Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever.
+1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
+1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
+1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (from Microsoft Research Asia) released with the paper [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) by Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei.
+1. **[LayoutXLM](https://huggingface.co/docs/transformers/model_doc/layoutxlm)** (from Microsoft Research Asia) released with the paper [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) by Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei.
+1. **[LED](https://huggingface.co/docs/transformers/model_doc/led)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
+1. **[LeViT](https://huggingface.co/docs/transformers/model_doc/levit)** (from Meta AI) released with the paper [LeViT: A Vision Transformer in ConvNet's Clothing for Faster Inference](https://arxiv.org/abs/2104.01136) by Ben Graham, Alaaeldin El-Nouby, Hugo Touvron, Pierre Stock, Armand Joulin, Hervé Jégou, Matthijs Douze.
+1. **[Longformer](https://huggingface.co/docs/transformers/model_doc/longformer)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
+1. **[LongT5](https://huggingface.co/docs/transformers/model_doc/longt5)** (from Google AI) released with the paper [LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/abs/2112.07916) by Mandy Guo, Joshua Ainslie, David Uthus, Santiago Ontanon, Jianmo Ni, Yun-Hsuan Sung, Yinfei Yang.
+1. **[LUKE](https://huggingface.co/docs/transformers/model_doc/luke)** (from Studio Ousia) released with the paper [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
+1. **[LXMERT](https://huggingface.co/docs/transformers/model_doc/lxmert)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
+1. **[M-CTC-T](https://huggingface.co/docs/transformers/model_doc/mctct)** (from Facebook) released with the paper [Pseudo-Labeling For Massively Multilingual Speech Recognition](https://arxiv.org/abs/2111.00161) by Loren Lugosch, Tatiana Likhomanenko, Gabriel Synnaeve, and Ronan Collobert.
+1. **[M2M100](https://huggingface.co/docs/transformers/model_doc/m2m_100)** (from Facebook) released with the paper [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
+1. **[MarianMT](https://huggingface.co/docs/transformers/model_doc/marian)** Machine translation models trained using [OPUS](http://opus.nlpl.eu/) data by Jörg Tiedemann. The [Marian Framework](https://marian-nmt.github.io/) is being developed by the Microsoft Translator Team.
+1. **[MaskFormer](https://huggingface.co/docs/transformers/model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov.
+1. **[mBART](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
+1. **[mBART-50](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Translation with Extensible Multilingual Pretraining and Finetuning](https://arxiv.org/abs/2008.00401) by Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan.
+1. **[Megatron-BERT](https://huggingface.co/docs/transformers/model_doc/megatron-bert)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
+1. **[Megatron-GPT2](https://huggingface.co/docs/transformers/model_doc/megatron_gpt2)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
+1. **[mLUKE](https://huggingface.co/docs/transformers/model_doc/mluke)** (from Studio Ousia) released with the paper [mLUKE: The Power of Entity Representations in Multilingual Pretrained Language Models](https://arxiv.org/abs/2110.08151) by Ryokan Ri, Ikuya Yamada, and Yoshimasa Tsuruoka.
+1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (from CMU/Google Brain) released with the paper [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) by Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou.
+1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (from Apple) released with the paper [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) by Sachin Mehta and Mohammad Rastegari.
+1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
+1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
+1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
+1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
+1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
+1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
+1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
+1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
+1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
+1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, and Peter J. Liu.
+1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
+1. **[PhoBERT](https://huggingface.co/docs/transformers/model_doc/phobert)** (from VinAI Research) released with the paper [PhoBERT: Pre-trained language models for Vietnamese](https://www.aclweb.org/anthology/2020.findings-emnlp.92/) by Dat Quoc Nguyen and Anh Tuan Nguyen.
+1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (from UCLA NLP) released with the paper [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
+1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (from Sea AI Labs) released with the paper [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) by Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng.
+1. **[ProphetNet](https://huggingface.co/docs/transformers/model_doc/prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
+1. **[QDQBert](https://huggingface.co/docs/transformers/model_doc/qdqbert)** (from NVIDIA) released with the paper [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://arxiv.org/abs/2004.09602) by Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev and Paulius Micikevicius.
+1. **[RAG](https://huggingface.co/docs/transformers/model_doc/rag)** (from Facebook) released with the paper [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) by Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela.
+1. **[REALM](https://huggingface.co/docs/transformers/model_doc/realm.html)** (from Google Research) released with the paper [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) by Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang.
+1. **[Reformer](https://huggingface.co/docs/transformers/model_doc/reformer)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
+1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Platforms) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár.
+1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/abs/2010.12821) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
+1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.
+1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (from Facebook), released together with the paper [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
+1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/abs/2104.09864) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
+1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
+1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
+1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
+1. **[SpeechToTextTransformer](https://huggingface.co/docs/transformers/model_doc/speech_to_text)** (from Facebook), released together with the paper [fairseq S2T: Fast Speech-to-Text Modeling with fairseq](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Dmytro Okhonko, Juan Pino.
+1. **[SpeechToTextTransformer2](https://huggingface.co/docs/transformers/model_doc/speech_to_text_2)** (from Facebook), released together with the paper [Large-Scale Self- and Semi-Supervised Learning for Speech Translation](https://arxiv.org/abs/2104.06678) by Changhan Wang, Anne Wu, Juan Pino, Alexei Baevski, Michael Auli, Alexis Conneau.
+1. **[Splinter](https://huggingface.co/docs/transformers/model_doc/splinter)** (from Tel Aviv University), released together with the paper [Few-Shot Question Answering by Pretraining Span Selection](https://arxiv.org/abs/2101.00438) by Ori Ram, Yuval Kirstain, Jonathan Berant, Amir Globerson, Omer Levy.
+1. **[SqueezeBERT](https://huggingface.co/docs/transformers/model_doc/squeezebert)** (from Berkeley) released with the paper [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) by Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer.
+1. **[Swin Transformer](https://huggingface.co/docs/transformers/model_doc/swin)** (from Microsoft) released with the paper [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) by Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo.
+1. **[Swin Transformer V2](https://huggingface.co/docs/transformers/model_doc/swinv2)** (from Microsoft) released with the paper [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883) by Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo.
+1. **[T5](https://huggingface.co/docs/transformers/model_doc/t5)** (from Google AI) released with the paper [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://arxiv.org/abs/1910.10683) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
+1. **[T5v1.1](https://huggingface.co/docs/transformers/model_doc/t5v1.1)** (from Google AI) released in the repository [google-research/text-to-text-transfer-transformer](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/released_checkpoints.md#t511) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
+1. **[TAPAS](https://huggingface.co/docs/transformers/model_doc/tapas)** (from Google AI) released with the paper [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
+1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (from Microsoft Research) released with the paper [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) by Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou.
+1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine
+1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
+1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (from Microsoft), released together with the paper [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) by Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei.
+1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler
+1. **[UniSpeech](https://huggingface.co/docs/transformers/model_doc/unispeech)** (from Microsoft Research) released with the paper [UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data](https://arxiv.org/abs/2101.07597) by Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei, Michael Zeng, Xuedong Huang.
+1. **[UniSpeechSat](https://huggingface.co/docs/transformers/model_doc/unispeech-sat)** (from Microsoft Research) released with the paper [UNISPEECH-SAT: UNIVERSAL SPEECH REPRESENTATION LEARNING WITH SPEAKER AWARE PRE-TRAINING](https://arxiv.org/abs/2110.05752) by Sanyuan Chen, Yu Wu, Chengyi Wang, Zhengyang Chen, Zhuo Chen, Shujie Liu, Jian Wu, Yao Qian, Furu Wei, Jinyu Li, Xiangzhan Yu.
+1. **[VAN](https://huggingface.co/docs/transformers/model_doc/van)** (from Tsinghua University and Nankai University) released with the paper [Visual Attention Network](https://arxiv.org/abs/2202.09741) by Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu.
+1. **[VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)** (from Multimedia Computing Group, Nanjing University) released with the paper [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://arxiv.org/abs/2203.12602) by Zhan Tong, Yibing Song, Jue Wang, Limin Wang.
+1. **[ViLT](https://huggingface.co/docs/transformers/model_doc/vilt)** (from NAVER AI Lab/Kakao Enterprise/Kakao Brain) released with the paper [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://arxiv.org/abs/2102.03334) by Wonjae Kim, Bokyung Son, Ildoo Kim.
+1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
+1. **[VisualBERT](https://huggingface.co/docs/transformers/model_doc/visual_bert)** (from UCLA NLP) released with the paper [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) by Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
+1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
+1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
+1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
+1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (from Facebook AI) released with the paper [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) by Qiantong Xu, Alexei Baevski, Michael Auli.
+1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
+1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
+1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
+1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
+1. **[XLM-ProphetNet](https://huggingface.co/docs/transformers/model_doc/xlm-prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
+1. **[XLM-RoBERTa](https://huggingface.co/docs/transformers/model_doc/xlm-roberta)** (from Facebook AI), released together with the paper [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.
+1. **[XLM-RoBERTa-XL](https://huggingface.co/docs/transformers/model_doc/xlm-roberta-xl)** (from Facebook AI), released together with the paper [Larger-Scale Transformers for Multilingual Masked Language Modeling](https://arxiv.org/abs/2105.00572) by Naman Goyal, Jingfei Du, Myle Ott, Giri Anantharaman, Alexis Conneau.
+1. **[XLNet](https://huggingface.co/docs/transformers/model_doc/xlnet)** (from Google/CMU) released with the paper [​XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) by Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
+1. **[XLS-R](https://huggingface.co/docs/transformers/model_doc/xls_r)** (from Facebook AI) released with the paper [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) by Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli.
+1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
+1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (from Huazhong University of Science & Technology) released with the paper [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
+1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (from the University of Wisconsin - Madison) released with the paper [You Only Sample (Almost) Once: Linear Cost Self-Attention Via Bernoulli Sampling](https://arxiv.org/abs/2111.09714) by Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh.
+1. Want to contribute a new model? We have added a **detailed guide and templates** to guide you in the process of adding a new model. You can find them in the [`templates`](./templates) folder of the repository. Be sure to check the [contributing guidelines](./CONTRIBUTING.md) and contact the maintainers or open an issue to collect feedbacks before starting your PR.
 
 To check if each model has an implementation in Flax, PyTorch or TensorFlow, or has an associated tokenizer backed by the 🤗 Tokenizers library, refer to [this table](https://huggingface.co/docs/transformers/index#supported-frameworks).
 
-These implementations have been tested on several datasets (see the example scripts) and should match the performance of the original implementations. You can find more details on performance in the Examples section of the [documentation](https://github.com/huggingface/transformers/tree/main/examples).
+These implementations have been tested on several datasets (see the example scripts) and should match the performance of the original implementations. You can find more details on performance in the Examples section of the [documentation](https://huggingface.co/docs/transformers/examples).
 
 
 ## Learn more
@@ -301,6 +414,7 @@ These implementations have been tested on several datasets (see the example scri
 | [Training and fine-tuning](https://huggingface.co/docs/transformers/training) | Using the models provided by 🤗 Transformers in a PyTorch/TensorFlow training loop and the `Trainer` API |
 | [Quick tour: Fine-tuning/usage scripts](https://github.com/huggingface/transformers/tree/main/examples) | Example scripts for fine-tuning models on a wide range of tasks |
 | [Model sharing and uploading](https://huggingface.co/docs/transformers/model_sharing) | Upload and share your fine-tuned models with the community |
+| [Migration](https://huggingface.co/docs/transformers/migration) | Migrate to 🤗 Transformers from `pytorch-transformers` or `pytorch-pretrained-bert` |
 
 ## Citation
 

README_ko.md

@@ -0,0 +1,383 @@
+<!---
+Copyright 2020 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+-->
+
+<p align="center">
+    <br>
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_logo_name.png" width="400"/>
+    <br>
+<p>
+<p align="center">
+    <a href="https://circleci.com/gh/huggingface/transformers">
+        <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
+    </a>
+    <a href="https://github.com/huggingface/transformers/blob/main/LICENSE">
+        <img alt="GitHub" src="https://img.shields.io/github/license/huggingface/transformers.svg?color=blue">
+    </a>
+    <a href="https://huggingface.co/docs/transformers/index">
+        <img alt="Documentation" src="https://img.shields.io/website/http/huggingface.co/docs/transformers/index.svg?down_color=red&down_message=offline&up_message=online">
+    </a>
+    <a href="https://github.com/huggingface/transformers/releases">
+        <img alt="GitHub release" src="https://img.shields.io/github/release/huggingface/transformers.svg">
+    </a>
+    <a href="https://github.com/huggingface/transformers/blob/main/CODE_OF_CONDUCT.md">
+        <img alt="Contributor Covenant" src="https://img.shields.io/badge/Contributor%20Covenant-v2.0%20adopted-ff69b4.svg">
+    </a>
+    <a href="https://zenodo.org/badge/latestdoi/155220641"><img src="https://zenodo.org/badge/155220641.svg" alt="DOI"></a>
+</p>
+
+<h4 align="center">
+    <p>
+        <a href="https://github.com/huggingface/transformers/">English</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hans.md">简体中文</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hant.md">繁體中文</a> |
+        <b>한국어</b>
+    <p>
+</h4>
+
+<h3 align="center">
+    <p> Jax, Pytorch, TensorFlow를 위한 최첨단 자연어처리</p>
+</h3>
+
+<h3 align="center">
+    <a href="https://hf.co/course"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/course_banner.png"></a>
+</h3>
+
+🤗 Transformers는 분류, 정보 추출, 질문 답변, 요약, 번역, 문장 생성 등을 100개 이상의 언어로 수행할 수 있는 수천개의 사전학습된 모델을 제공합니다. 우리의 목표는 모두가 최첨단의 NLP 기술을 쉽게 사용하는 것입니다.
+
+🤗 Transformers는 이러한 사전학습 모델을 빠르게 다운로드해 특정 텍스트에 사용하고, 원하는 데이터로 fine-tuning해 커뮤니티나 우리의 [모델 허브](https://huggingface.co/models)에 공유할 수 있도록 API를 제공합니다. 또한, 모델 구조를 정의하는 각 파이썬 모듈은 완전히 독립적이여서 연구 실험을 위해 손쉽게 수정할 수 있습니다.
+
+🤗 Transformers는 가장 유명한 3개의 딥러닝 라이브러리를 지원합니다. 이들은 서로 완벽히 연동됩니다 — [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/), [TensorFlow](https://www.tensorflow.org/). 간단하게 이 라이브러리 중 하나로 모델을 학습하고, 또 다른 라이브러리로 추론을 위해 모델을 불러올 수 있습니다. 
+
+## 온라인 데모
+
+대부분의 모델을 [모델 허브](https://huggingface.co/models) 페이지에서 바로 테스트해볼 수 있습니다. 공개 및 비공개 모델을 위한 [비공개 모델 호스팅, 버전 관리, 추론 API](https://huggingface.co/pricing)도 제공합니다.
+
+예시:
+- [BERT로 마스킹된 단어 완성하기](https://huggingface.co/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
+- [Electra를 이용한 개체명 인식](https://huggingface.co/dbmdz/electra-large-discriminator-finetuned-conll03-english?text=My+name+is+Sarah+and+I+live+in+London+city)
+- [GPT-2로 텍스트 생성하기](https://huggingface.co/gpt2?text=A+long+time+ago%2C+)
+- [RoBERTa로 자연어 추론하기](https://huggingface.co/roberta-large-mnli?text=The+dog+was+lost.+Nobody+lost+any+animal)
+- [BART를 이용한 요약](https://huggingface.co/facebook/bart-large-cnn?text=The+tower+is+324+metres+%281%2C063+ft%29+tall%2C+about+the+same+height+as+an+81-storey+building%2C+and+the+tallest+structure+in+Paris.+Its+base+is+square%2C+measuring+125+metres+%28410+ft%29+on+each+side.+During+its+construction%2C+the+Eiffel+Tower+surpassed+the+Washington+Monument+to+become+the+tallest+man-made+structure+in+the+world%2C+a+title+it+held+for+41+years+until+the+Chrysler+Building+in+New+York+City+was+finished+in+1930.+It+was+the+first+structure+to+reach+a+height+of+300+metres.+Due+to+the+addition+of+a+broadcasting+aerial+at+the+top+of+the+tower+in+1957%2C+it+is+now+taller+than+the+Chrysler+Building+by+5.2+metres+%2817+ft%29.+Excluding+transmitters%2C+the+Eiffel+Tower+is+the+second+tallest+free-standing+structure+in+France+after+the+Millau+Viaduct)
+- [DistilBERT를 이용한 질문 답변](https://huggingface.co/distilbert-base-uncased-distilled-squad?text=Which+name+is+also+used+to+describe+the+Amazon+rainforest+in+English%3F&context=The+Amazon+rainforest+%28Portuguese%3A+Floresta+Amaz%C3%B4nica+or+Amaz%C3%B4nia%3B+Spanish%3A+Selva+Amaz%C3%B3nica%2C+Amazon%C3%ADa+or+usually+Amazonia%3B+French%3A+For%C3%AAt+amazonienne%3B+Dutch%3A+Amazoneregenwoud%29%2C+also+known+in+English+as+Amazonia+or+the+Amazon+Jungle%2C+is+a+moist+broadleaf+forest+that+covers+most+of+the+Amazon+basin+of+South+America.+This+basin+encompasses+7%2C000%2C000+square+kilometres+%282%2C700%2C000+sq+mi%29%2C+of+which+5%2C500%2C000+square+kilometres+%282%2C100%2C000+sq+mi%29+are+covered+by+the+rainforest.+This+region+includes+territory+belonging+to+nine+nations.+The+majority+of+the+forest+is+contained+within+Brazil%2C+with+60%25+of+the+rainforest%2C+followed+by+Peru+with+13%25%2C+Colombia+with+10%25%2C+and+with+minor+amounts+in+Venezuela%2C+Ecuador%2C+Bolivia%2C+Guyana%2C+Suriname+and+French+Guiana.+States+or+departments+in+four+nations+contain+%22Amazonas%22+in+their+names.+The+Amazon+represents+over+half+of+the+planet%27s+remaining+rainforests%2C+and+comprises+the+largest+and+most+biodiverse+tract+of+tropical+rainforest+in+the+world%2C+with+an+estimated+390+billion+individual+trees+divided+into+16%2C000+species)
+- [T5로 번역하기](https://huggingface.co/t5-base?text=My+name+is+Wolfgang+and+I+live+in+Berlin)
+
+**[Transformer와 글쓰기](https://transformer.huggingface.co)** 는 이 저장소의 텍스트 생성 능력에 관한 Hugging Face 팀의 공식 데모입니다. 
+
+## Hugging Face 팀의 커스텀 지원을 원한다면
+
+<a target="_blank" href="https://huggingface.co/support">
+    <img alt="HuggingFace Expert Acceleration Program" src="https://huggingface.co/front/thumbnails/support.png" style="max-width: 600px; border: 1px solid #eee; border-radius: 4px; box-shadow: 0 1px 2px 0 rgba(0, 0, 0, 0.05);">
+</a><br>
+
+## 퀵 투어
+
+원하는 텍스트에 바로 모델을 사용할 수 있도록, 우리는 `pipeline` API를 제공합니다. Pipeline은 사전학습 모델과 그 모델을 학습할 때 적용한 전처리 방식을 하나로 합칩니다. 다음은 긍정적인 텍스트와 부정적인 텍스트를 분류하기 위해 pipeline을 사용한 간단한 예시입니다:
+
+```python
+>>> from transformers import pipeline
+
+# Allocate a pipeline for sentiment-analysis
+>>> classifier = pipeline('sentiment-analysis')
+>>> classifier('We are very happy to introduce pipeline to the transformers repository.')
+[{'label': 'POSITIVE', 'score': 0.9996980428695679}]
+```
+
+코드의 두번째 줄은 pipeline이 사용하는 사전학습 모델을 다운로드하고 캐시로 저장합니다. 세번째 줄에선 그 모델이 주어진 텍스트를 평가합니다. 여기서 모델은 99.97%의 확률로 텍스트가 긍정적이라고 평가했습니다.
+
+많은 NLP 과제들을 `pipeline`으로 바로 수행할 수 있습니다. 예를 들어, 질문과 문맥이 주어지면 손쉽게 답변을 추출할 수 있습니다:
+
+``` python
+>>> from transformers import pipeline
+
+# Allocate a pipeline for question-answering
+>>> question_answerer = pipeline('question-answering')
+>>> question_answerer({
+...     'question': 'What is the name of the repository ?',
+...     'context': 'Pipeline has been included in the huggingface/transformers repository'
+... })
+{'score': 0.30970096588134766, 'start': 34, 'end': 58, 'answer': 'huggingface/transformers'}
+
+```
+
+답변뿐만 아니라, 여기에 사용된 사전학습 모델은 확신도와 토크나이즈된 문장 속 답변의 시작점, 끝점까지 반환합니다. [이 튜토리얼](https://huggingface.co/docs/transformers/task_summary)에서 `pipeline` API가 지원하는 다양한 과제를 확인할 수 있습니다.
+
+코드 3줄로 원하는 과제에 맞게 사전학습 모델을 다운로드 받고 사용할 수 있습니다. 다음은 PyTorch 버전입니다:
+```python
+>>> from transformers import AutoTokenizer, AutoModel
+
+>>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
+>>> model = AutoModel.from_pretrained("bert-base-uncased")
+
+>>> inputs = tokenizer("Hello world!", return_tensors="pt")
+>>> outputs = model(**inputs)
+```
+다음은 TensorFlow 버전입니다:
+```python
+>>> from transformers import AutoTokenizer, TFAutoModel
+
+>>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
+>>> model = TFAutoModel.from_pretrained("bert-base-uncased")
+
+>>> inputs = tokenizer("Hello world!", return_tensors="tf")
+>>> outputs = model(**inputs)
+```
+
+토크나이저는 사전학습 모델의 모든 전처리를 책임집니다. 그리고 (위의 예시처럼) 1개의 스트링이나 리스트도 처리할 수 있습니다. 토크나이저는 딕셔너리를 반환하는데, 이는 다운스트림 코드에 사용하거나 언패킹 연산자 ** 를 이용해 모델에 바로 전달할 수도 있습니다.
+
+모델 자체는 일반적으로 사용되는 [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module)나 [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model)입니다. [이 튜토리얼](https://huggingface.co/transformers/training.html)은 이러한 모델을 표준적인 PyTorch나 TensorFlow 학습 과정에서 사용하는 방법, 또는 새로운 데이터로 fine-tune하기 위해 `Trainer` API를 사용하는 방법을 설명해줍니다.
+
+## 왜 transformers를 사용해야 할까요?
+
+1. 손쉽게 사용할 수 있는 최첨단 모델:
+    - NLU와 NLG 과제에서 뛰어난 성능을 보입니다.
+    - 교육자 실무자에게 진입 장벽이 낮습니다.
+    - 3개의 클래스만 배우면 바로 사용할 수 있습니다.
+    - 하나의 API로 모든 사전학습 모델을 사용할 수 있습니다.
+
+1. 더 적은 계산 비용, 더 적은 탄소 발자국:
+    - 연구자들은 모델을 계속 다시 학습시키는 대신 학습된 모델을 공유할 수 있습니다.
+    - 실무자들은 학습에 필요한 시간과 비용을 절약할 수 있습니다.
+    - 수십개의 모델 구조, 2,000개 이상의 사전학습 모델, 100개 이상의 언어로 학습된 모델 등.
+
+1. 모델의 각 생애주기에 적합한 프레임워크:
+    - 코드 3줄로 최첨단 모델을 학습하세요.
+    - 자유롭게 모델을 TF2.0나 PyTorch 프레임워크로 변환하세요.
+    - 학습, 평가, 공개 등 각 단계에 맞는 프레임워크를 원하는대로 선택하세요.
+
+1. 필요한 대로 모델이나 예시를 커스터마이즈하세요:
+    - 우리는 저자가 공개한 결과를 재현하기 위해 각 모델 구조의 예시를 제공합니다.
+    - 모델 내부 구조는 가능한 일관적으로 공개되어 있습니다.
+    - 빠른 실험을 위해 모델 파일은 라이브러리와 독립적으로 사용될 수 있습니다.
+
+## 왜 transformers를 사용하지 말아야 할까요?
+
+- 이 라이브러리는 신경망 블록을 만들기 위한 모듈이 아닙니다. 연구자들이 여러 파일을 살펴보지 않고 바로 각 모델을 사용할 수 있도록, 모델 파일 코드의 추상화 수준을 적정하게 유지했습니다.
+- 학습 API는 모든 모델에 적용할 수 있도록 만들어지진 않았지만, 라이브러리가 제공하는 모델들에 적용할 수 있도록 최적화되었습니다. 일반적인 머신 러닝을 위해선, 다른 라이브러리를 사용하세요.
+- 가능한 많은 사용 예시를 보여드리고 싶어서, [예시 폴더](https://github.com/huggingface/transformers/tree/main/examples)의 스크립트를 준비했습니다. 이 스크립트들을 수정 없이 특정한 문제에 바로 적용하지 못할 수 있습니다. 필요에 맞게 일부 코드를 수정해야 할 수 있습니다.
+
+## 설치
+
+### pip로 설치하기
+
+이 저장소는 Python 3.6+, Flax 0.3.2+, PyTorch 1.3.1+, TensorFlow 2.3+에서 테스트 되었습니다.
+
+[가상 환경](https://docs.python.org/3/library/venv.html)에 🤗 Transformers를 설치하세요. Python 가상 환경에 익숙하지 않다면, [사용자 가이드](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/)를 확인하세요.
+
+우선, 사용할 Python 버전으로 가상 환경을 만들고 실행하세요.
+
+그 다음, Flax, PyTorch, TensorFlow 중 적어도 하나는 설치해야 합니다.
+플랫폼에 맞는 설치 명령어를 확인하기 위해 [TensorFlow 설치 페이지](https://www.tensorflow.org/install/), [PyTorch 설치 페이지](https://pytorch.org/get-started/locally/#start-locally), [Flax 설치 페이지](https://github.com/google/flax#quick-install)를 확인하세요.
+
+이들 중 적어도 하나가 설치되었다면, 🤗 Transformers는 다음과 같이 pip을 이용해 설치할 수 있습니다:
+
+```bash
+pip install transformers
+```
+
+예시들을 체험해보고 싶거나, 최최최첨단 코드를 원하거나, 새로운 버전이 나올 때까지 기다릴 수 없다면 [라이브러리를 소스에서 바로 설치](https://huggingface.co/docs/transformers/installation#installing-from-source)하셔야 합니다.
+
+### conda로 설치하기
+
+Transformers 버전 v4.0.0부터, conda 채널이 생겼습니다: `huggingface`.
+
+🤗 Transformers는 다음과 같이 conda로 설치할 수 있습니다:
+
+```shell script
+conda install -c huggingface transformers
+```
+
+Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는 방법을 확인하세요.
+
+## 모델 구조
+
+**🤗 Transformers가 제공하는 [모든 모델 체크포인트](https://huggingface.co/models)** 는 huggingface.co [모델 허브](https://huggingface.co)에 완벽히 연동되어 있습니다. [개인](https://huggingface.co/users)과 [기관](https://huggingface.co/organizations)이 모델 허브에 직접 업로드할 수 있습니다.
+
+현재 사용 가능한 모델 체크포인트의 개수: ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
+
+🤗 Transformers는 다음 모델들을 제공합니다 (각 모델의 요약은 [여기](https://huggingface.co/docs/transformers/model_summary)서 확인하세요):
+
+1. **[ALBERT](https://huggingface.co/docs/transformers/model_doc/albert)** (from Google Research and the Toyota Technological Institute at Chicago) released with the paper [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://arxiv.org/abs/1909.11942), by Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut.
+1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/pdf/1910.13461.pdf) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
+1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (from École polytechnique) released with the paper [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) by Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis.
+1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (from VinAI Research) released with the paper [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701) by Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen.
+1. **[BEiT](https://huggingface.co/docs/transformers/model_doc/beit)** (from Microsoft) released with the paper [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) by Hangbo Bao, Li Dong, Furu Wei.
+1. **[BERT](https://huggingface.co/docs/transformers/model_doc/bert)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
+1. **[BERT For Sequence Generation](https://huggingface.co/docs/transformers/model_doc/bert-generation)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (from VinAI Research) released with the paper [BERTweet: A pre-trained language model for English Tweets](https://aclanthology.org/2020.emnlp-demos.2/) by Dat Quoc Nguyen, Thanh Vu and Anh Tuan Nguyen.
+1. **[BigBird-Pegasus](https://huggingface.co/docs/transformers/model_doc/bigbird_pegasus)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
+1. **[BigBird-RoBERTa](https://huggingface.co/docs/transformers/model_doc/big_bird)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
+1. **[Blenderbot](https://huggingface.co/docs/transformers/model_doc/blenderbot)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
+1. **[BlenderbotSmall](https://huggingface.co/docs/transformers/model_doc/blenderbot-small)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
+1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigSicence Workshop](https://bigscience.huggingface.co/).
+1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
+1. **[ByT5](https://huggingface.co/docs/transformers/model_doc/byt5)** (from Google Research) released with the paper [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) by Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel.
+1. **[CamemBERT](https://huggingface.co/docs/transformers/model_doc/camembert)** (from Inria/Facebook/Sorbonne) released with the paper [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
+1. **[CANINE](https://huggingface.co/docs/transformers/model_doc/canine)** (from Google Research) released with the paper [CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation](https://arxiv.org/abs/2103.06874) by Jonathan H. Clark, Dan Garrette, Iulia Turc, John Wieting.
+1. **[CLIP](https://huggingface.co/docs/transformers/model_doc/clip)** (from OpenAI) released with the paper [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) by Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever.
+1. **[CodeGen](https://huggingface.co/docs/transformers/model_doc/codegen)** (from Salesforce) released with the paper [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) by Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong.
+1. **[ConvBERT](https://huggingface.co/docs/transformers/model_doc/convbert)** (from YituTech) released with the paper [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
+1. **[ConvNeXT](https://huggingface.co/docs/transformers/model_doc/convnext)** (from Facebook AI) released with the paper [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) by Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, Saining Xie.
+1. **[CPM](https://huggingface.co/docs/transformers/model_doc/cpm)** (from Tsinghua University) released with the paper [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun.
+1. **[CTRL](https://huggingface.co/docs/transformers/model_doc/ctrl)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
+1. **[CvT](https://huggingface.co/docs/transformers/model_doc/cvt)** (from Microsoft) released with the paper [CvT: Introducing Convolutions to Vision Transformers](https://arxiv.org/abs/2103.15808) by Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, Lei Zhang.
+1. **[Data2Vec](https://huggingface.co/docs/transformers/model_doc/data2vec)** (from Facebook) released with the paper [Data2Vec:  A General Framework for Self-supervised Learning in Speech, Vision and Language](https://arxiv.org/abs/2202.03555) by Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu, Michael Auli.
+1. **[DeBERTa](https://huggingface.co/docs/transformers/model_doc/deberta)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
+1. **[DeBERTa-v2](https://huggingface.co/docs/transformers/model_doc/deberta-v2)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
+1. **[Decision Transformer](https://huggingface.co/docs/transformers/model_doc/decision_transformer)** (from Berkeley/Facebook/Google) released with the paper [Decision Transformer: Reinforcement Learning via Sequence Modeling](https://arxiv.org/abs/2106.01345) by Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Michael Laskin, Pieter Abbeel, Aravind Srinivas, Igor Mordatch.
+1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
+1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (from Facebook) released with the paper [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko.
+1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (from Microsoft Research) released with the paper [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.
+1. **[DistilBERT](https://huggingface.co/docs/transformers/model_doc/distilbert)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/main/examples/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/main/examples/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/main/examples/distillation) and a German version of DistilBERT.
+1. **[DiT](https://huggingface.co/docs/transformers/model_doc/dit)** (from Microsoft Research) released with the paper [DiT: Self-supervised Pre-training for Document Image Transformer](https://arxiv.org/abs/2203.02378) by Junlong Li, Yiheng Xu, Tengchao Lv, Lei Cui, Cha Zhang, Furu Wei.
+1. **[Donut](https://huggingface.co/docs/transformers/model_doc/donut)** (from NAVER) released with the paper [OCR-free Document Understanding Transformer](https://arxiv.org/abs/2111.15664) by Geewook Kim, Teakgyu Hong, Moonbin Yim, Jeongyeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, Seunghyun Park.
+1. **[DPR](https://huggingface.co/docs/transformers/model_doc/dpr)** (from Facebook) released with the paper [Dense Passage Retrieval for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) by Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
+1. **[DPT](https://huggingface.co/docs/transformers/master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.
+1. **[ELECTRA](https://huggingface.co/docs/transformers/model_doc/electra)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
+1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (from Google Research) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+1. **[ERNIE](https://huggingface.co/docs/transformers/model_doc/ernie)** (from Baidu) released with the paper [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu.
+1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
+1. **[FLAVA](https://huggingface.co/docs/transformers/model_doc/flava)** (from Facebook AI) released with the paper [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) by Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela.
+1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
+1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
+1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
+1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
+1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
+1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach 
+1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
+1. **[GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj)** (from EleutherAI) released in the repository [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) by Ben Wang and Aran Komatsuzaki.
+1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (from UCSD, NVIDIA) released with the paper [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) by Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang.
+1. **[Hubert](https://huggingface.co/docs/transformers/model_doc/hubert)** (from Facebook) released with the paper [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://arxiv.org/abs/2106.07447) by Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed.
+1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (from Berkeley) released with the paper [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer.
+1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (from OpenAI) released with the paper [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) by Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever.
+1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
+1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
+1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (from Microsoft Research Asia) released with the paper [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) by Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei.
+1. **[LayoutXLM](https://huggingface.co/docs/transformers/model_doc/layoutxlm)** (from Microsoft Research Asia) released with the paper [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) by Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei.
+1. **[LED](https://huggingface.co/docs/transformers/model_doc/led)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
+1. **[LeViT](https://huggingface.co/docs/transformers/model_doc/levit)** (from Meta AI) released with the paper [LeViT: A Vision Transformer in ConvNet's Clothing for Faster Inference](https://arxiv.org/abs/2104.01136) by Ben Graham, Alaaeldin El-Nouby, Hugo Touvron, Pierre Stock, Armand Joulin, Hervé Jégou, Matthijs Douze.
+1. **[Longformer](https://huggingface.co/docs/transformers/model_doc/longformer)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
+1. **[LongT5](https://huggingface.co/docs/transformers/model_doc/longt5)** (from Google AI) released with the paper [LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/abs/2112.07916) by Mandy Guo, Joshua Ainslie, David Uthus, Santiago Ontanon, Jianmo Ni, Yun-Hsuan Sung, Yinfei Yang.
+1. **[LUKE](https://huggingface.co/docs/transformers/model_doc/luke)** (from Studio Ousia) released with the paper [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
+1. **[LXMERT](https://huggingface.co/docs/transformers/model_doc/lxmert)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
+1. **[M-CTC-T](https://huggingface.co/docs/transformers/model_doc/mctct)** (from Facebook) released with the paper [Pseudo-Labeling For Massively Multilingual Speech Recognition](https://arxiv.org/abs/2111.00161) by Loren Lugosch, Tatiana Likhomanenko, Gabriel Synnaeve, and Ronan Collobert.
+1. **[M2M100](https://huggingface.co/docs/transformers/model_doc/m2m_100)** (from Facebook) released with the paper [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
+1. **[MarianMT](https://huggingface.co/docs/transformers/model_doc/marian)** Machine translation models trained using [OPUS](http://opus.nlpl.eu/) data by Jörg Tiedemann. The [Marian Framework](https://marian-nmt.github.io/) is being developed by the Microsoft Translator Team.
+1. **[MaskFormer](https://huggingface.co/docs/transformers/model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov.
+1. **[mBART](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
+1. **[mBART-50](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Translation with Extensible Multilingual Pretraining and Finetuning](https://arxiv.org/abs/2008.00401) by Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan.
+1. **[Megatron-BERT](https://huggingface.co/docs/transformers/model_doc/megatron-bert)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
+1. **[Megatron-GPT2](https://huggingface.co/docs/transformers/model_doc/megatron_gpt2)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
+1. **[mLUKE](https://huggingface.co/docs/transformers/model_doc/mluke)** (from Studio Ousia) released with the paper [mLUKE: The Power of Entity Representations in Multilingual Pretrained Language Models](https://arxiv.org/abs/2110.08151) by Ryokan Ri, Ikuya Yamada, and Yoshimasa Tsuruoka.
+1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (from CMU/Google Brain) released with the paper [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) by Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou.
+1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (from Apple) released with the paper [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) by Sachin Mehta and Mohammad Rastegari.
+1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
+1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
+1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
+1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
+1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
+1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
+1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
+1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
+1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
+1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, Peter J. Liu. 
+1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
+1. **[PhoBERT](https://huggingface.co/docs/transformers/model_doc/phobert)** (from VinAI Research) released with the paper [PhoBERT: Pre-trained language models for Vietnamese](https://www.aclweb.org/anthology/2020.findings-emnlp.92/) by Dat Quoc Nguyen and Anh Tuan Nguyen.
+1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (from UCLA NLP) released with the paper [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
+1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (from Sea AI Labs) released with the paper [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) by Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng.
+1. **[ProphetNet](https://huggingface.co/docs/transformers/model_doc/prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
+1. **[QDQBert](https://huggingface.co/docs/transformers/model_doc/qdqbert)** (from NVIDIA) released with the paper [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://arxiv.org/abs/2004.09602) by Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev and Paulius Micikevicius.
+1. **[RAG](https://huggingface.co/docs/transformers/model_doc/rag)** (from Facebook) released with the paper [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) by Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela.
+1. **[REALM](https://huggingface.co/docs/transformers/model_doc/realm.html)** (from Google Research) released with the paper [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) by Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang.
+1. **[Reformer](https://huggingface.co/docs/transformers/model_doc/reformer)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
+1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Research) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár. 
+1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/pdf/2010.12821.pdf) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
+1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun. 
+1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (from Facebook), released together with the paper a [Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
+1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper a [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/pdf/2104.09864v1.pdf) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
+1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
+1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
+1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
+1. **[SpeechToTextTransformer](https://huggingface.co/docs/transformers/model_doc/speech_to_text)** (from Facebook), released together with the paper [fairseq S2T: Fast Speech-to-Text Modeling with fairseq](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Dmytro Okhonko, Juan Pino.
+1. **[SpeechToTextTransformer2](https://huggingface.co/docs/transformers/model_doc/speech_to_text_2)** (from Facebook), released together with the paper [Large-Scale Self- and Semi-Supervised Learning for Speech Translation](https://arxiv.org/abs/2104.06678) by Changhan Wang, Anne Wu, Juan Pino, Alexei Baevski, Michael Auli, Alexis Conneau.
+1. **[Splinter](https://huggingface.co/docs/transformers/model_doc/splinter)** (from Tel Aviv University), released together with the paper [Few-Shot Question Answering by Pretraining Span Selection](https://arxiv.org/abs/2101.00438) by Ori Ram, Yuval Kirstain, Jonathan Berant, Amir Globerson, Omer Levy.
+1. **[SqueezeBERT](https://huggingface.co/docs/transformers/model_doc/squeezebert)** (from Berkeley) released with the paper [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) by Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer.
+1. **[Swin Transformer](https://huggingface.co/docs/transformers/model_doc/swin)** (from Microsoft) released with the paper [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) by Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo.
+1. **[Swin Transformer V2](https://huggingface.co/docs/transformers/model_doc/swinv2)** (from Microsoft) released with the paper [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883) by Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo.
+1. **[T5](https://huggingface.co/docs/transformers/model_doc/t5)** (from Google AI) released with the paper [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://arxiv.org/abs/1910.10683) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
+1. **[T5v1.1](https://huggingface.co/docs/transformers/model_doc/t5v1.1)** (from Google AI) released in the repository [google-research/text-to-text-transfer-transformer](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/released_checkpoints.md#t511) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
+1. **[TAPAS](https://huggingface.co/docs/transformers/model_doc/tapas)** (from Google AI) released with the paper [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
+1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (from Microsoft Research) released with the paper [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) by Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou.
+1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine 
+1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
+1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (from Microsoft), released together with the paper [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) by Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei.
+1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler 
+1. **[UniSpeech](https://huggingface.co/docs/transformers/model_doc/unispeech)** (from Microsoft Research) released with the paper [UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data](https://arxiv.org/abs/2101.07597) by Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei, Michael Zeng, Xuedong Huang.
+1. **[UniSpeechSat](https://huggingface.co/docs/transformers/model_doc/unispeech-sat)** (from Microsoft Research) released with the paper [UNISPEECH-SAT: UNIVERSAL SPEECH REPRESENTATION LEARNING WITH SPEAKER AWARE PRE-TRAINING](https://arxiv.org/abs/2110.05752) by Sanyuan Chen, Yu Wu, Chengyi Wang, Zhengyang Chen, Zhuo Chen, Shujie Liu, Jian Wu, Yao Qian, Furu Wei, Jinyu Li, Xiangzhan Yu.
+1. **[VAN](https://huggingface.co/docs/transformers/model_doc/van)** (from Tsinghua University and Nankai University) released with the paper [Visual Attention Network](https://arxiv.org/pdf/2202.09741.pdf) by Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu.
+1. **[VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)** (from Multimedia Computing Group, Nanjing University) released with the paper [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://arxiv.org/abs/2203.12602) by Zhan Tong, Yibing Song, Jue Wang, Limin Wang.
+1. **[ViLT](https://huggingface.co/docs/transformers/model_doc/vilt)** (from NAVER AI Lab/Kakao Enterprise/Kakao Brain) released with the paper [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://arxiv.org/abs/2102.03334) by Wonjae Kim, Bokyung Son, Ildoo Kim.
+1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby.
+1. **[VisualBERT](https://huggingface.co/docs/transformers/model_doc/visual_bert)** (from UCLA NLP) released with the paper [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) by Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
+1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
+1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
+1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
+1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (from Facebook AI) released with the paper [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) by Qiantong Xu, Alexei Baevski, Michael Auli.
+1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
+1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
+1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li. 
+1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
+1. **[XLM-ProphetNet](https://huggingface.co/docs/transformers/model_doc/xlm-prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
+1. **[XLM-RoBERTa](https://huggingface.co/docs/transformers/model_doc/xlm-roberta)** (from Facebook AI), released together with the paper [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.
+1. **[XLM-RoBERTa-XL](https://huggingface.co/docs/transformers/model_doc/xlm-roberta-xl)** (from Facebook AI) released with the paper [Larger-Scale Transformers for Multilingual Masked Language Modeling](https://arxiv.org/abs/2105.00572) by Naman Goyal, Jingfei Du, Myle Ott, Giri Anantharaman, Alexis Conneau.
+1. **[XLNet](https://huggingface.co/docs/transformers/model_doc/xlnet)** (from Google/CMU) released with the paper [​XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) by Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
+1. **[XLS-R](https://huggingface.co/docs/transformers/model_doc/xls_r)** (from Facebook AI) released with the paper [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) by Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli.
+1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
+1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (from Huazhong University of Science & Technology) released with the paper [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
+1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (from the University of Wisconsin - Madison) released with the paper [You Only Sample (Almost) by Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh.
+1. 새로운 모델을 올리고 싶나요? 우리가 **상세한 가이드와 템플릿** 으로 새로운 모델을 올리도록 도와드릴게요. 가이드와 템플릿은 이 저장소의 [`templates`](./templates) 폴더에서 확인하실 수 있습니다. [컨트리뷰션 가이드라인](./CONTRIBUTING.md)을 꼭 확인해주시고, PR을 올리기 전에 메인테이너에게 연락하거나 이슈를 오픈해 피드백을 받으시길 바랍니다. 
+
+각 모델이 Flax, PyTorch, TensorFlow으로 구현되었는지 또는 🤗 Tokenizers 라이브러리가 지원하는 토크나이저를 사용하는지 확인하려면, [이 표](https://huggingface.co/docs/transformers/index#supported-frameworks)를 확인하세요.
+
+이 구현은 여러 데이터로 검증되었고 (예시 스크립트를 참고하세요) 오리지널 구현의 성능과 같아야 합니다. [도큐먼트](https://huggingface.co/docs/transformers/examples)의 Examples 섹션에서 성능에 대한 자세한 설명을 확인할 수 있습니다.
+
+## 더 알아보기
+
+| 섹션 | 설명 |
+|-|-|
+| [도큐먼트](https://huggingface.co/transformers/) | 전체 API 도큐먼트와 튜토리얼 |
+| [과제 요약](https://huggingface.co/docs/transformers/task_summary) | 🤗 Transformers가 지원하는 과제들 |
+| [전처리 튜토리얼](https://huggingface.co/docs/transformers/preprocessing) | `Tokenizer` 클래스를 이용해 모델을 위한 데이터 준비하기 |
+| [학습과 fine-tuning](https://huggingface.co/docs/transformers/training) | 🤗 Transformers가 제공하는 모델 PyTorch/TensorFlow 학습 과정과 `Trainer` API에서 사용하기 |
+| [퀵 투어: Fine-tuning/사용 스크립트](https://github.com/huggingface/transformers/tree/main/examples) | 다양한 과제에서 모델 fine-tuning하는 예시 스크립트 |
+| [모델 공유 및 업로드](https://huggingface.co/docs/transformers/model_sharing) | 커뮤니티에 fine-tune된 모델을 업로드 및 공유하기 |
+| [마이그레이션](https://huggingface.co/docs/transformers/migration) | `pytorch-transformers`나 `pytorch-pretrained-bert`에서 🤗 Transformers로 이동하기|
+
+## 인용
+
+🤗 Transformers 라이브러리를 인용하고 싶다면, 이 [논문](https://www.aclweb.org/anthology/2020.emnlp-demos.6/)을 인용해 주세요:
+```bibtex
+@inproceedings{wolf-etal-2020-transformers,
+    title = "Transformers: State-of-the-Art Natural Language Processing",
+    author = "Thomas Wolf and Lysandre Debut and Victor Sanh and Julien Chaumond and Clement Delangue and Anthony Moi and Pierric Cistac and Tim Rault and Rémi Louf and Morgan Funtowicz and Joe Davison and Sam Shleifer and Patrick von Platen and Clara Ma and Yacine Jernite and Julien Plu and Canwen Xu and Teven Le Scao and Sylvain Gugger and Mariama Drame and Quentin Lhoest and Alexander M. Rush",
+    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",
+    month = oct,
+    year = "2020",
+    address = "Online",
+    publisher = "Association for Computational Linguistics",
+    url = "https://www.aclweb.org/anthology/2020.emnlp-demos.6",
+    pages = "38--45"
+}
+```

README_zh-hans.md

@@ -0,0 +1,408 @@
+<!---
+Copyright 2020 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+-->
+
+<!---
+A useful guide for English-Chinese translation of Hugging Face documentation
+- Add space around English words and numbers when they appear between Chinese characters. E.g., 共 100 多种语言; 使用 transformers 库。
+- Use square quotes, e.g.,「引用」
+
+Dictionary
+
+Hugging Face: 抱抱脸
+token: 词符（并用括号标注原英文）
+tokenize: 词符化（并用括号标注原英文）
+tokenizer: 词符化器（并用括号标注原英文）
+transformer: transformer（不翻译）
+pipeline: 流水线
+API: API (不翻译）
+inference: 推理
+Trainer: 训练器。当作为类名出现时不翻译。
+pretrained/pretrain: 预训练
+finetune: 微调
+community: 社区
+example: 当特指仓库中 example 目录时翻译为「用例」
+Python data structures (e.g., list, set, dict): 翻译为列表，集合，词典，并用括号标注原英文
+NLP/Natural Language Processing: 以 NLP 出现时不翻译，以 Natural Language Processing 出现时翻译为自然语言处理
+checkpoint: 检查点
+-->
+
+<p align="center">
+    <br>
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_logo_name.png" width="400"/>
+    <br>
+<p>
+<p align="center">
+    <a href="https://circleci.com/gh/huggingface/transformers">
+        <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
+    </a>
+    <a href="https://github.com/huggingface/transformers/blob/main/LICENSE">
+        <img alt="GitHub" src="https://img.shields.io/github/license/huggingface/transformers.svg?color=blue">
+    </a>
+    <a href="https://huggingface.co/docs/transformers/index">
+        <img alt="Documentation" src="https://img.shields.io/website/http/huggingface.co/docs/transformers/index.svg?down_color=red&down_message=offline&up_message=online">
+    </a>
+    <a href="https://github.com/huggingface/transformers/releases">
+        <img alt="GitHub release" src="https://img.shields.io/github/release/huggingface/transformers.svg">
+    </a>
+    <a href="https://github.com/huggingface/transformers/blob/main/CODE_OF_CONDUCT.md">
+        <img alt="Contributor Covenant" src="https://img.shields.io/badge/Contributor%20Covenant-v2.0%20adopted-ff69b4.svg">
+    </a>
+    <a href="https://zenodo.org/badge/latestdoi/155220641"><img src="https://zenodo.org/badge/155220641.svg" alt="DOI"></a>
+</p>
+
+<h4 align="center">
+    <p>
+        <a href="https://github.com/huggingface/transformers/">English</a> |
+        <b>简体中文</b> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hant.md">繁體中文</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_ko.md">한국어</a>
+    <p>
+</h4>
+
+<h3 align="center">
+    <p>为 Jax、PyTorch 和 TensorFlow 打造的先进的自然语言处理</p>
+</h3>
+
+<h3 align="center">
+    <a href="https://hf.co/course"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/course_banner.png"></a>
+</h3>
+
+🤗 Transformers 提供了数以千计的预训练模型，支持 100 多种语言的文本分类、信息抽取、问答、摘要、翻译、文本生成。它的宗旨让最先进的 NLP 技术人人易用。
+
+🤗 Transformers 提供了便于快速下载和使用的API，让你可以把预训练模型用在给定文本、在你的数据集上微调然后通过 [model hub](https://huggingface.co/models) 与社区共享。同时，每个定义的 Python 模块均完全独立，方便修改和快速研究实验。
+
+🤗 Transformers 支持三个最热门的深度学习库： [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/) and [TensorFlow](https://www.tensorflow.org/) — 并与之无缝整合。你可以直接使用一个框架训练你的模型然后用另一个加载和推理。
+
+## 在线演示
+
+你可以直接在模型页面上测试大多数 [model hub](https://huggingface.co/models) 上的模型。 我们也提供了 [私有模型托管、模型版本管理以及推理API](https://huggingface.co/pricing)。
+
+这里是一些例子：
+- [用 BERT 做掩码填词](https://huggingface.co/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
+- [用 Electra 做命名实体识别](https://huggingface.co/dbmdz/electra-large-discriminator-finetuned-conll03-english?text=My+name+is+Sarah+and+I+live+in+London+city)
+- [用 GPT-2 做文本生成](https://huggingface.co/gpt2?text=A+long+time+ago%2C+)
+- [用 RoBERTa 做自然语言推理](https://huggingface.co/roberta-large-mnli?text=The+dog+was+lost.+Nobody+lost+any+animal)
+- [用 BART 做文本摘要](https://huggingface.co/facebook/bart-large-cnn?text=The+tower+is+324+metres+%281%2C063+ft%29+tall%2C+about+the+same+height+as+an+81-storey+building%2C+and+the+tallest+structure+in+Paris.+Its+base+is+square%2C+measuring+125+metres+%28410+ft%29+on+each+side.+During+its+construction%2C+the+Eiffel+Tower+surpassed+the+Washington+Monument+to+become+the+tallest+man-made+structure+in+the+world%2C+a+title+it+held+for+41+years+until+the+Chrysler+Building+in+New+York+City+was+finished+in+1930.+It+was+the+first+structure+to+reach+a+height+of+300+metres.+Due+to+the+addition+of+a+broadcasting+aerial+at+the+top+of+the+tower+in+1957%2C+it+is+now+taller+than+the+Chrysler+Building+by+5.2+metres+%2817+ft%29.+Excluding+transmitters%2C+the+Eiffel+Tower+is+the+second+tallest+free-standing+structure+in+France+after+the+Millau+Viaduct)
+- [用 DistilBERT 做问答](https://huggingface.co/distilbert-base-uncased-distilled-squad?text=Which+name+is+also+used+to+describe+the+Amazon+rainforest+in+English%3F&context=The+Amazon+rainforest+%28Portuguese%3A+Floresta+Amaz%C3%B4nica+or+Amaz%C3%B4nia%3B+Spanish%3A+Selva+Amaz%C3%B3nica%2C+Amazon%C3%ADa+or+usually+Amazonia%3B+French%3A+For%C3%AAt+amazonienne%3B+Dutch%3A+Amazoneregenwoud%29%2C+also+known+in+English+as+Amazonia+or+the+Amazon+Jungle%2C+is+a+moist+broadleaf+forest+that+covers+most+of+the+Amazon+basin+of+South+America.+This+basin+encompasses+7%2C000%2C000+square+kilometres+%282%2C700%2C000+sq+mi%29%2C+of+which+5%2C500%2C000+square+kilometres+%282%2C100%2C000+sq+mi%29+are+covered+by+the+rainforest.+This+region+includes+territory+belonging+to+nine+nations.+The+majority+of+the+forest+is+contained+within+Brazil%2C+with+60%25+of+the+rainforest%2C+followed+by+Peru+with+13%25%2C+Colombia+with+10%25%2C+and+with+minor+amounts+in+Venezuela%2C+Ecuador%2C+Bolivia%2C+Guyana%2C+Suriname+and+French+Guiana.+States+or+departments+in+four+nations+contain+%22Amazonas%22+in+their+names.+The+Amazon+represents+over+half+of+the+planet%27s+remaining+rainforests%2C+and+comprises+the+largest+and+most+biodiverse+tract+of+tropical+rainforest+in+the+world%2C+with+an+estimated+390+billion+individual+trees+divided+into+16%2C000+species)
+- [用 T5 做翻译](https://huggingface.co/t5-base?text=My+name+is+Wolfgang+and+I+live+in+Berlin)
+
+**[Write With Transformer](https://transformer.huggingface.co)**，由抱抱脸团队打造，是一个文本生成的官方 demo。
+
+## 如果你在寻找由抱抱脸团队提供的定制化支持服务
+
+<a target="_blank" href="https://huggingface.co/support">
+    <img alt="HuggingFace Expert Acceleration Program" src="https://huggingface.co/front/thumbnails/support.png" style="max-width: 600px; border: 1px solid #eee; border-radius: 4px; box-shadow: 0 1px 2px 0 rgba(0, 0, 0, 0.05);">
+</a><br>
+
+## 快速上手
+
+我们为快速使用模型提供了 `pipeline` （流水线）API。流水线聚合了预训练模型和对应的文本预处理。下面是一个快速使用流水线去判断正负面情绪的例子：
+
+```python
+>>> from transformers import pipeline
+
+# 使用情绪分析流水线
+>>> classifier = pipeline('sentiment-analysis')
+>>> classifier('We are very happy to introduce pipeline to the transformers repository.')
+[{'label': 'POSITIVE', 'score': 0.9996980428695679}]
+```
+
+第二行代码下载并缓存了流水线使用的预训练模型，而第三行代码则在给定的文本上进行了评估。这里的答案“正面” (positive) 具有 99 的置信度。
+
+许多的 NLP 任务都有开箱即用的预训练流水线。比如说，我们可以轻松的从给定文本中抽取问题答案：
+
+``` python
+>>> from transformers import pipeline
+
+# 使用问答流水线
+>>> question_answerer = pipeline('question-answering')
+>>> question_answerer({
+...     'question': 'What is the name of the repository ?',
+...     'context': 'Pipeline has been included in the huggingface/transformers repository'
+... })
+{'score': 0.30970096588134766, 'start': 34, 'end': 58, 'answer': 'huggingface/transformers'}
+
+```
+
+除了给出答案，预训练模型还给出了对应的置信度分数、答案在词符化 (tokenized) 后的文本中开始和结束的位置。你可以从[这个教程](https://huggingface.co/docs/transformers/task_summary)了解更多流水线API支持的任务。
+
+要在你的任务上下载和使用任意预训练模型也很简单，只需三行代码。这里是 PyTorch 版的示例：
+```python
+>>> from transformers import AutoTokenizer, AutoModel
+
+>>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
+>>> model = AutoModel.from_pretrained("bert-base-uncased")
+
+>>> inputs = tokenizer("Hello world!", return_tensors="pt")
+>>> outputs = model(**inputs)
+```
+这里是等效的 TensorFlow 代码：
+```python
+>>> from transformers import AutoTokenizer, TFAutoModel
+
+>>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
+>>> model = TFAutoModel.from_pretrained("bert-base-uncased")
+
+>>> inputs = tokenizer("Hello world!", return_tensors="tf")
+>>> outputs = model(**inputs)
+```
+
+词符化器 (tokenizer) 为所有的预训练模型提供了预处理，并可以直接对单个字符串进行调用（比如上面的例子）或对列表 (list) 调用。它会输出一个你可以在下游代码里使用或直接通过 `**` 解包表达式传给模型的词典 (dict)。
+
+模型本身是一个常规的 [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) 或 [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model)（取决于你的后端），可以常规方式使用。 [这个教程](https://huggingface.co/transformers/training.html)解释了如何将这样的模型整合到经典的 PyTorch 或 TensorFlow 训练循环中，或是如何使用我们的 `Trainer` 训练器）API 来在一个新的数据集上快速微调。
+
+## 为什么要用 transformers？
+
+1. 便于使用的先进模型：
+    - NLU 和 NLG 上表现优越
+    - 对教学和实践友好且低门槛
+    - 高级抽象，只需了解三个类
+    - 对所有模型统一的API
+
+1. 更低计算开销，更少的碳排放：
+    - 研究人员可以分享已训练的模型而非每次从头开始训练
+    - 工程师可以减少计算用时和生产环境开销
+    - 数十种模型架构、两千多个预训练模型、100多种语言支持
+
+1. 对于模型生命周期的每一个部分都面面俱到：
+    - 训练先进的模型，只需 3 行代码
+    - 模型在不同深度学习框架间任意转移，随你心意
+    - 为训练、评估和生产选择最适合的框架，衔接无缝
+
+1. 为你的需求轻松定制专属模型和用例：
+    - 我们为每种模型架构提供了多个用例来复现原论文结果
+    - 模型内部结构保持透明一致
+    - 模型文件可单独使用，方便魔改和快速实验
+
+## 什么情况下我不该用 transformers？
+
+- 本库并不是模块化的神经网络工具箱。模型文件中的代码特意呈若璞玉，未经额外抽象封装，以便研究人员快速迭代魔改而不致溺于抽象和文件跳转之中。
+- `Trainer` API 并非兼容任何模型，只为本库之模型优化。若是在寻找适用于通用机器学习的训练循环实现，请另觅他库。
+- 尽管我们已尽力而为，[examples 目录](https://github.com/huggingface/transformers/tree/main/examples)中的脚本也仅为用例而已。对于你的特定问题，它们并不一定开箱即用，可能需要改几行代码以适之。
+
+## 安装
+
+### 使用 pip
+
+这个仓库已在 Python 3.6+、Flax 0.3.2+、PyTorch 1.3.1+ 和 TensorFlow 2.3+ 下经过测试。
+
+你可以在[虚拟环境](https://docs.python.org/3/library/venv.html)中安装 🤗 Transformers。如果你还不熟悉 Python 的虚拟环境，请阅此[用户说明](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/)。
+
+首先，用你打算使用的版本的 Python 创建一个虚拟环境并激活。
+
+然后，你需要安装 Flax、PyTorch 或 TensorFlow 其中之一。关于在你使用的平台上安装这些框架，请参阅 [TensorFlow 安装页](https://www.tensorflow.org/install/), [PyTorch 安装页](https://pytorch.org/get-started/locally/#start-locally) 或 [Flax 安装页](https://github.com/google/flax#quick-install)。
+
+当这些后端之一安装成功后， 🤗 Transformers 可依此安装：
+
+```bash
+pip install transformers
+```
+
+如果你想要试试用例或者想在正式发布前使用最新的开发中代码，你得[从源代码安装](https://huggingface.co/docs/transformers/installation#installing-from-source)。
+
+### 使用 conda
+
+自 Transformers 4.0.0 版始，我们有了一个 conda 频道： `huggingface`。
+
+🤗 Transformers 可以通过 conda 依此安装：
+
+```shell script
+conda install -c huggingface transformers
+```
+
+要通过 conda 安装 Flax、PyTorch 或 TensorFlow 其中之一，请参阅它们各自安装页的说明。
+
+## 模型架构
+
+🤗 Transformers 支持的[**所有的模型检查点**](https://huggingface.co/models)由[用户](https://huggingface.co/users)和[组织](https://huggingface.co/organizations)上传，均与 huggingface.co [model hub](https://huggingface.co) 无缝整合。
+
+目前的检查点数量： ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
+
+🤗 Transformers 目前支持如下的架构（模型概述请阅[这里](https://huggingface.co/docs/transformers/model_summary)）：
+
+1. **[ALBERT](https://huggingface.co/docs/transformers/model_doc/albert)** (来自 Google Research and the Toyota Technological Institute at Chicago) 伴随论文 [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://arxiv.org/abs/1909.11942), 由 Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut 发布。
+1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (来自 Facebook) 伴随论文 [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/pdf/1910.13461.pdf) 由 Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer 发布。
+1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (来自 École polytechnique) 伴随论文 [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) 由 Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis 发布。
+1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (来自 VinAI Research) 伴随论文 [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701) 由 Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen 发布。
+1. **[BEiT](https://huggingface.co/docs/transformers/model_doc/beit)** (来自 Microsoft) 伴随论文 [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) 由 Hangbo Bao, Li Dong, Furu Wei 发布。
+1. **[BERT](https://huggingface.co/docs/transformers/model_doc/bert)** (来自 Google) 伴随论文 [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) 由 Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova 发布。
+1. **[BERT For Sequence Generation](https://huggingface.co/docs/transformers/model_doc/bert-generation)** (来自 Google) 伴随论文 [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) 由 Sascha Rothe, Shashi Narayan, Aliaksei Severyn 发布。
+1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (来自 VinAI Research) 伴随论文 [BERTweet: A pre-trained language model for English Tweets](https://aclanthology.org/2020.emnlp-demos.2/) 由 Dat Quoc Nguyen, Thanh Vu and Anh Tuan Nguyen 发布。
+1. **[BigBird-Pegasus](https://huggingface.co/docs/transformers/model_doc/bigbird_pegasus)** (来自 Google Research) 伴随论文 [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) 由 Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed 发布。
+1. **[BigBird-RoBERTa](https://huggingface.co/docs/transformers/model_doc/big_bird)** (来自 Google Research) 伴随论文 [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) 由 Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed 发布。
+1. **[Blenderbot](https://huggingface.co/docs/transformers/model_doc/blenderbot)** (来自 Facebook) 伴随论文 [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) 由 Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston 发布。
+1. **[BlenderbotSmall](https://huggingface.co/docs/transformers/model_doc/blenderbot-small)** (来自 Facebook) 伴随论文 [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) 由 Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston 发布。
+1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigSicence Workshop](https://bigscience.huggingface.co/). 
+1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (来自 Alexa) 伴随论文 [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) 由 Adrian de Wynter and Daniel J. Perry 发布。
+1. **[ByT5](https://huggingface.co/docs/transformers/model_doc/byt5)** (来自 Google Research) 伴随论文 [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) 由 Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel 发布。
+1. **[CamemBERT](https://huggingface.co/docs/transformers/model_doc/camembert)** (来自 Inria/Facebook/Sorbonne) 伴随论文 [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) 由 Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot 发布。
+1. **[CANINE](https://huggingface.co/docs/transformers/model_doc/canine)** (来自 Google Research) 伴随论文 [CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation](https://arxiv.org/abs/2103.06874) 由 Jonathan H. Clark, Dan Garrette, Iulia Turc, John Wieting 发布。
+1. **[CLIP](https://huggingface.co/docs/transformers/model_doc/clip)** (来自 OpenAI) 伴随论文 [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) 由 Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever 发布。
+1. **[CodeGen](https://huggingface.co/docs/transformers/model_doc/codegen)** (来自 Salesforce) 伴随论文 [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) 由 Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong 发布。
+1. **[ConvBERT](https://huggingface.co/docs/transformers/model_doc/convbert)** (来自 YituTech) 伴随论文 [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) 由 Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan 发布。
+1. **[ConvNeXT](https://huggingface.co/docs/transformers/model_doc/convnext)** (来自 Facebook AI) 伴随论文 [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) 由 Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, Saining Xie 发布。
+1. **[CPM](https://huggingface.co/docs/transformers/model_doc/cpm)** (来自 Tsinghua University) 伴随论文 [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) 由 Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun 发布。
+1. **[CTRL](https://huggingface.co/docs/transformers/model_doc/ctrl)** (来自 Salesforce) 伴随论文 [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) 由 Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher 发布。
+1. **[CvT](https://huggingface.co/docs/transformers/model_doc/cvt)** (来自 Microsoft) 伴随论文 [CvT: Introducing Convolutions to Vision Transformers](https://arxiv.org/abs/2103.15808) 由 Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, Lei Zhang 发布。
+1. **[Data2Vec](https://huggingface.co/docs/transformers/model_doc/data2vec)** (来自 Facebook) 伴随论文 [Data2Vec:  A General Framework for Self-supervised Learning in Speech, Vision and Language](https://arxiv.org/abs/2202.03555) 由 Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu, Michael Auli 发布。
+1. **[DeBERTa](https://huggingface.co/docs/transformers/model_doc/deberta)** (来自 Microsoft) 伴随论文 [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) 由 Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen 发布。
+1. **[DeBERTa-v2](https://huggingface.co/docs/transformers/model_doc/deberta-v2)** (来自 Microsoft) 伴随论文 [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) 由 Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen 发布。
+1. **[Decision Transformer](https://huggingface.co/docs/transformers/model_doc/decision_transformer)** (来自 Berkeley/Facebook/Google) 伴随论文 [Decision Transformer: Reinforcement Learning via Sequence Modeling](https://arxiv.org/abs/2106.01345) 由 Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Michael Laskin, Pieter Abbeel, Aravind Srinivas, Igor Mordatch 发布。
+1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (来自 Facebook) 伴随论文 [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) 由 Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou 发布。
+1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (来自 Facebook) 伴随论文 [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) 由 Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko 发布。
+1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (来自 Microsoft Research) 伴随论文 [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) 由 Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan 发布。
+1. **[DistilBERT](https://huggingface.co/docs/transformers/model_doc/distilbert)** (来自 HuggingFace), 伴随论文 [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) 由 Victor Sanh, Lysandre Debut and Thomas Wolf 发布。 同样的方法也应用于压缩 GPT-2 到 [DistilGPT2](https://github.com/huggingface/transformers/tree/main/examples/distillation), RoBERTa 到 [DistilRoBERTa](https://github.com/huggingface/transformers/tree/main/examples/distillation), Multilingual BERT 到 [DistilmBERT](https://github.com/huggingface/transformers/tree/main/examples/distillation) 和德语版 DistilBERT。
+1. **[DiT](https://huggingface.co/docs/transformers/model_doc/dit)** (来自 Microsoft Research) 伴随论文 [DiT: Self-supervised Pre-training for Document Image Transformer](https://arxiv.org/abs/2203.02378) 由 Junlong Li, Yiheng Xu, Tengchao Lv, Lei Cui, Cha Zhang, Furu Wei 发布。
+1. **[Donut](https://huggingface.co/docs/transformers/model_doc/donut)** (来自 NAVER) 伴随论文 [OCR-free Document Understanding Transformer](https://arxiv.org/abs/2111.15664) 由 Geewook Kim, Teakgyu Hong, Moonbin Yim, Jeongyeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, Seunghyun Park 发布。
+1. **[DPR](https://huggingface.co/docs/transformers/model_doc/dpr)** (来自 Facebook) 伴随论文 [Dense Passage Retrieval for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) 由 Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih 发布。
+1. **[DPT](https://huggingface.co/docs/transformers/master/model_doc/dpt)** (来自 Intel Labs) 伴随论文 [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) 由 René Ranftl, Alexey Bochkovskiy, Vladlen Koltun 发布。
+1. **[ELECTRA](https://huggingface.co/docs/transformers/model_doc/electra)** (来自 Google Research/Stanford University) 伴随论文 [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) 由 Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning 发布。
+1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (来自 Google Research) 伴随论文 [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) 由 Sascha Rothe, Shashi Narayan, Aliaksei Severyn 发布。
+1. **[ERNIE](https://huggingface.co/docs/transformers/model_doc/ernie)** (来自 Baidu) 伴随论文 [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu 发布。
+1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (来自 CNRS) 伴随论文 [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) 由 Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab 发布。
+1. **[FLAVA](https://huggingface.co/docs/transformers/model_doc/flava)** (来自 Facebook AI) 伴随论文 [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) 由 Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela 发布。
+1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (来自 Google Research) 伴随论文 [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) 由 James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon 发布。
+1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (来自 CMU/Google Brain) 伴随论文 [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) 由 Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le 发布。
+1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (来自 KAIST) 伴随论文 [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) 由 Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim 发布。
+1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (来自 OpenAI) 伴随论文 [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) 由 Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever 发布。
+1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (来自 EleutherAI) 随仓库 [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) 发布。作者为 Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy 发布。
+1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach 
+1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (来自 OpenAI) 伴随论文 [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) 由 Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever** 发布。
+1. **[GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj)** (来自 EleutherAI) 伴随论文 [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) 由 Ben Wang and Aran Komatsuzaki 发布。
+1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (来自 UCSD, NVIDIA) 伴随论文 [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) 由 Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang 发布。
+1. **[Hubert](https://huggingface.co/docs/transformers/model_doc/hubert)** (来自 Facebook) 伴随论文 [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://arxiv.org/abs/2106.07447) 由 Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed 发布。
+1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (来自 Berkeley) 伴随论文 [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) 由 Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer 发布。
+1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (来自 OpenAI) 伴随论文 [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) 由 Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever 发布。
+1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (来自 Microsoft Research Asia) 伴随论文 [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) 由 Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou 发布。
+1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (来自 Microsoft Research Asia) 伴随论文 [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) 由 Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou 发布。
+1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (来自 Microsoft Research Asia) 伴随论文 [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) 由 Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei 发布。
+1. **[LayoutXLM](https://huggingface.co/docs/transformers/model_doc/layoutxlm)** (来自 Microsoft Research Asia) 伴随论文 [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) 由 Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei 发布。
+1. **[LED](https://huggingface.co/docs/transformers/model_doc/led)** (来自 AllenAI) 伴随论文 [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) 由 Iz Beltagy, Matthew E. Peters, Arman Cohan 发布。
+1. **[LeViT](https://huggingface.co/docs/transformers/model_doc/levit)** (来自 Meta AI) 伴随论文 [LeViT: A Vision Transformer in ConvNet's Clothing for Faster Inference](https://arxiv.org/abs/2104.01136) 由 Ben Graham, Alaaeldin El-Nouby, Hugo Touvron, Pierre Stock, Armand Joulin, Hervé Jégou, Matthijs Douze 发布。
+1. **[Longformer](https://huggingface.co/docs/transformers/model_doc/longformer)** (来自 AllenAI) 伴随论文 [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) 由 Iz Beltagy, Matthew E. Peters, Arman Cohan 发布。
+1. **[LongT5](https://huggingface.co/docs/transformers/model_doc/longt5)** (来自 Google AI) released 伴随论文 [LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/abs/2112.07916) 由 Mandy Guo, Joshua Ainslie, David Uthus, Santiago Ontanon, Jianmo Ni, Yun-Hsuan Sung, Yinfei Yang 发布。
+1. **[LUKE](https://huggingface.co/docs/transformers/model_doc/luke)** (来自 Studio Ousia) 伴随论文 [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) 由 Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto 发布。
+1. **[LXMERT](https://huggingface.co/docs/transformers/model_doc/lxmert)** (来自 UNC Chapel Hill) 伴随论文 [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) 由 Hao Tan and Mohit Bansal 发布。
+1. **[M-CTC-T](https://huggingface.co/docs/transformers/model_doc/mctct)** (来自 Facebook) 伴随论文 [Pseudo-Labeling For Massively Multilingual Speech Recognition](https://arxiv.org/abs/2111.00161) 由 Loren Lugosch, Tatiana Likhomanenko, Gabriel Synnaeve, and Ronan Collobert 发布。
+1. **[M2M100](https://huggingface.co/docs/transformers/model_doc/m2m_100)** (来自 Facebook) 伴随论文 [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) 由 Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin 发布。
+1. **[MarianMT](https://huggingface.co/docs/transformers/model_doc/marian)** 用 [OPUS](http://opus.nlpl.eu/) 数据训练的机器翻译模型由 Jörg Tiedemann 发布。[Marian Framework](https://marian-nmt.github.io/) 由微软翻译团队开发。
+1. **[MaskFormer](https://huggingface.co/docs/transformers/model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov 
+1. **[mBART](https://huggingface.co/docs/transformers/model_doc/mbart)** (来自 Facebook) 伴随论文 [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) 由 Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer 发布。
+1. **[mBART-50](https://huggingface.co/docs/transformers/model_doc/mbart)** (来自 Facebook) 伴随论文 [Multilingual Translation with Extensible Multilingual Pretraining and Finetuning](https://arxiv.org/abs/2008.00401) 由 Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan 发布。
+1. **[Megatron-BERT](https://huggingface.co/docs/transformers/model_doc/megatron-bert)** (来自 NVIDIA) 伴随论文 [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) 由 Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro 发布。
+1. **[Megatron-GPT2](https://huggingface.co/docs/transformers/model_doc/megatron_gpt2)** (来自 NVIDIA) 伴随论文 [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) 由 Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro 发布。
+1. **[mLUKE](https://huggingface.co/docs/transformers/model_doc/mluke)** (来自 Studio Ousia) 伴随论文 [mLUKE: The Power of Entity Representations in Multilingual Pretrained Language Models](https://arxiv.org/abs/2110.08151) 由 Ryokan Ri, Ikuya Yamada, and Yoshimasa Tsuruoka 发布。
+1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (来自 CMU/Google Brain) 伴随论文 [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) 由 Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou 发布。
+1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (来自 Apple) 伴随论文 [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) 由 Sachin Mehta and Mohammad Rastegari 发布。
+1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (来自 Microsoft Research) 伴随论文 [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) 由 Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu 发布。
+1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (来自 Google AI) 伴随论文 [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) 由 Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel 发布。
+1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (来自 中国人民大学 AI Box) 伴随论文 [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) 由 Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen 发布。 
+1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (来自华为诺亚方舟实验室) 伴随论文 [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) 由 Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu 发布。
+1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (来自 Meta) 伴随论文 [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) 由 the NLLB team 发布。
+1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (来自 the University of Wisconsin - Madison) 伴随论文 [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) 由 Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh 发布。
+1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (来自 Meta AI) 伴随论文 [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) 由 Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al 发布。
+1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (来自 Google AI) 伴随论文 [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) 由 Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby 发布。
+1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (来自 Google) 伴随论文 [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) 由 Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu 发布。
+1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (来自 Google) 伴随论文 [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) 由 Jason Phang, Yao Zhao, Peter J. Liu 发布。  
+1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (来自 Deepmind) 伴随论文 [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) 由 Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira 发布。
+1. **[PhoBERT](https://huggingface.co/docs/transformers/model_doc/phobert)** (来自 VinAI Research) 伴随论文 [PhoBERT: Pre-trained language models for Vietnamese](https://www.aclweb.org/anthology/2020.findings-emnlp.92/) 由 Dat Quoc Nguyen and Anh Tuan Nguyen 发布。
+1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (来自 UCLA NLP) 伴随论文 [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) 由 Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang 发布。
+1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (来自 Sea AI Labs) 伴随论文 [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) 由 Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng 发布。
+1. **[ProphetNet](https://huggingface.co/docs/transformers/model_doc/prophetnet)** (来自 Microsoft Research) 伴随论文 [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) 由 Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou 发布。
+1. **[QDQBert](https://huggingface.co/docs/transformers/model_doc/qdqbert)** (来自 NVIDIA) 伴随论文 [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://arxiv.org/abs/2004.09602) 由 Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev and Paulius Micikevicius 发布。
+1. **[RAG](https://huggingface.co/docs/transformers/model_doc/rag)** (来自 Facebook) 伴随论文 [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) 由 Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela 发布。
+1. **[REALM](https://huggingface.co/docs/transformers/model_doc/realm.html)** (来自 Google Research) 伴随论文 [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) 由 Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang 发布。
+1. **[Reformer](https://huggingface.co/docs/transformers/model_doc/reformer)** (来自 Google Research) 伴随论文 [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) 由 Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya 发布。
+1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Research) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár. 
+1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (来自 Google Research) 伴随论文 [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/pdf/2010.12821.pdf) 由 Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder 发布。
+1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun. 
+1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (来自 Facebook), 伴随论文 [Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) 由 Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov 发布。
+1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (来自 ZhuiyiTechnology), 伴随论文 [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/pdf/2104.09864v1.pdf) 由 Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu 发布。
+1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (来自 NVIDIA) 伴随论文 [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) 由 Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo 发布。
+1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (来自 ASAPP) 伴随论文 [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) 由 Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi 发布。
+1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (来自 ASAPP) 伴随论文 [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) 由 Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi 发布。
+1. **[SpeechToTextTransformer](https://huggingface.co/docs/transformers/model_doc/speech_to_text)** (来自 Facebook), 伴随论文 [fairseq S2T: Fast Speech-to-Text Modeling with fairseq](https://arxiv.org/abs/2010.05171) 由 Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Dmytro Okhonko, Juan Pino 发布。
+1. **[SpeechToTextTransformer2](https://huggingface.co/docs/transformers/model_doc/speech_to_text_2)** (来自 Facebook) 伴随论文 [Large-Scale Self- and Semi-Supervised Learning for Speech Translation](https://arxiv.org/abs/2104.06678) 由 Changhan Wang, Anne Wu, Juan Pino, Alexei Baevski, Michael Auli, Alexis Conneau 发布。
+1. **[Splinter](https://huggingface.co/docs/transformers/model_doc/splinter)** (来自 Tel Aviv University) 伴随论文 [Few-Shot Question Answering by Pretraining Span Selection](https://arxiv.org/abs/2101.00438) 由 Ori Ram, Yuval Kirstain, Jonathan Berant, Amir Globerson, Omer Levy 发布。
+1. **[SqueezeBERT](https://huggingface.co/docs/transformers/model_doc/squeezebert)** (来自 Berkeley) 伴随论文 [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) 由 Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer 发布。
+1. **[Swin Transformer](https://huggingface.co/docs/transformers/model_doc/swin)** (来自 Microsoft) 伴随论文 [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) 由 Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo 发布。
+1. **[Swin Transformer V2](https://huggingface.co/docs/transformers/model_doc/swinv2)** (来自 Microsoft) 伴随论文 [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883) 由 Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo 发布。
+1. **[T5](https://huggingface.co/docs/transformers/model_doc/t5)** (来自 Google AI) 伴随论文 [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://arxiv.org/abs/1910.10683) 由 Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu 发布。
+1. **[T5v1.1](https://huggingface.co/docs/transformers/model_doc/t5v1.1)** (来自 Google AI) 伴随论文 [google-research/text-to-text-transfer-transformer](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/released_checkpoints.md#t511) 由 Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu 发布。
+1. **[TAPAS](https://huggingface.co/docs/transformers/model_doc/tapas)** (来自 Google AI) 伴随论文 [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) 由 Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos 发布。
+1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (来自 Microsoft Research) 伴随论文 [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) 由 Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou 发布。
+1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine 
+1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (来自 Google/CMU) 伴随论文 [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) 由 Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov 发布。
+1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (来自 Microsoft) 伴随论文 [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) 由 Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei 发布。
+1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler 
+1. **[UniSpeech](https://huggingface.co/docs/transformers/model_doc/unispeech)** (来自 Microsoft Research) 伴随论文 [UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data](https://arxiv.org/abs/2101.07597) 由 Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei, Michael Zeng, Xuedong Huang 发布。
+1. **[UniSpeechSat](https://huggingface.co/docs/transformers/model_doc/unispeech-sat)** (来自 Microsoft Research) 伴随论文 [UNISPEECH-SAT: UNIVERSAL SPEECH REPRESENTATION LEARNING WITH SPEAKER AWARE PRE-TRAINING](https://arxiv.org/abs/2110.05752) 由 Sanyuan Chen, Yu Wu, Chengyi Wang, Zhengyang Chen, Zhuo Chen, Shujie Liu, Jian Wu, Yao Qian, Furu Wei, Jinyu Li, Xiangzhan Yu 发布。
+1. **[VAN](https://huggingface.co/docs/transformers/model_doc/van)** (来自 Tsinghua University and Nankai University) 伴随论文 [Visual Attention Network](https://arxiv.org/pdf/2202.09741.pdf) 由 Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu 发布。
+1. **[VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)** (来自 Multimedia Computing Group, Nanjing University) 伴随论文 [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://arxiv.org/abs/2203.12602) 由 Zhan Tong, Yibing Song, Jue Wang, Limin Wang 发布。
+1. **[ViLT](https://huggingface.co/docs/transformers/model_doc/vilt)** (来自 NAVER AI Lab/Kakao Enterprise/Kakao Brain) 伴随论文 [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://arxiv.org/abs/2102.03334) 由 Wonjae Kim, Bokyung Son, Ildoo Kim 发布。
+1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (来自 Google AI) 伴随论文 [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) 由 Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby 发布。
+1. **[VisualBERT](https://huggingface.co/docs/transformers/model_doc/visual_bert)** (来自 UCLA NLP) 伴随论文 [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) 由 Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang 发布。
+1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (来自 Meta AI) 伴随论文 [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) 由 Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick 发布。
+1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (来自 Facebook AI) 伴随论文 [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) 由 Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli 发布。
+1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (来自 Facebook AI) 伴随论文 [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) 由 Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino 发布。
+1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (来自 Facebook AI) 伴随论文 [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) 由 Qiantong Xu, Alexei Baevski, Michael Auli 发布。
+1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
+1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (来自 Microsoft Research) 伴随论文 [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) 由 Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling 发布。
+1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li. 
+1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (来自 Facebook) 伴随论文 [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) 由 Guillaume Lample and Alexis Conneau 发布。
+1. **[XLM-ProphetNet](https://huggingface.co/docs/transformers/model_doc/xlm-prophetnet)** (来自 Microsoft Research) 伴随论文 [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) 由 Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou 发布。
+1. **[XLM-RoBERTa](https://huggingface.co/docs/transformers/model_doc/xlm-roberta)** (来自 Facebook AI), 伴随论文 [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) 由 Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov 发布。
+1. **[XLM-RoBERTa-XL](https://huggingface.co/docs/transformers/model_doc/xlm-roberta-xl)** (来自 Facebook AI) 伴随论文 [Larger-Scale Transformers for Multilingual Masked Language Modeling](https://arxiv.org/abs/2105.00572) 由 Naman Goyal, Jingfei Du, Myle Ott, Giri Anantharaman, Alexis Conneau 发布。
+1. **[XLNet](https://huggingface.co/docs/transformers/model_doc/xlnet)** (来自 Google/CMU) 伴随论文 [XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) 由 Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le 发布。
+1. **[XLS-R](https://huggingface.co/docs/transformers/model_doc/xls_r)** (来自 Facebook AI) 伴随论文 [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) 由 Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli 发布。
+1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (来自 Facebook AI) 伴随论文 [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) 由 Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli 发布。
+1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (来自 Huazhong University of Science & Technology) 伴随论文 [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) 由 Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu 发布。
+1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (来自 the University of Wisconsin - Madison) 伴随论文 [You Only Sample (Almost) 由 Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh 发布。
+1. 想要贡献新的模型？我们这里有一份**详细指引和模板**来引导你添加新的模型。你可以在 [`templates`](./templates) 目录中找到他们。记得查看 [贡献指南](./CONTRIBUTING.md) 并在开始写 PR 前联系维护人员或开一个新的 issue 来获得反馈。
+
+要检查某个模型是否已有 Flax、PyTorch 或 TensorFlow 的实现，或其是否在 🤗 Tokenizers 库中有对应词符化器（tokenizer），敬请参阅[此表](https://huggingface.co/docs/transformers/index#supported-frameworks)。
+
+这些实现均已于多个数据集测试（请参看用例脚本）并应于原版实现表现相当。你可以在用例文档的[此节](https://huggingface.co/docs/transformers/examples)中了解表现的细节。
+
+
+## 了解更多
+
+| 章节 | 描述 |
+|-|-|
+| [文档](https://huggingface.co/transformers/) | 完整的 API 文档和教程 |
+| [任务总结](https://huggingface.co/docs/transformers/task_summary) | 🤗 Transformers 支持的任务 |
+| [预处理教程](https://huggingface.co/docs/transformers/preprocessing) | 使用 `Tokenizer` 来为模型准备数据 |
+| [训练和微调](https://huggingface.co/docs/transformers/training) | 在 PyTorch/TensorFlow 的训练循环或 `Trainer` API 中使用 🤗 Transformers 提供的模型 |
+| [快速上手：微调和用例脚本](https://github.com/huggingface/transformers/tree/main/examples) | 为各种任务提供的用例脚本 |
+| [模型分享和上传](https://huggingface.co/docs/transformers/model_sharing) | 和社区上传和分享你微调的模型 |
+| [迁移](https://huggingface.co/docs/transformers/migration) | 从 `pytorch-transformers` 或 `pytorch-pretrained-bert` 迁移到 🤗 Transformers |
+
+## 引用
+
+我们已将此库的[论文](https://www.aclweb.org/anthology/2020.emnlp-demos.6/)正式发表，如果你使用了 🤗 Transformers 库，请引用:
+```bibtex
+@inproceedings{wolf-etal-2020-transformers,
+    title = "Transformers: State-of-the-Art Natural Language Processing",
+    author = "Thomas Wolf and Lysandre Debut and Victor Sanh and Julien Chaumond and Clement Delangue and Anthony Moi and Pierric Cistac and Tim Rault and Rémi Louf and Morgan Funtowicz and Joe Davison and Sam Shleifer and Patrick von Platen and Clara Ma and Yacine Jernite and Julien Plu and Canwen Xu and Teven Le Scao and Sylvain Gugger and Mariama Drame and Quentin Lhoest and Alexander M. Rush",
+    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",
+    month = oct,
+    year = "2020",
+    address = "Online",
+    publisher = "Association for Computational Linguistics",
+    url = "https://www.aclweb.org/anthology/2020.emnlp-demos.6",
+    pages = "38--45"
+}
+```

README_zh-hant.md

@@ -0,0 +1,420 @@
+<!---
+Copyright 2020 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+-->
+
+<!---
+A useful guide for English-Traditional Chinese translation of Hugging Face documentation
+- Add space around English words and numbers when they appear between Chinese characters. E.g., 共 100 多種語言; 使用 transformers 函式庫。
+- Use square quotes, e.g.,「引用」
+- Some of terms in the file can be found at National Academy for Educational Research (https://terms.naer.edu.tw/), an official website providing bilingual translations between English and Traditional Chinese.
+
+Dictionary
+
+API: API (不翻譯）
+add: 加入
+checkpoint: 檢查點
+code: 程式碼
+community: 社群
+confidence: 信賴度
+dataset: 資料集
+documentation: 文件
+example: 基本翻譯為「範例」，或依語意翻為「例子」
+finetune: 微調
+Hugging Face: Hugging Face（不翻譯）
+implementation: 實作
+inference: 推論
+library: 函式庫
+module: 模組
+NLP/Natural Language Processing: 以 NLP 出現時不翻譯，以 Natural Language Processing 出現時翻譯為自然語言處理
+online demos: 線上Demo
+pipeline: pipeline（不翻譯）
+pretrained/pretrain: 預訓練
+Python data structures (e.g., list, set, dict): 翻譯為串列，集合，字典，並用括號標註原英文
+repository: repository（不翻譯）
+summary: 概覽
+token-: token-（不翻譯）
+Trainer: Trainer（不翻譯）
+transformer: transformer（不翻譯）
+tutorial: 教學
+user: 使用者
+-->
+
+<p align="center">
+    <br>
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_logo_name.png" width="400"/>
+    <br>
+<p>
+<p align="center">
+    <a href="https://circleci.com/gh/huggingface/transformers">
+        <img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main">
+    </a>
+    <a href="https://github.com/huggingface/transformers/blob/main/LICENSE">
+        <img alt="GitHub" src="https://img.shields.io/github/license/huggingface/transformers.svg?color=blue">
+    </a>
+    <a href="https://huggingface.co/docs/transformers/index">
+        <img alt="Documentation" src="https://img.shields.io/website/http/huggingface.co/docs/transformers/index.svg?down_color=red&down_message=offline&up_message=online">
+    </a>
+    <a href="https://github.com/huggingface/transformers/releases">
+        <img alt="GitHub release" src="https://img.shields.io/github/release/huggingface/transformers.svg">
+    </a>
+    <a href="https://github.com/huggingface/transformers/blob/main/CODE_OF_CONDUCT.md">
+        <img alt="Contributor Covenant" src="https://img.shields.io/badge/Contributor%20Covenant-v2.0%20adopted-ff69b4.svg">
+    </a>
+    <a href="https://zenodo.org/badge/latestdoi/155220641"><img src="https://zenodo.org/badge/155220641.svg" alt="DOI"></a>
+</p>
+
+<h4 align="center">
+    <p>
+        <a href="https://github.com/huggingface/transformers/">English</a> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_zh-hans.md">简体中文</a> |
+        <b>繁體中文</b> |
+        <a href="https://github.com/huggingface/transformers/blob/main/README_ko.md">한국어</a>
+    <p>
+</h4>
+
+<h3 align="center">
+    <p>為 Jax、PyTorch 以及 TensorFlow 打造的先進自然語言處理函式庫</p>
+</h3>
+
+<h3 align="center">
+    <a href="https://hf.co/course"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/course_banner.png"></a>
+</h3>
+
+🤗 Transformers 提供了數以千計的預訓練模型，支援 100 多種語言的文本分類、資訊擷取、問答、摘要、翻譯、文本生成。它的宗旨是讓最先進的 NLP 技術人人易用。
+
+🤗 Transformers 提供了便於快速下載和使用的API，讓你可以將預訓練模型用在給定文本、在你的資料集上微調然後經由 [model hub](https://huggingface.co/models) 與社群共享。同時，每個定義的 Python 模組架構均完全獨立，方便修改和快速研究實驗。
+
+🤗 Transformers 支援三個最熱門的深度學習函式庫： [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/) 以及 [TensorFlow](https://www.tensorflow.org/) — 並與之完美整合。你可以直接使用其中一個框架訓練你的模型，然後用另一個載入和推論。
+
+## 線上Demo
+
+你可以直接在 [model hub](https://huggingface.co/models) 上測試大多數的模型。我們也提供了 [私有模型託管、模型版本管理以及推論API](https://huggingface.co/pricing)。
+
+這裡是一些範例：
+- [用 BERT 做遮蓋填詞](https://huggingface.co/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
+- [用 Electra 做專有名詞辨識](https://huggingface.co/dbmdz/electra-large-discriminator-finetuned-conll03-english?text=My+name+is+Sarah+and+I+live+in+London+city)
+- [用 GPT-2 做文本生成](https://huggingface.co/gpt2?text=A+long+time+ago%2C+)
+- [用 RoBERTa 做自然語言推論](https://huggingface.co/roberta-large-mnli?text=The+dog+was+lost.+Nobody+lost+any+animal)
+- [用 BART 做文本摘要](https://huggingface.co/facebook/bart-large-cnn?text=The+tower+is+324+metres+%281%2C063+ft%29+tall%2C+about+the+same+height+as+an+81-storey+building%2C+and+the+tallest+structure+in+Paris.+Its+base+is+square%2C+measuring+125+metres+%28410+ft%29+on+each+side.+During+its+construction%2C+the+Eiffel+Tower+surpassed+the+Washington+Monument+to+become+the+tallest+man-made+structure+in+the+world%2C+a+title+it+held+for+41+years+until+the+Chrysler+Building+in+New+York+City+was+finished+in+1930.+It+was+the+first+structure+to+reach+a+height+of+300+metres.+Due+to+the+addition+of+a+broadcasting+aerial+at+the+top+of+the+tower+in+1957%2C+it+is+now+taller+than+the+Chrysler+Building+by+5.2+metres+%2817+ft%29.+Excluding+transmitters%2C+the+Eiffel+Tower+is+the+second+tallest+free-standing+structure+in+France+after+the+Millau+Viaduct)
+- [用 DistilBERT 做問答](https://huggingface.co/distilbert-base-uncased-distilled-squad?text=Which+name+is+also+used+to+describe+the+Amazon+rainforest+in+English%3F&context=The+Amazon+rainforest+%28Portuguese%3A+Floresta+Amaz%C3%B4nica+or+Amaz%C3%B4nia%3B+Spanish%3A+Selva+Amaz%C3%B3nica%2C+Amazon%C3%ADa+or+usually+Amazonia%3B+French%3A+For%C3%AAt+amazonienne%3B+Dutch%3A+Amazoneregenwoud%29%2C+also+known+in+English+as+Amazonia+or+the+Amazon+Jungle%2C+is+a+moist+broadleaf+forest+that+covers+most+of+the+Amazon+basin+of+South+America.+This+basin+encompasses+7%2C000%2C000+square+kilometres+%282%2C700%2C000+sq+mi%29%2C+of+which+5%2C500%2C000+square+kilometres+%282%2C100%2C000+sq+mi%29+are+covered+by+the+rainforest.+This+region+includes+territory+belonging+to+nine+nations.+The+majority+of+the+forest+is+contained+within+Brazil%2C+with+60%25+of+the+rainforest%2C+followed+by+Peru+with+13%25%2C+Colombia+with+10%25%2C+and+with+minor+amounts+in+Venezuela%2C+Ecuador%2C+Bolivia%2C+Guyana%2C+Suriname+and+French+Guiana.+States+or+departments+in+four+nations+contain+%22Amazonas%22+in+their+names.+The+Amazon+represents+over+half+of+the+planet%27s+remaining+rainforests%2C+and+comprises+the+largest+and+most+biodiverse+tract+of+tropical+rainforest+in+the+world%2C+with+an+estimated+390+billion+individual+trees+divided+into+16%2C000+species)
+- [用 T5 做翻譯](https://huggingface.co/t5-base?text=My+name+is+Wolfgang+and+I+live+in+Berlin)
+
+**[Write With Transformer](https://transformer.huggingface.co)**，由 Hugging Face 團隊所打造，是一個文本生成的官方 demo。
+
+## 如果你在尋找由 Hugging Face 團隊所提供的客製化支援服務
+
+<a target="_blank" href="https://huggingface.co/support">
+    <img alt="HuggingFace Expert Acceleration Program" src="https://huggingface.co/front/thumbnails/support.png" style="max-width: 600px; border: 1px solid #eee; border-radius: 4px; box-shadow: 0 1px 2px 0 rgba(0, 0, 0, 0.05);">
+</a><br>
+
+## 快速上手
+
+我們為快速使用模型提供了 `pipeline` API。 Pipeline 包含了預訓練模型和對應的文本預處理。下面是一個快速使用 pipeline 去判斷正負面情緒的例子：
+
+```python
+>>> from transformers import pipeline
+
+# 使用情緒分析 pipeline
+>>> classifier = pipeline('sentiment-analysis')
+>>> classifier('We are very happy to introduce pipeline to the transformers repository.')
+[{'label': 'POSITIVE', 'score': 0.9996980428695679}]
+```
+
+第二行程式碼下載並快取 pipeline 使用的預訓練模型，而第三行程式碼則在給定的文本上進行了評估。這裡的答案“正面” (positive) 具有 99.97% 的信賴度。
+
+許多的 NLP 任務都有隨選即用的預訓練 `pipeline`。例如，我們可以輕鬆地從給定文本中擷取問題答案：
+
+``` python
+>>> from transformers import pipeline
+
+# 使用問答 pipeline
+>>> question_answerer = pipeline('question-answering')
+>>> question_answerer({
+...     'question': 'What is the name of the repository ?',
+...     'context': 'Pipeline has been included in the huggingface/transformers repository'
+... })
+{'score': 0.30970096588134766, 'start': 34, 'end': 58, 'answer': 'huggingface/transformers'}
+
+```
+
+除了提供問題解答，預訓練模型還提供了對應的信賴度分數以及解答在 tokenized 後的文本中開始和結束的位置。你可以從[這個教學](https://huggingface.co/docs/transformers/task_summary)了解更多 `pipeline` API支援的任務。
+
+要在你的任務中下載和使用任何預訓練模型很簡單，只需三行程式碼。這裡是 PyTorch 版的範例：
+```python
+>>> from transformers import AutoTokenizer, AutoModel
+
+>>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
+>>> model = AutoModel.from_pretrained("bert-base-uncased")
+
+>>> inputs = tokenizer("Hello world!", return_tensors="pt")
+>>> outputs = model(**inputs)
+```
+這裡是對應的 TensorFlow 程式碼：
+```python
+>>> from transformers import AutoTokenizer, TFAutoModel
+
+>>> tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
+>>> model = TFAutoModel.from_pretrained("bert-base-uncased")
+
+>>> inputs = tokenizer("Hello world!", return_tensors="tf")
+>>> outputs = model(**inputs)
+```
+
+Tokenizer 為所有的預訓練模型提供了預處理，並可以直接轉換單一字串（比如上面的例子）或串列 (list)。它會輸出一個的字典 (dict) 讓你可以在下游程式碼裡使用或直接藉由 `**` 運算式傳給模型。
+
+模型本身是一個常規的 [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) 或 [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model)（取決於你的後端），可依常規方式使用。 [這個教學](https://huggingface.co/transformers/training.html)解釋了如何將這樣的模型整合到一般的 PyTorch 或 TensorFlow 訓練迴圈中，或是如何使用我們的 `Trainer` API 在一個新的資料集上快速進行微調。
+
+## 為什麼要用 transformers？
+
+1. 便於使用的先進模型：
+    - NLU 和 NLG 上性能卓越
+    - 對教學和實作友好且低門檻
+    - 高度抽象，使用者只須學習 3 個類別
+    - 對所有模型使用的制式化API
+
+1. 更低的運算成本，更少的碳排放：
+    - 研究人員可以分享已訓練的模型而非每次從頭開始訓練
+    - 工程師可以減少計算時間以及生產成本
+    - 數十種模型架構、兩千多個預訓練模型、100多種語言支援
+
+1. 對於模型生命週期的每一個部分都面面俱到：
+    - 訓練先進的模型，只需 3 行程式碼
+    - 模型可以在不同深度學習框架之間任意轉換
+    - 為訓練、評估和生產選擇最適合的框架，並完美銜接
+
+1. 為你的需求輕鬆客製化專屬模型和範例：
+    - 我們為每種模型架構提供了多個範例來重現原論文結果
+    - 一致的模型內部架構
+    - 模型檔案可單獨使用，便於修改和快速實驗
+
+## 什麼情況下我不該用 transformers？
+
+- 本函式庫並不是模組化的神經網絡工具箱。模型文件中的程式碼並未做額外的抽象封裝，以便研究人員快速地翻閱及修改程式碼，而不會深陷複雜的類別包裝之中。
+- `Trainer` API 並非相容任何模型，它只為本函式庫中的模型最佳化。對於一般的機器學習用途，請使用其他函式庫。
+- 儘管我們已盡力而為，[examples 目錄](https://github.com/huggingface/transformers/tree/main/examples)中的腳本也僅為範例而已。對於特定問題，它們並不一定隨選即用，可能需要修改幾行程式碼以符合需求。
+
+## 安裝
+
+### 使用 pip
+
+這個 Repository 已在 Python 3.6+、Flax 0.3.2+、PyTorch 1.3.1+ 和 TensorFlow 2.3+ 下經過測試。
+
+你可以在[虛擬環境](https://docs.python.org/3/library/venv.html)中安裝 🤗 Transformers。如果你還不熟悉 Python 的虛擬環境，請閱此[使用者指引](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/)。
+
+首先，用你打算使用的版本的 Python 創建一個虛擬環境並進入。
+
+然後，你需要安裝 Flax、PyTorch 或 TensorFlow 其中之一。對於該如何在你使用的平台上安裝這些框架，請參閱 [TensorFlow 安裝頁面](https://www.tensorflow.org/install/), [PyTorch 安裝頁面](https://pytorch.org/get-started/locally/#start-locally) 或 [Flax 安裝頁面](https://github.com/google/flax#quick-install)。
+
+當其中一個後端安裝成功後，🤗 Transformers 可依此安裝：
+
+```bash
+pip install transformers
+```
+
+如果你想要試試範例或者想在正式發布前使用最新開發中的程式碼，你必須[從原始碼安裝](https://huggingface.co/docs/transformers/installation#installing-from-source)。
+
+### 使用 conda
+
+自 Transformers 4.0.0 版始，我們有了一個 conda channel： `huggingface`。
+
+🤗 Transformers 可以藉由 conda 依此安裝：
+
+```shell script
+conda install -c huggingface transformers
+```
+
+要藉由 conda 安裝 Flax、PyTorch 或 TensorFlow 其中之一，請參閱它們各自安裝頁面的說明。
+
+## 模型架構
+
+**🤗 Transformers 支援的[所有的模型檢查點](https://huggingface.co/models)**，由[使用者](https://huggingface.co/users)和[組織](https://huggingface.co/organizations)上傳，均與 huggingface.co [model hub](https://huggingface.co) 完美結合。
+
+目前的檢查點數量： ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
+
+🤗 Transformers 目前支援以下的架構（模型概覽請參閱[這裡](https://huggingface.co/docs/transformers/model_summary)）：
+
+1. **[ALBERT](https://huggingface.co/docs/transformers/model_doc/albert)** (from Google Research and the Toyota Technological Institute at Chicago) released with the paper [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://arxiv.org/abs/1909.11942), by Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, Radu Soricut.
+1. **[BART](https://huggingface.co/docs/transformers/model_doc/bart)** (from Facebook) released with the paper [BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension](https://arxiv.org/pdf/1910.13461.pdf) by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer.
+1. **[BARThez](https://huggingface.co/docs/transformers/model_doc/barthez)** (from École polytechnique) released with the paper [BARThez: a Skilled Pretrained French Sequence-to-Sequence Model](https://arxiv.org/abs/2010.12321) by Moussa Kamal Eddine, Antoine J.-P. Tixier, Michalis Vazirgiannis.
+1. **[BARTpho](https://huggingface.co/docs/transformers/model_doc/bartpho)** (from VinAI Research) released with the paper [BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese](https://arxiv.org/abs/2109.09701) by Nguyen Luong Tran, Duong Minh Le and Dat Quoc Nguyen.
+1. **[BEiT](https://huggingface.co/docs/transformers/model_doc/beit)** (from Microsoft) released with the paper [BEiT: BERT Pre-Training of Image Transformers](https://arxiv.org/abs/2106.08254) by Hangbo Bao, Li Dong, Furu Wei.
+1. **[BERT](https://huggingface.co/docs/transformers/model_doc/bert)** (from Google) released with the paper [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova.
+1. **[BERT For Sequence Generation](https://huggingface.co/docs/transformers/model_doc/bert-generation)** (from Google) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+1. **[BERTweet](https://huggingface.co/docs/transformers/model_doc/bertweet)** (from VinAI Research) released with the paper [BERTweet: A pre-trained language model for English Tweets](https://aclanthology.org/2020.emnlp-demos.2/) by Dat Quoc Nguyen, Thanh Vu and Anh Tuan Nguyen.
+1. **[BigBird-Pegasus](https://huggingface.co/docs/transformers/model_doc/bigbird_pegasus)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
+1. **[BigBird-RoBERTa](https://huggingface.co/docs/transformers/model_doc/big_bird)** (from Google Research) released with the paper [Big Bird: Transformers for Longer Sequences](https://arxiv.org/abs/2007.14062) by Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, Amr Ahmed.
+1. **[Blenderbot](https://huggingface.co/docs/transformers/model_doc/blenderbot)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
+1. **[BlenderbotSmall](https://huggingface.co/docs/transformers/model_doc/blenderbot-small)** (from Facebook) released with the paper [Recipes for building an open-domain chatbot](https://arxiv.org/abs/2004.13637) by Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston.
+1. **[BLOOM](https://huggingface.co/docs/transformers/model_doc/bloom)** (from BigScience workshop) released by the [BigSicence Workshop](https://bigscience.huggingface.co/). 
+1. **[BORT](https://huggingface.co/docs/transformers/model_doc/bort)** (from Alexa) released with the paper [Optimal Subarchitecture Extraction For BERT](https://arxiv.org/abs/2010.10499) by Adrian de Wynter and Daniel J. Perry.
+1. **[ByT5](https://huggingface.co/docs/transformers/model_doc/byt5)** (from Google Research) released with the paper [ByT5: Towards a token-free future with pre-trained byte-to-byte models](https://arxiv.org/abs/2105.13626) by Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, Colin Raffel.
+1. **[CamemBERT](https://huggingface.co/docs/transformers/model_doc/camembert)** (from Inria/Facebook/Sorbonne) released with the paper [CamemBERT: a Tasty French Language Model](https://arxiv.org/abs/1911.03894) by Louis Martin*, Benjamin Muller*, Pedro Javier Ortiz Suárez*, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah and Benoît Sagot.
+1. **[CANINE](https://huggingface.co/docs/transformers/model_doc/canine)** (from Google Research) released with the paper [CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation](https://arxiv.org/abs/2103.06874) by Jonathan H. Clark, Dan Garrette, Iulia Turc, John Wieting.
+1. **[CLIP](https://huggingface.co/docs/transformers/model_doc/clip)** (from OpenAI) released with the paper [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) by Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever.
+1. **[CodeGen](https://huggingface.co/docs/transformers/model_doc/codegen)** (from Salesforce) released with the paper [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) by Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong.
+1. **[ConvBERT](https://huggingface.co/docs/transformers/model_doc/convbert)** (from YituTech) released with the paper [ConvBERT: Improving BERT with Span-based Dynamic Convolution](https://arxiv.org/abs/2008.02496) by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan.
+1. **[ConvNeXT](https://huggingface.co/docs/transformers/model_doc/convnext)** (from Facebook AI) released with the paper [A ConvNet for the 2020s](https://arxiv.org/abs/2201.03545) by Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, Saining Xie.
+1. **[CPM](https://huggingface.co/docs/transformers/model_doc/cpm)** (from Tsinghua University) released with the paper [CPM: A Large-scale Generative Chinese Pre-trained Language Model](https://arxiv.org/abs/2012.00413) by Zhengyan Zhang, Xu Han, Hao Zhou, Pei Ke, Yuxian Gu, Deming Ye, Yujia Qin, Yusheng Su, Haozhe Ji, Jian Guan, Fanchao Qi, Xiaozhi Wang, Yanan Zheng, Guoyang Zeng, Huanqi Cao, Shengqi Chen, Daixuan Li, Zhenbo Sun, Zhiyuan Liu, Minlie Huang, Wentao Han, Jie Tang, Juanzi Li, Xiaoyan Zhu, Maosong Sun.
+1. **[CTRL](https://huggingface.co/docs/transformers/model_doc/ctrl)** (from Salesforce) released with the paper [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://arxiv.org/abs/1909.05858) by Nitish Shirish Keskar*, Bryan McCann*, Lav R. Varshney, Caiming Xiong and Richard Socher.
+1. **[CvT](https://huggingface.co/docs/transformers/model_doc/cvt)** (from Microsoft) released with the paper [CvT: Introducing Convolutions to Vision Transformers](https://arxiv.org/abs/2103.15808) by Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, Lei Zhang.
+1. **[Data2Vec](https://huggingface.co/docs/transformers/model_doc/data2vec)** (from Facebook) released with the paper [Data2Vec:  A General Framework for Self-supervised Learning in Speech, Vision and Language](https://arxiv.org/abs/2202.03555) by Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu, Michael Auli.
+1. **[DeBERTa](https://huggingface.co/docs/transformers/model_doc/deberta)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
+1. **[DeBERTa-v2](https://huggingface.co/docs/transformers/model_doc/deberta-v2)** (from Microsoft) released with the paper [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://arxiv.org/abs/2006.03654) by Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen.
+1. **[Decision Transformer](https://huggingface.co/docs/transformers/model_doc/decision_transformer)** (from Berkeley/Facebook/Google) released with the paper [Decision Transformer: Reinforcement Learning via Sequence Modeling](https://arxiv.org/abs/2106.01345) by Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Michael Laskin, Pieter Abbeel, Aravind Srinivas, Igor Mordatch.
+1. **[DeiT](https://huggingface.co/docs/transformers/model_doc/deit)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
+1. **[DETR](https://huggingface.co/docs/transformers/model_doc/detr)** (from Facebook) released with the paper [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko.
+1. **[DialoGPT](https://huggingface.co/docs/transformers/model_doc/dialogpt)** (from Microsoft Research) released with the paper [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.
+1. **[DistilBERT](https://huggingface.co/docs/transformers/model_doc/distilbert)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/main/examples/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/main/examples/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/main/examples/distillation) and a German version of DistilBERT.
+1. **[DiT](https://huggingface.co/docs/transformers/model_doc/dit)** (from Microsoft Research) released with the paper [DiT: Self-supervised Pre-training for Document Image Transformer](https://arxiv.org/abs/2203.02378) by Junlong Li, Yiheng Xu, Tengchao Lv, Lei Cui, Cha Zhang, Furu Wei.
+1. **[Donut](https://huggingface.co/docs/transformers/model_doc/donut)** (from NAVER) released with the paper [OCR-free Document Understanding Transformer](https://arxiv.org/abs/2111.15664) by Geewook Kim, Teakgyu Hong, Moonbin Yim, Jeongyeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, Seunghyun Park.
+1. **[DPR](https://huggingface.co/docs/transformers/model_doc/dpr)** (from Facebook) released with the paper [Dense Passage Retrieval for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) by Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
+1. **[DPT](https://huggingface.co/docs/transformers/master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.
+1. **[ELECTRA](https://huggingface.co/docs/transformers/model_doc/electra)** (from Google Research/Stanford University) released with the paper [ELECTRA: Pre-training text encoders as discriminators rather than generators](https://arxiv.org/abs/2003.10555) by Kevin Clark, Minh-Thang Luong, Quoc V. Le, Christopher D. Manning.
+1. **[EncoderDecoder](https://huggingface.co/docs/transformers/model_doc/encoder-decoder)** (from Google Research) released with the paper [Leveraging Pre-trained Checkpoints for Sequence Generation Tasks](https://arxiv.org/abs/1907.12461) by Sascha Rothe, Shashi Narayan, Aliaksei Severyn.
+1. **[ERNIE](https://huggingface.co/docs/transformers/model_doc/ernie)** (from Baidu) released with the paper [ERNIE: Enhanced Representation through Knowledge Integration](https://arxiv.org/abs/1904.09223) by Yu Sun, Shuohuan Wang, Yukun Li, Shikun Feng, Xuyi Chen, Han Zhang, Xin Tian, Danxiang Zhu, Hao Tian, Hua Wu.
+1. **[FlauBERT](https://huggingface.co/docs/transformers/model_doc/flaubert)** (from CNRS) released with the paper [FlauBERT: Unsupervised Language Model Pre-training for French](https://arxiv.org/abs/1912.05372) by Hang Le, Loïc Vial, Jibril Frej, Vincent Segonne, Maximin Coavoux, Benjamin Lecouteux, Alexandre Allauzen, Benoît Crabbé, Laurent Besacier, Didier Schwab.
+1. **[FLAVA](https://huggingface.co/docs/transformers/model_doc/flava)** (from Facebook AI) released with the paper [FLAVA: A Foundational Language And Vision Alignment Model](https://arxiv.org/abs/2112.04482) by Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela.
+1. **[FNet](https://huggingface.co/docs/transformers/model_doc/fnet)** (from Google Research) released with the paper [FNet: Mixing Tokens with Fourier Transforms](https://arxiv.org/abs/2105.03824) by James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, Santiago Ontanon.
+1. **[Funnel Transformer](https://huggingface.co/docs/transformers/model_doc/funnel)** (from CMU/Google Brain) released with the paper [Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing](https://arxiv.org/abs/2006.03236) by Zihang Dai, Guokun Lai, Yiming Yang, Quoc V. Le.
+1. **[GLPN](https://huggingface.co/docs/transformers/model_doc/glpn)** (from KAIST) released with the paper [Global-Local Path Networks for Monocular Depth Estimation with Vertical CutDepth](https://arxiv.org/abs/2201.07436) by Doyeon Kim, Woonghyun Ga, Pyungwhan Ahn, Donggyu Joo, Sehwan Chun, Junmo Kim.
+1. **[GPT](https://huggingface.co/docs/transformers/model_doc/openai-gpt)** (from OpenAI) released with the paper [Improving Language Understanding by Generative Pre-Training](https://blog.openai.com/language-unsupervised/) by Alec Radford, Karthik Narasimhan, Tim Salimans and Ilya Sutskever.
+1. **[GPT Neo](https://huggingface.co/docs/transformers/model_doc/gpt_neo)** (from EleutherAI) released in the repository [EleutherAI/gpt-neo](https://github.com/EleutherAI/gpt-neo) by Sid Black, Stella Biderman, Leo Gao, Phil Wang and Connor Leahy.
+1. **[GPT NeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox)** (from EleutherAI) released with the paper [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://arxiv.org/abs/2204.06745) by Sid Black, Stella Biderman, Eric Hallahan, Quentin Anthony, Leo Gao, Laurence Golding, Horace He, Connor Leahy, Kyle McDonell, Jason Phang, Michael Pieler, USVSN Sai Prashanth, Shivanshu Purohit, Laria Reynolds, Jonathan Tow, Ben Wang, Samuel Weinbach 
+1. **[GPT-2](https://huggingface.co/docs/transformers/model_doc/gpt2)** (from OpenAI) released with the paper [Language Models are Unsupervised Multitask Learners](https://blog.openai.com/better-language-models/) by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
+1. **[GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj)** (from EleutherAI) released with the paper [kingoflolz/mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax/) by Ben Wang and Aran Komatsuzaki.
+1. **[GroupViT](https://huggingface.co/docs/transformers/model_doc/groupvit)** (from UCSD, NVIDIA) released with the paper [GroupViT: Semantic Segmentation Emerges from Text Supervision](https://arxiv.org/abs/2202.11094) by Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang.
+1. **[Hubert](https://huggingface.co/docs/transformers/model_doc/hubert)** (from Facebook) released with the paper [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://arxiv.org/abs/2106.07447) by Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, Abdelrahman Mohamed.
+1. **[I-BERT](https://huggingface.co/docs/transformers/model_doc/ibert)** (from Berkeley) released with the paper [I-BERT: Integer-only BERT Quantization](https://arxiv.org/abs/2101.01321) by Sehoon Kim, Amir Gholami, Zhewei Yao, Michael W. Mahoney, Kurt Keutzer.
+1. **[ImageGPT](https://huggingface.co/docs/transformers/model_doc/imagegpt)** (from OpenAI) released with the paper [Generative Pretraining from Pixels](https://openai.com/blog/image-gpt/) by Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, Ilya Sutskever.
+1. **[LayoutLM](https://huggingface.co/docs/transformers/model_doc/layoutlm)** (from Microsoft Research Asia) released with the paper [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) by Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou.
+1. **[LayoutLMv2](https://huggingface.co/docs/transformers/model_doc/layoutlmv2)** (from Microsoft Research Asia) released with the paper [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) by Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou.
+1. **[LayoutLMv3](https://huggingface.co/docs/transformers/model_doc/layoutlmv3)** (from Microsoft Research Asia) released with the paper [LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking](https://arxiv.org/abs/2204.08387) by Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei.
+1. **[LayoutXLM](https://huggingface.co/docs/transformers/model_doc/layoutxlm)** (from Microsoft Research Asia) released with the paper [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) by Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei.
+1. **[LED](https://huggingface.co/docs/transformers/model_doc/led)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
+1. **[LeViT](https://huggingface.co/docs/transformers/model_doc/levit)** (from Meta AI) released with the paper [LeViT: A Vision Transformer in ConvNet's Clothing for Faster Inference](https://arxiv.org/abs/2104.01136) by Ben Graham, Alaaeldin El-Nouby, Hugo Touvron, Pierre Stock, Armand Joulin, Hervé Jégou, Matthijs Douze.
+1. **[Longformer](https://huggingface.co/docs/transformers/model_doc/longformer)** (from AllenAI) released with the paper [Longformer: The Long-Document Transformer](https://arxiv.org/abs/2004.05150) by Iz Beltagy, Matthew E. Peters, Arman Cohan.
+1. **[LongT5](https://huggingface.co/docs/transformers/model_doc/longt5)** (from Google AI) released with the paper [LongT5: Efficient Text-To-Text Transformer for Long Sequences](https://arxiv.org/abs/2112.07916) by Mandy Guo, Joshua Ainslie, David Uthus, Santiago Ontanon, Jianmo Ni, Yun-Hsuan Sung, Yinfei Yang.
+1. **[LUKE](https://huggingface.co/docs/transformers/model_doc/luke)** (from Studio Ousia) released with the paper [LUKE: Deep Contextualized Entity Representations with Entity-aware Self-attention](https://arxiv.org/abs/2010.01057) by Ikuya Yamada, Akari Asai, Hiroyuki Shindo, Hideaki Takeda, Yuji Matsumoto.
+1. **[LXMERT](https://huggingface.co/docs/transformers/model_doc/lxmert)** (from UNC Chapel Hill) released with the paper [LXMERT: Learning Cross-Modality Encoder Representations from Transformers for Open-Domain Question Answering](https://arxiv.org/abs/1908.07490) by Hao Tan and Mohit Bansal.
+1. **[M-CTC-T](https://huggingface.co/docs/transformers/model_doc/mctct)** (from Facebook) released with the paper [Pseudo-Labeling For Massively Multilingual Speech Recognition](https://arxiv.org/abs/2111.00161) by Loren Lugosch, Tatiana Likhomanenko, Gabriel Synnaeve, and Ronan Collobert.
+1. **[M2M100](https://huggingface.co/docs/transformers/model_doc/m2m_100)** (from Facebook) released with the paper [Beyond English-Centric Multilingual Machine Translation](https://arxiv.org/abs/2010.11125) by Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El-Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, Naman Goyal, Tom Birch, Vitaliy Liptchinsky, Sergey Edunov, Edouard Grave, Michael Auli, Armand Joulin.
+1. **[MarianMT](https://huggingface.co/docs/transformers/model_doc/marian)** Machine translation models trained using [OPUS](http://opus.nlpl.eu/) data by Jörg Tiedemann. The [Marian Framework](https://marian-nmt.github.io/) is being developed by the Microsoft Translator Team.
+1. **[MaskFormer](https://huggingface.co/docs/transformers/model_doc/maskformer)** (from Meta and UIUC) released with the paper [Per-Pixel Classification is Not All You Need for Semantic Segmentation](https://arxiv.org/abs/2107.06278) by Bowen Cheng, Alexander G. Schwing, Alexander Kirillov 
+1. **[mBART](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Denoising Pre-training for Neural Machine Translation](https://arxiv.org/abs/2001.08210) by Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, Luke Zettlemoyer.
+1. **[mBART-50](https://huggingface.co/docs/transformers/model_doc/mbart)** (from Facebook) released with the paper [Multilingual Translation with Extensible Multilingual Pretraining and Finetuning](https://arxiv.org/abs/2008.00401) by Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan.
+1. **[Megatron-BERT](https://huggingface.co/docs/transformers/model_doc/megatron-bert)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
+1. **[Megatron-GPT2](https://huggingface.co/docs/transformers/model_doc/megatron_gpt2)** (from NVIDIA) released with the paper [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/abs/1909.08053) by Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper and Bryan Catanzaro.
+1. **[mLUKE](https://huggingface.co/docs/transformers/model_doc/mluke)** (from Studio Ousia) released with the paper [mLUKE: The Power of Entity Representations in Multilingual Pretrained Language Models](https://arxiv.org/abs/2110.08151) by Ryokan Ri, Ikuya Yamada, and Yoshimasa Tsuruoka.
+1. **[MobileBERT](https://huggingface.co/docs/transformers/model_doc/mobilebert)** (from CMU/Google Brain) released with the paper [MobileBERT: a Compact Task-Agnostic BERT for Resource-Limited Devices](https://arxiv.org/abs/2004.02984) by Zhiqing Sun, Hongkun Yu, Xiaodan Song, Renjie Liu, Yiming Yang, and Denny Zhou.
+1. **[MobileViT](https://huggingface.co/docs/transformers/model_doc/mobilevit)** (from Apple) released with the paper [MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer](https://arxiv.org/abs/2110.02178) by Sachin Mehta and Mohammad Rastegari.
+1. **[MPNet](https://huggingface.co/docs/transformers/model_doc/mpnet)** (from Microsoft Research) released with the paper [MPNet: Masked and Permuted Pre-training for Language Understanding](https://arxiv.org/abs/2004.09297) by Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, Tie-Yan Liu.
+1. **[MT5](https://huggingface.co/docs/transformers/model_doc/mt5)** (from Google AI) released with the paper [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel.
+1. **[MVP](https://huggingface.co/docs/transformers/model_doc/mvp)** (from RUC AI Box) released with the paper [MVP: Multi-task Supervised Pre-training for Natural Language Generation](https://arxiv.org/abs/2206.12131) by Tianyi Tang, Junyi Li, Wayne Xin Zhao and Ji-Rong Wen.
+1. **[Nezha](https://huggingface.co/docs/transformers/model_doc/nezha)** (from Huawei Noah’s Ark Lab) released with the paper [NEZHA: Neural Contextualized Representation for Chinese Language Understanding](https://arxiv.org/abs/1909.00204) by Junqiu Wei, Xiaozhe Ren, Xiaoguang Li, Wenyong Huang, Yi Liao, Yasheng Wang, Jiashu Lin, Xin Jiang, Xiao Chen and Qun Liu.
+1. **[NLLB](https://huggingface.co/docs/transformers/model_doc/nllb)** (from Meta) released with the paper [No Language Left Behind: Scaling Human-Centered Machine Translation](https://arxiv.org/abs/2207.04672) by the NLLB team.
+1. **[Nyströmformer](https://huggingface.co/docs/transformers/model_doc/nystromformer)** (from the University of Wisconsin - Madison) released with the paper [Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention](https://arxiv.org/abs/2102.03902) by Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, Vikas Singh.
+1. **[OPT](https://huggingface.co/docs/transformers/master/model_doc/opt)** (from Meta AI) released with the paper [OPT: Open Pre-trained Transformer Language Models](https://arxiv.org/abs/2205.01068) by Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen et al.
+1. **[OWL-ViT](https://huggingface.co/docs/transformers/model_doc/owlvit)** (from Google AI) released with the paper [Simple Open-Vocabulary Object Detection with Vision Transformers](https://arxiv.org/abs/2205.06230) by Matthias Minderer, Alexey Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby.
+1. **[Pegasus](https://huggingface.co/docs/transformers/model_doc/pegasus)** (from Google) released with the paper [PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization](https://arxiv.org/abs/1912.08777) by Jingqing Zhang, Yao Zhao, Mohammad Saleh and Peter J. Liu.
+1. **[PEGASUS-X](https://huggingface.co/docs/transformers/model_doc/pegasus_x)** (from Google) released with the paper [Investigating Efficiently Extending Transformers for Long Input Summarization](https://arxiv.org/abs/2208.04347) by Jason Phang, Yao Zhao, Peter J. Liu.
+1. **[Perceiver IO](https://huggingface.co/docs/transformers/model_doc/perceiver)** (from Deepmind) released with the paper [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://arxiv.org/abs/2107.14795) by Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, Olivier Hénaff, Matthew M. Botvinick, Andrew Zisserman, Oriol Vinyals, João Carreira.
+1. **[PhoBERT](https://huggingface.co/docs/transformers/model_doc/phobert)** (from VinAI Research) released with the paper [PhoBERT: Pre-trained language models for Vietnamese](https://www.aclweb.org/anthology/2020.findings-emnlp.92/) by Dat Quoc Nguyen and Anh Tuan Nguyen.
+1. **[PLBart](https://huggingface.co/docs/transformers/model_doc/plbart)** (from UCLA NLP) released with the paper [Unified Pre-training for Program Understanding and Generation](https://arxiv.org/abs/2103.06333) by Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, Kai-Wei Chang.
+1. **[PoolFormer](https://huggingface.co/docs/transformers/model_doc/poolformer)** (from Sea AI Labs) released with the paper [MetaFormer is Actually What You Need for Vision](https://arxiv.org/abs/2111.11418) by Yu, Weihao and Luo, Mi and Zhou, Pan and Si, Chenyang and Zhou, Yichen and Wang, Xinchao and Feng, Jiashi and Yan, Shuicheng.
+1. **[ProphetNet](https://huggingface.co/docs/transformers/model_doc/prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
+1. **[QDQBert](https://huggingface.co/docs/transformers/model_doc/qdqbert)** (from NVIDIA) released with the paper [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://arxiv.org/abs/2004.09602) by Hao Wu, Patrick Judd, Xiaojie Zhang, Mikhail Isaev and Paulius Micikevicius.
+1. **[RAG](https://huggingface.co/docs/transformers/model_doc/rag)** (from Facebook) released with the paper [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) by Patrick Lewis, Ethan Perez, Aleksandara Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, Douwe Kiela.
+1. **[REALM](https://huggingface.co/docs/transformers/model_doc/realm.html)** (from Google Research) released with the paper [REALM: Retrieval-Augmented Language Model Pre-Training](https://arxiv.org/abs/2002.08909) by Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang.
+1. **[Reformer](https://huggingface.co/docs/transformers/model_doc/reformer)** (from Google Research) released with the paper [Reformer: The Efficient Transformer](https://arxiv.org/abs/2001.04451) by Nikita Kitaev, Łukasz Kaiser, Anselm Levskaya.
+1. **[RegNet](https://huggingface.co/docs/transformers/model_doc/regnet)** (from META Research) released with the paper [Designing Network Design Space](https://arxiv.org/abs/2003.13678) by Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, Piotr Dollár. 
+1. **[RemBERT](https://huggingface.co/docs/transformers/model_doc/rembert)** (from Google Research) released with the paper [Rethinking embedding coupling in pre-trained language models](https://arxiv.org/pdf/2010.12821.pdf) by Hyung Won Chung, Thibault Févry, Henry Tsai, M. Johnson, Sebastian Ruder.
+1. **[ResNet](https://huggingface.co/docs/transformers/model_doc/resnet)** (from Microsoft Research) released with the paper [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun. 
+1. **[RoBERTa](https://huggingface.co/docs/transformers/model_doc/roberta)** (from Facebook), released together with the paper a [Robustly Optimized BERT Pretraining Approach](https://arxiv.org/abs/1907.11692) by Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov.
+1. **[RoFormer](https://huggingface.co/docs/transformers/model_doc/roformer)** (from ZhuiyiTechnology), released together with the paper a [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://arxiv.org/pdf/2104.09864v1.pdf) by Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu.
+1. **[SegFormer](https://huggingface.co/docs/transformers/model_doc/segformer)** (from NVIDIA) released with the paper [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203) by Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo.
+1. **[SEW](https://huggingface.co/docs/transformers/model_doc/sew)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
+1. **[SEW-D](https://huggingface.co/docs/transformers/model_doc/sew_d)** (from ASAPP) released with the paper [Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition](https://arxiv.org/abs/2109.06870) by Felix Wu, Kwangyoun Kim, Jing Pan, Kyu Han, Kilian Q. Weinberger, Yoav Artzi.
+1. **[SpeechToTextTransformer](https://huggingface.co/docs/transformers/model_doc/speech_to_text)** (from Facebook), released together with the paper [fairseq S2T: Fast Speech-to-Text Modeling with fairseq](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Dmytro Okhonko, Juan Pino.
+1. **[SpeechToTextTransformer2](https://huggingface.co/docs/transformers/model_doc/speech_to_text_2)** (from Facebook) released with the paper [Large-Scale Self- and Semi-Supervised Learning for Speech Translation](https://arxiv.org/abs/2104.06678) by Changhan Wang, Anne Wu, Juan Pino, Alexei Baevski, Michael Auli, Alexis Conneau.
+1. **[Splinter](https://huggingface.co/docs/transformers/model_doc/splinter)** (from Tel Aviv University) released with the paper [Few-Shot Question Answering by Pretraining Span Selection](https://arxiv.org/abs/2101.00438) by Ori Ram, Yuval Kirstain, Jonathan Berant, Amir Globerson, Omer Levy.
+1. **[SqueezeBERT](https://huggingface.co/docs/transformers/model_doc/squeezebert)** (from Berkeley) released with the paper [SqueezeBERT: What can computer vision teach NLP about efficient neural networks?](https://arxiv.org/abs/2006.11316) by Forrest N. Iandola, Albert E. Shaw, Ravi Krishna, and Kurt W. Keutzer.
+1. **[Swin Transformer](https://huggingface.co/docs/transformers/model_doc/swin)** (from Microsoft) released with the paper [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://arxiv.org/abs/2103.14030) by Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo.
+1. **[Swin Transformer V2](https://huggingface.co/docs/transformers/model_doc/swinv2)** (from Microsoft) released with the paper [Swin Transformer V2: Scaling Up Capacity and Resolution](https://arxiv.org/abs/2111.09883) by Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, Baining Guo.
+1. **[T5](https://huggingface.co/docs/transformers/model_doc/t5)** (from Google AI) released with the paper [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://arxiv.org/abs/1910.10683) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
+1. **[T5v1.1](https://huggingface.co/docs/transformers/model_doc/t5v1.1)** (from Google AI) released with the paper [google-research/text-to-text-transfer-transformer](https://github.com/google-research/text-to-text-transfer-transformer/blob/main/released_checkpoints.md#t511) by Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu.
+1. **[TAPAS](https://huggingface.co/docs/transformers/model_doc/tapas)** (from Google AI) released with the paper [TAPAS: Weakly Supervised Table Parsing via Pre-training](https://arxiv.org/abs/2004.02349) by Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller, Francesco Piccinno and Julian Martin Eisenschlos.
+1. **[TAPEX](https://huggingface.co/docs/transformers/model_doc/tapex)** (from Microsoft Research) released with the paper [TAPEX: Table Pre-training via Learning a Neural SQL Executor](https://arxiv.org/abs/2107.07653) by Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, Jian-Guang Lou.
+1. **[Trajectory Transformer](https://huggingface.co/docs/transformers/model_doc/trajectory_transformers)** (from the University of California at Berkeley) released with the paper [Offline Reinforcement Learning as One Big Sequence Modeling Problem](https://arxiv.org/abs/2106.02039) by Michael Janner, Qiyang Li, Sergey Levine 
+1. **[Transformer-XL](https://huggingface.co/docs/transformers/model_doc/transfo-xl)** (from Google/CMU) released with the paper [Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context](https://arxiv.org/abs/1901.02860) by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov.
+1. **[TrOCR](https://huggingface.co/docs/transformers/model_doc/trocr)** (from Microsoft) released with the paper [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282) by Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, Furu Wei.
+1. **[UL2](https://huggingface.co/docs/transformers/model_doc/ul2)** (from Google Research) released with the paper [Unifying Language Learning Paradigms](https://arxiv.org/abs/2205.05131v1) by Yi Tay, Mostafa Dehghani, Vinh Q. Tran, Xavier Garcia, Dara Bahri, Tal Schuster, Huaixiu Steven Zheng, Neil Houlsby, Donald Metzler 
+1. **[UniSpeech](https://huggingface.co/docs/transformers/model_doc/unispeech)** (from Microsoft Research) released with the paper [UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data](https://arxiv.org/abs/2101.07597) by Chengyi Wang, Yu Wu, Yao Qian, Kenichi Kumatani, Shujie Liu, Furu Wei, Michael Zeng, Xuedong Huang.
+1. **[UniSpeechSat](https://huggingface.co/docs/transformers/model_doc/unispeech-sat)** (from Microsoft Research) released with the paper [UNISPEECH-SAT: UNIVERSAL SPEECH REPRESENTATION LEARNING WITH SPEAKER AWARE PRE-TRAINING](https://arxiv.org/abs/2110.05752) by Sanyuan Chen, Yu Wu, Chengyi Wang, Zhengyang Chen, Zhuo Chen, Shujie Liu, Jian Wu, Yao Qian, Furu Wei, Jinyu Li, Xiangzhan Yu.
+1. **[VAN](https://huggingface.co/docs/transformers/model_doc/van)** (from Tsinghua University and Nankai University) released with the paper [Visual Attention Network](https://arxiv.org/pdf/2202.09741.pdf) by Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, Shi-Min Hu.
+1. **[VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)** (from Multimedia Computing Group, Nanjing University) released with the paper [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://arxiv.org/abs/2203.12602) by Zhan Tong, Yibing Song, Jue Wang, Limin Wang.
+1. **[ViLT](https://huggingface.co/docs/transformers/model_doc/vilt)** (from NAVER AI Lab/Kakao Enterprise/Kakao Brain) released with the paper [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://arxiv.org/abs/2102.03334) by Wonjae Kim, Bokyung Son, Ildoo Kim.
+1. **[Vision Transformer (ViT)](https://huggingface.co/docs/transformers/model_doc/vit)** (from Google AI) released with the paper [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://arxiv.org/abs/2010.11929) by Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, Neil Houlsby. 
+1. **[VisualBERT](https://huggingface.co/docs/transformers/model_doc/visual_bert)** (from UCLA NLP) released with the paper [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://arxiv.org/pdf/1908.03557) by Liunian Harold Li, Mark Yatskar, Da Yin, Cho-Jui Hsieh, Kai-Wei Chang.
+1. **[ViTMAE](https://huggingface.co/docs/transformers/model_doc/vit_mae)** (from Meta AI) released with the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377) by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
+1. **[Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/wav2vec2)** (from Facebook AI) released with the paper [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://arxiv.org/abs/2006.11477) by Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli.
+1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
+1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (from Facebook AI) released with the paper [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) by Qiantong Xu, Alexei Baevski, Michael Auli.
+1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
+1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
+1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li. 
+1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
+1. **[XLM-ProphetNet](https://huggingface.co/docs/transformers/model_doc/xlm-prophetnet)** (from Microsoft Research) released with the paper [ProphetNet: Predicting Future N-gram for Sequence-to-Sequence Pre-training](https://arxiv.org/abs/2001.04063) by Yu Yan, Weizhen Qi, Yeyun Gong, Dayiheng Liu, Nan Duan, Jiusheng Chen, Ruofei Zhang and Ming Zhou.
+1. **[XLM-RoBERTa](https://huggingface.co/docs/transformers/model_doc/xlm-roberta)** (from Facebook AI), released together with the paper [Unsupervised Cross-lingual Representation Learning at Scale](https://arxiv.org/abs/1911.02116) by Alexis Conneau*, Kartikay Khandelwal*, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer and Veselin Stoyanov.
+1. **[XLM-RoBERTa-XL](https://huggingface.co/docs/transformers/model_doc/xlm-roberta-xl)** (from Facebook AI) released with the paper [Larger-Scale Transformers for Multilingual Masked Language Modeling](https://arxiv.org/abs/2105.00572) by Naman Goyal, Jingfei Du, Myle Ott, Giri Anantharaman, Alexis Conneau.
+1. **[XLNet](https://huggingface.co/docs/transformers/model_doc/xlnet)** (from Google/CMU) released with the paper [​XLNet: Generalized Autoregressive Pretraining for Language Understanding](https://arxiv.org/abs/1906.08237) by Zhilin Yang*, Zihang Dai*, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, Quoc V. Le.
+1. **[XLS-R](https://huggingface.co/docs/transformers/model_doc/xls_r)** (from Facebook AI) released with the paper [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://arxiv.org/abs/2111.09296) by Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, Michael Auli.
+1. **[XLSR-Wav2Vec2](https://huggingface.co/docs/transformers/model_doc/xlsr_wav2vec2)** (from Facebook AI) released with the paper [Unsupervised Cross-Lingual Representation Learning For Speech Recognition](https://arxiv.org/abs/2006.13979) by Alexis Conneau, Alexei Baevski, Ronan Collobert, Abdelrahman Mohamed, Michael Auli.
+1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (from Huazhong University of Science & Technology) released with the paper [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666) by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
+1. **[YOSO](https://huggingface.co/docs/transformers/model_doc/yoso)** (from the University of Wisconsin - Madison) released with the paper [You Only Sample (Almost) by Zhanpeng Zeng, Yunyang Xiong, Sathya N. Ravi, Shailesh Acharya, Glenn Fung, Vikas Singh.
+1. 想要貢獻新的模型？我們這裡有一份**詳細指引和模板**來引導你加入新的模型。你可以在 [`templates`](./templates) 目錄中找到它們。記得查看[貢獻指引](./CONTRIBUTING.md)並在開始寫 PR 前聯繫維護人員或開一個新的 issue 來獲得 feedbacks。
+
+要檢查某個模型是否已有 Flax、PyTorch 或 TensorFlow 的實作，或其是否在🤗 Tokenizers 函式庫中有對應的 tokenizer，敬請參閱[此表](https://huggingface.co/docs/transformers/index#supported-frameworks)。
+
+這些實作均已於多個資料集測試（請參閱範例腳本）並應與原版實作表現相當。你可以在範例文件的[此節](https://huggingface.co/docs/transformers/examples)中了解實作的細節。
+
+
+## 了解更多
+
+| 章節 | 描述 |
+|-|-|
+| [文件](https://huggingface.co/transformers/) | 完整的 API 文件和教學 |
+| [任務概覽](https://huggingface.co/docs/transformers/task_summary) | 🤗 Transformers 支援的任務 |
+| [預處理教學](https://huggingface.co/docs/transformers/preprocessing) | 使用 `Tokenizer` 來為模型準備資料 |
+| [訓練和微調](https://huggingface.co/docs/transformers/training) | 使用 PyTorch/TensorFlow 的內建的訓練方式或於 `Trainer` API 中使用 🤗 Transformers 提供的模型 |
+| [快速上手：微調和範例腳本](https://github.com/huggingface/transformers/tree/main/examples) | 為各種任務提供的範例腳本 |
+| [模型分享和上傳](https://huggingface.co/docs/transformers/model_sharing) | 上傳並與社群分享你微調的模型 |
+| [遷移](https://huggingface.co/docs/transformers/migration) | 從 `pytorch-transformers` 或 `pytorch-pretrained-bert` 遷移到 🤗 Transformers |
+
+## 引用
+
+我們已將此函式庫的[論文](https://www.aclweb.org/anthology/2020.emnlp-demos.6/)正式發表。如果你使用了 🤗 Transformers 函式庫，可以引用：
+```bibtex
+@inproceedings{wolf-etal-2020-transformers,
+    title = "Transformers: State-of-the-Art Natural Language Processing",
+    author = "Thomas Wolf and Lysandre Debut and Victor Sanh and Julien Chaumond and Clement Delangue and Anthony Moi and Pierric Cistac and Tim Rault and Rémi Louf and Morgan Funtowicz and Joe Davison and Sam Shleifer and Patrick von Platen and Clara Ma and Yacine Jernite and Julien Plu and Canwen Xu and Teven Le Scao and Sylvain Gugger and Mariama Drame and Quentin Lhoest and Alexander M. Rush",
+    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",
+    month = oct,
+    year = "2020",
+    address = "Online",
+    publisher = "Association for Computational Linguistics",
+    url = "https://www.aclweb.org/anthology/2020.emnlp-demos.6",
+    pages = "38--45"
+}
+```

SECURITY.md

@@ -1,40 +0,0 @@
-# Security Policy
-
-## Hugging Face Hub, remote artefacts, and remote code
-
-Transformers is open-source software that is tightly coupled to the Hugging Face Hub. While you have the ability to use it
-offline with pre-downloaded model weights, it provides a very simple way to download, use, and manage models locally.
-
-When downloading artefacts that have been uploaded by others on any platform, you expose yourself to risks. Please
-read below for the security recommendations in order to keep your runtime and local environment safe.
-
-### Remote artefacts
-
-Models uploaded on the Hugging Face Hub come in different formats. We heavily recommend uploading and downloading
-models in the [`safetensors`](https://github.com/huggingface/safetensors) format (which is the default prioritized
-by the transformers library), as developed specifically to prevent arbitrary code execution on your system.
-
-To avoid loading models from unsafe formats(e.g. [pickle](https://docs.python.org/3/library/pickle.html), you should use the `use_safetensors` parameter. If doing so, in the event that no .safetensors file is present, transformers will error when loading the model.
-
-### Remote code
-
-#### Modeling
-
-Transformers supports many model architectures, but is also the bridge between your Python runtime and models that
-are stored in model repositories on the Hugging Face Hub.
-
-These models require the `trust_remote_code=True` parameter to be set when using them; please **always** verify
-the content of the modeling files when using this argument. We recommend setting a revision in order to ensure you
-protect yourself from updates on the repository.
-
-#### Tools
-
-Through the `Agent` framework, remote tools can be downloaded to be used by the Agent. You're to specify these tools
-yourself, but please keep in mind that their code will be run on your machine if the Agent chooses to run them.
-
-Please inspect the code of the tools before passing them to the Agent to protect your runtime and local setup.
-
-## Reporting a Vulnerability
-
-🤗 Please feel free to submit vulnerability reports to our private bug bounty program at https://hackerone.com/hugging_face. You'll need to request access to the program by emailing security@huggingface.co.
-Note that you'll need to be invited to our program, so send us a quick email at security@huggingface.co if you've found a vulnerability.

awesome-transformers.md

@@ -1,609 +0,0 @@
-# Awesome projects built with Transformers
-
-This page lists awesome projects built on top of Transformers. Transformers is more than a toolkit to use pretrained
-models: it's a community of projects built around it and the Hugging Face Hub. We want Transformers to enable
-developers, researchers, students, professors, engineers, and anyone else to build their dream projects.
-
-In this list, we showcase incredibly impactful and novel projects that have pushed the field forward. We celebrate
-100 of these projects as we reach the milestone of 100k stars as a community; but we're very open to pull requests
-adding other projects to the list. If you believe a project should be here and it's not, then please, open a PR 
-to add it.
-
-## [gpt4all](https://github.com/nomic-ai/gpt4all)
-
-[gpt4all](https://github.com/nomic-ai/gpt4all) is an ecosystem of open-source chatbots trained on massive collections of clean assistant data including code, stories and dialogue. It offers open-source, large language models such as LLaMA and GPT-J trained in an assistant-style.
-
-Keywords: Open-source, LLaMa, GPT-J, instruction, assistant
-
-## [recommenders](https://github.com/microsoft/recommenders)
-
-This repository contains examples and best practices for building recommendation systems, provided as Jupyter notebooks. It goes over several aspects required to build efficient recommendation systems: data preparation, modeling, evaluation, model selection & optimization, as well as operationalization
-
-Keywords: Recommender systems, AzureML
-
-## [IOPaint](https://github.com/Sanster/IOPaint)
-
-Image inpainting tool powered by Stable Diffusion. Remove any unwanted object, defect, people from your pictures or erase and replace anything on your pictures.
-
-Keywords: inpainting, SD, Stable Diffusion
-
-## [flair](https://github.com/flairNLP/flair)
-
-FLAIR is a powerful PyTorch NLP framework, convering several important tasks: NER, sentiment-analysis, part-of-speech tagging, text and document embeddings, among other things.
-
-Keywords: NLP, text embedding, document embedding, biomedical, NER, PoS, sentiment-analysis
-
-## [mindsdb](https://github.com/mindsdb/mindsdb)
-
-MindsDB is a low-code ML platform, which automates and integrates several ML frameworks into the data stack as "AI Tables" to streamline the integration of AI into applications, making it accessible to developers of all skill levels.
-
-Keywords: Database, low-code, AI table
-
-## [langchain](https://github.com/hwchase17/langchain)
-
-[langchain](https://github.com/hwchase17/langchain) is aimed at assisting in the development of apps merging both LLMs and other sources of knowledge. The library allows chaining calls to applications, creating a sequence across many tools.
-
-Keywords: LLMs, Large Language Models, Agents, Chains
-
-## [LlamaIndex](https://github.com/jerryjliu/llama_index)
-
-[LlamaIndex](https://github.com/jerryjliu/llama_index) is a project that provides a central interface to connect your LLM's with external data. It provides various kinds of indices and retreival mechanisms to perform different LLM tasks and obtain knowledge-augmented results.
-
-Keywords: LLMs, Large Language Models, Data Retrieval, Indices, Knowledge Augmentation 
-
-## [ParlAI](https://github.com/facebookresearch/ParlAI)
-
-[ParlAI](https://github.com/facebookresearch/ParlAI) is a python framework for sharing, training and testing dialogue models, from open-domain chitchat, to task-oriented dialogue, to visual question answering. It provides more than 100 datasets under the same API, a large zoo of pretrained models, a set of agents, and has several integrations.
-
-Keywords: Dialogue, Chatbots, VQA, Datasets, Agents
-
-## [sentence-transformers](https://github.com/UKPLab/sentence-transformers)
-
-This framework provides an easy method to compute dense vector representations for sentences, paragraphs, and images. The models are based on transformer networks like BERT / RoBERTa / XLM-RoBERTa etc. and achieve state-of-the-art performance in various task. Text is embedding in vector space such that similar text is close and can efficiently be found using cosine similarity.
-
-Keywords: Dense vector representations, Text embeddings, Sentence embeddings
-
-## [ludwig](https://github.com/ludwig-ai/ludwig)
-
-Ludwig is a declarative machine learning framework that makes it easy to define machine learning pipelines using a simple and flexible data-driven configuration system. Ludwig is targeted at a wide variety of AI tasks. It provides a data-driven configuration system, training, prediction, and evaluation scripts, as well as a programmatic API.
-
-Keywords: Declarative, Data-driven, ML Framework
-
-## [InvokeAI](https://github.com/invoke-ai/InvokeAI)
-
-[InvokeAI](https://github.com/invoke-ai/InvokeAI) is an engine for Stable Diffusion models, aimed at professionals, artists, and enthusiasts. It leverages the latest AI-driven technologies through CLI as well as a WebUI.
-
-Keywords: Stable-Diffusion, WebUI, CLI
-
-## [PaddleNLP](https://github.com/PaddlePaddle/PaddleNLP)
-
-[PaddleNLP](https://github.com/PaddlePaddle/PaddleNLP) is an easy-to-use and powerful NLP library particularly targeted at the Chinese languages. It has support for multiple pre-trained model zoos, and supports a wide-range of NLP tasks from research to industrial applications.
-
-Keywords: NLP, Chinese, Research, Industry
-
-## [stanza](https://github.com/stanfordnlp/stanza)
-
-The Stanford NLP Group's official Python NLP library. It contains support for running various accurate natural language processing tools on 60+ languages and for accessing the Java Stanford CoreNLP software from Python.
-
-Keywords: NLP, Multilingual, CoreNLP
-
-## [DeepPavlov](https://github.com/deeppavlov/DeepPavlov)
-
-[DeepPavlov](https://github.com/deeppavlov/DeepPavlov) is an open-source conversational AI library. It is designed for the development of production ready chat-bots and complex conversational systems, as well as research in the area of NLP and, particularly, of dialog systems.
-
-Keywords: Conversational, Chatbot, Dialog
-
-## [alpaca-lora](https://github.com/tloen/alpaca-lora)
-
-Alpaca-lora contains code for reproducing the Stanford Alpaca results using low-rank adaptation (LoRA). The repository provides training (fine-tuning) as well as generation scripts.
-
-Keywords: LoRA, Parameter-efficient fine-tuning
-
-## [imagen-pytorch](https://github.com/lucidrains/imagen-pytorch)
-
-An open-source Implementation of Imagen, Google's closed-source Text-to-Image Neural Network that beats DALL-E2. As of release, it is the new SOTA for text-to-image synthesis.
-
-Keywords: Imagen, Text-to-image
-
-## [adapters](https://github.com/adapter-hub/adapters)
-
-[adapters](https://github.com/adapter-hub/adapters) is an extension of HuggingFace's Transformers library, integrating adapters into state-of-the-art language models by incorporating AdapterHub, a central repository for pre-trained adapter modules. It is a drop-in replacement for transformers, which is regularly updated to stay up-to-date with the developments of transformers.
-
-Keywords: Adapters, LoRA, Parameter-efficient fine-tuning, Hub
-
-## [NeMo](https://github.com/NVIDIA/NeMo)
-
-NVIDIA [NeMo](https://github.com/NVIDIA/NeMo) is a conversational AI toolkit built for researchers working on automatic speech recognition (ASR), text-to-speech synthesis (TTS), large language models (LLMs), and natural language processing (NLP). The primary objective of [NeMo](https://github.com/NVIDIA/NeMo) is to help researchers from industry and academia to reuse prior work (code and pretrained models) and make it easier to create new https://developer.nvidia.com/conversational-ai#started.
-
-Keywords: Conversational, ASR, TTS, LLMs, NLP
-
-## [Runhouse](https://github.com/run-house/runhouse)
-
-[Runhouse](https://github.com/run-house/runhouse) allows to send code and data to any of your compute or data infra, all in Python, and continue to interact with them normally from your existing code and environment. Runhouse developers mention:
-
-> Think of it as an expansion pack to your Python interpreter that lets it take detours to remote machines or manipulate remote data.
-
-Keywords: MLOps, Infrastructure, Data storage, Modeling
-
-## [MONAI](https://github.com/Project-MONAI/MONAI)
-
-[MONAI](https://github.com/Project-MONAI/MONAI) is a PyTorch-based, open-source framework for deep learning in healthcare imaging, part of PyTorch Ecosystem. Its ambitions are:
-- developing a community of academic, industrial and clinical researchers collaborating on a common foundation;
-- creating state-of-the-art, end-to-end training workflows for healthcare imaging;
-- providing researchers with the optimized and standardized way to create and evaluate deep learning models.
-
-Keywords: Healthcare imaging, Training, Evaluation
-
-## [simpletransformers](https://github.com/ThilinaRajapakse/simpletransformers)
-
-Simple Transformers lets you quickly train and evaluate Transformer models. Only 3 lines of code are needed to initialize, train, and evaluate a model. It supports a wide variety of NLP tasks.
-
-Keywords: Framework, simplicity, NLP
-
-## [JARVIS](https://github.com/microsoft/JARVIS)
-
-[JARVIS](https://github.com/microsoft/JARVIS) is a system attempting to merge LLMs such as GPT-4 with the rest of the open-source ML community: leveraging up to 60 downstream models in order to perform tasks identified by the LLM.
-
-Keywords: LLM, Agents, HF Hub
-
-## [transformers.js](https://xenova.github.io/transformers.js/)
-
-[transformers.js](https://xenova.github.io/transformers.js/) is a JavaScript library targeted at running models from transformers directly within the browser.
-
-Keywords: Transformers, JavaScript, browser
-
-## [bumblebee](https://github.com/elixir-nx/bumblebee)
-
-Bumblebee provides pre-trained Neural Network models on top of Axon, a neural networks library for the Elixir language. It includes integration with 🤗 Models, allowing anyone to download and perform Machine Learning tasks with few lines of code.
-
-Keywords: Elixir, Axon
-
-## [argilla](https://github.com/argilla-io/argilla)
-
-Argilla is an open-source platform providing advanced NLP labeling, monitoring, and workspaces. It is compatible with many open source ecosystems such as Hugging Face, Stanza, FLAIR, and others.
-
-Keywords: NLP, Labeling, Monitoring, Workspaces
-
-## [haystack](https://github.com/deepset-ai/haystack)
-
-Haystack is an open source NLP framework to interact with your data using Transformer models and LLMs. It offers production-ready tools to quickly build complex decision making, question answering, semantic search, text generation applications, and more.
-
-Keywords: NLP, Framework, LLM
-
-## [spaCy](https://github.com/explosion/spaCy)
-
-[spaCy](https://github.com/explosion/spaCy) is a library for advanced Natural Language Processing in Python and Cython. It's built on the very latest research, and was designed from day one to be used in real products. It offers support for transformers models through its third party package, spacy-transformers.
-
-Keywords: NLP, Framework
-
-## [speechbrain](https://github.com/speechbrain/speechbrain)
-
-SpeechBrain is an open-source and all-in-one conversational AI toolkit based on PyTorch.
-The goal is to create a single, flexible, and user-friendly toolkit that can be used to easily develop state-of-the-art speech technologies, including systems for speech recognition, speaker recognition, speech enhancement, speech separation, language identification, multi-microphone signal processing, and many others.
-
-Keywords: Conversational, Speech
-
-## [skorch](https://github.com/skorch-dev/skorch)
-
-Skorch is a scikit-learn compatible neural network library that wraps PyTorch. It has support for models within transformers, and tokenizers from tokenizers.
-
-Keywords: Scikit-Learn, PyTorch
-
-## [bertviz](https://github.com/jessevig/bertviz)
-
-BertViz is an interactive tool for visualizing attention in Transformer language models such as BERT, GPT2, or T5. It can be run inside a Jupyter or Colab notebook through a simple Python API that supports most Huggingface models.
-
-Keywords: Visualization, Transformers
-
-## [mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax)
-
-[mesh-transformer-jax](https://github.com/kingoflolz/mesh-transformer-jax) is a haiku library using the xmap/pjit operators in JAX for model parallelism of transformers. This library is designed for scalability up to approximately 40B parameters on TPUv3s. It was the library used to train the GPT-J model.
-
-Keywords: Haiku, Model parallelism, LLM, TPU
-
-## [deepchem](https://github.com/deepchem/deepchem)
-
-DeepChem aims to provide a high quality open-source toolchain that democratizes the use of deep-learning in drug discovery, materials science, quantum chemistry, and biology.
-
-Keywords: Drug discovery, Materials Science, Quantum Chemistry, Biology
-
-## [OpenNRE](https://github.com/thunlp/OpenNRE)
-
-An Open-Source Package for Neural Relation Extraction (NRE). It is targeted at a wide range of users, from newcomers to relation extraction, to developers, researchers, or students.
-
-Keywords: Neural Relation Extraction, Framework
-
-## [pycorrector](https://github.com/shibing624/pycorrector)
-
-PyCorrector is a Chinese Text Error Correction Tool. It uses a language model to detect errors, pinyin feature and shape feature to correct Chinese text errors. it can be used for Chinese Pinyin and stroke input method.
-
-Keywords: Chinese, Error correction tool, Language model, Pinyin
-
-## [nlpaug](https://github.com/makcedward/nlpaug)
-
-This python library helps you with augmenting nlp for machine learning projects. It is a lightweight library featuring synthetic data generation for improving model performance, support for audio and text, and compatibility with several ecosystems (scikit-learn, pytorch, tensorflow).
-
-Keywords: Data augmentation, Synthetic data generation, Audio, NLP
-
-## [dream-textures](https://github.com/carson-katri/dream-textures)
-
-[dream-textures](https://github.com/carson-katri/dream-textures) is a library targeted at bringing stable-diffusion support within Blender. It supports several use-cases, such as image generation, texture projection, inpainting/outpainting, ControlNet, and upscaling.
-
-Keywords: Stable-Diffusion, Blender
-
-## [seldon-core](https://github.com/SeldonIO/seldon-core)
-
-Seldon core converts your ML models (Tensorflow, Pytorch, H2o, etc.) or language wrappers (Python, Java, etc.) into production REST/GRPC microservices.
-Seldon handles scaling to thousands of production machine learning models and provides advanced machine learning capabilities out of the box including Advanced Metrics, Request Logging, Explainers, Outlier Detectors, A/B Tests, Canaries and more.
-
-Keywords: Microservices, Modeling, Language wrappers
-
-## [open_model_zoo](https://github.com/openvinotoolkit/open_model_zoo)
-
-This repository includes optimized deep learning models and a set of demos to expedite development of high-performance deep learning inference applications. Use these free pre-trained models instead of training your own models to speed-up the development and production deployment process.
-
-Keywords: Optimized models, Demos
-
-## [ml-stable-diffusion](https://github.com/apple/ml-stable-diffusion)
-
-ML-Stable-Diffusion is a repository by Apple bringing Stable Diffusion support to Core ML, on Apple Silicon devices. It supports stable diffusion checkpoints hosted on the Hugging Face Hub.
-
-Keywords: Stable Diffusion, Apple Silicon, Core ML
-
-## [stable-dreamfusion](https://github.com/ashawkey/stable-dreamfusion)
-
-Stable-Dreamfusion is a pytorch implementation of the text-to-3D model Dreamfusion, powered by the Stable Diffusion text-to-2D model.
-
-Keywords: Text-to-3D, Stable Diffusion
-
-## [txtai](https://github.com/neuml/txtai)
- 
-[txtai](https://github.com/neuml/txtai) is an open-source platform for semantic search and workflows powered by language models. txtai builds embeddings databases, which are a union of vector indexes and relational databases enabling similarity search with SQL. Semantic workflows connect language models together into unified applications.
-
-Keywords: Semantic search, LLM
-
-## [djl](https://github.com/deepjavalibrary/djl)
-
-Deep Java Library (DJL) is an open-source, high-level, engine-agnostic Java framework for deep learning. DJL is designed to be easy to get started with and simple to use for developers. DJL provides a native Java development experience and functions like any other regular Java library. DJL offers [a Java binding](https://github.com/deepjavalibrary/djl/tree/master/extensions/tokenizers) for HuggingFace Tokenizers and easy conversion toolkit for HuggingFace model to deploy in Java.
-
-Keywords: Java, Framework
-
-## [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness/)
-
-This project provides a unified framework to test generative language models on a large number of different evaluation tasks. It has support for more than 200 tasks, and supports different ecosystems: HF Transformers, GPT-NeoX, DeepSpeed, as well as the OpenAI API.
-
-Keywords: LLM, Evaluation, Few-shot
-
-## [gpt-neox](https://github.com/EleutherAI/gpt-neox)
-
-This repository records EleutherAI's library for training large-scale language models on GPUs. The framework is based on NVIDIA's Megatron Language Model and has been augmented with techniques from DeepSpeed as well as some novel optimizations. It is focused on training multi-billion-parameter models.
-
-Keywords: Training, LLM, Megatron, DeepSpeed
-
-## [muzic](https://github.com/microsoft/muzic)
-
-Muzic is a research project on AI music that empowers music understanding and generation with deep learning and artificial intelligence. Muzic was created by researchers from Microsoft Research Asia.
-
-Keywords: Music understanding, Music generation
-
-## [dalle-flow](https://github.com/jina-ai/dalle-flow)
-
-DALL·E Flow is an interactive workflow for generating high-definition images from a text prompt. Itt leverages DALL·E-Mega, GLID-3 XL, and Stable Diffusion to generate image candidates, and then calls CLIP-as-service to rank the candidates w.r.t. the prompt.
-The preferred candidate is fed to GLID-3 XL for diffusion, which often enriches the texture and background. Finally, the candidate is upscaled to 1024x1024 via SwinIR.
-
-Keywords: High-definition image generation, Stable Diffusion, DALL-E Mega, GLID-3 XL, CLIP, SwinIR
-
-## [lightseq](https://github.com/bytedance/lightseq)
-
-LightSeq is a high performance training and inference library for sequence processing and generation implemented in CUDA. It enables highly efficient computation of modern NLP and CV models such as BERT, GPT, Transformer, etc. It is therefore best useful for machine translation, text generation, image classification, and other sequence related tasks.
-
-Keywords: Training, Inference, Sequence Processing, Sequence Generation
-
-## [LaTeX-OCR](https://github.com/lukas-blecher/LaTeX-OCR)
-
-The goal of this project is to create a learning based system that takes an image of a math formula and returns corresponding LaTeX code.
-
-Keywords: OCR, LaTeX, Math formula
-
-## [open_clip](https://github.com/mlfoundations/open_clip)
-
-OpenCLIP is an open source implementation of OpenAI's CLIP.
-
-The goal of this repository is to enable training models with contrastive image-text supervision, and to investigate their properties such as robustness to distribution shift. 
-The starting point is an implementation of CLIP that matches the accuracy of the original CLIP models when trained on the same dataset. 
-
-Specifically, a ResNet-50 model trained with this codebase on OpenAI's 15 million image subset of YFCC achieves 32.7% top-1 accuracy on ImageNet.
-
-Keywords: CLIP, Open-source, Contrastive, Image-text
-
-## [dalle-playground](https://github.com/saharmor/dalle-playground)
-
-A playground to generate images from any text prompt using Stable Diffusion and Dall-E mini.
-
-Keywords: WebUI, Stable Diffusion, Dall-E mini
-
-## [FedML](https://github.com/FedML-AI/FedML)
-
-[FedML](https://github.com/FedML-AI/FedML) is a federated learning and analytics library enabling secure and collaborative machine learning on decentralized data anywhere at any scale.
-
-It supports large-scale cross-silo federated learning, and cross-device federated learning on smartphones/IoTs, and research simulation.
-
-Keywords: Federated Learning, Analytics, Collaborative ML, Decentralized
-
-## [gpt-code-clippy](https://github.com/CodedotAl/gpt-code-clippy)
-
-GPT-Code-Clippy (GPT-CC) is an open source version of GitHub Copilot, a language model -- based on GPT-3, called GPT-Codex -- that is fine-tuned on publicly available code from GitHub.
-
-Keywords: LLM, Code
-
-## [TextAttack](https://github.com/QData/TextAttack)
-
-[TextAttack](https://github.com/QData/TextAttack) 🐙 is a Python framework for adversarial attacks, data augmentation, and model training in NLP.
-
-Keywords: Adversarial attacks, Data augmentation, NLP
-
-## [OpenPrompt](https://github.com/thunlp/OpenPrompt)
-
-Prompt-learning is a paradigm to adapt pre-trained language models (PLMs) to downstream NLP tasks, which modify the input text with a textual template and directly uses PLMs to conduct pre-trained tasks. This library provides a standard, flexible and extensible framework to deploy the prompt-learning pipeline. [OpenPrompt](https://github.com/thunlp/OpenPrompt) supports loading PLMs directly from https://github.com/huggingface/transformers.
-
-## [text-generation-webui](https://github.com/oobabooga/text-generation-webui/)
-
-[text-generation-webui](https://github.com/oobabooga/text-generation-webui/) is a Gradio Web UI for running Large Language Models like LLaMA, llama.cpp, GPT-J, Pythia, OPT, and GALACTICA.
-
-Keywords: LLM, WebUI
-
-## [libra](https://github.com/Palashio/libra)
-
-An ergonomic machine learning [libra](https://github.com/Palashio/libra)ry for non-technical users. It focuses on ergonomics and on ensuring that training a model is as simple as it can be.
-
-Keywords: Ergonomic, Non-technical
-
-## [alibi](https://github.com/SeldonIO/alibi)
-
-Alibi is an open source Python library aimed at machine learning model inspection and interpretation. The focus of the library is to provide high-quality implementations of black-box, white-box, local and global explanation methods for classification and regression models.
-
-Keywords: Model inspection, Model interpretation, Black-box, White-box
-
-## [tortoise-tts](https://github.com/neonbjb/tortoise-tts)
-
-Tortoise is a text-to-speech program built with the following priorities: strong multi-voice capabilities, and highly realistic prosody and intonation.
-
-Keywords: Text-to-speech
-
-## [flower](https://github.com/adap/flower)
-
-Flower (flwr) is a framework for building federated learning systems. The design of Flower is based on a few guiding principles: customizability, extendability, framework agnosticity, and ease-of-use.
-
-Keywords: Federated learning systems, Customizable, Extendable, Framework-agnostic, Simplicity
-
-## [fast-bert](https://github.com/utterworks/fast-bert)
-
-Fast-Bert is a deep learning library that allows developers and data scientists to train and deploy BERT and XLNet based models for natural language processing tasks beginning with Text Classification. It is aimed at simplicity.
-
-Keywords: Deployment, BERT, XLNet
-
-## [towhee](https://github.com/towhee-io/towhee)
-
-Towhee makes it easy to build neural data processing pipelines for AI applications. We provide hundreds of models, algorithms, and transformations that can be used as standard pipeline building blocks. Users can use Towhee's Pythonic API to build a prototype of their pipeline and automatically optimize it for production-ready environments.
-
-Keywords: Data processing pipeline, Optimization
-
-## [alibi-detect](https://github.com/SeldonIO/alibi-detect)
-
-Alibi Detect is an open source Python library focused on outlier, adversarial and drift detection. The package aims to cover both online and offline detectors for tabular data, text, images and time series. Both TensorFlow and PyTorch backends are supported for drift detection.
-
-Keywords: Adversarial, Outlier, Drift detection
-
-## [FARM](https://github.com/deepset-ai/FARM)
-
-[FARM](https://github.com/deepset-ai/FARM) makes Transfer Learning with BERT & Co simple, fast and enterprise-ready. It's built upon transformers and provides additional features to simplify the life of developers: Parallelized preprocessing, highly modular design, multi-task learning, experiment tracking, easy debugging and close integration with AWS SageMaker.
-
-Keywords: Transfer Learning, Modular design, Multi-task learning, Experiment tracking
-
-## [aitextgen](https://github.com/minimaxir/aitextgen)
-
-A robust Python tool for text-based AI training and generation using OpenAI's GPT-2 and EleutherAI's GPT Neo/GPT-3 architecture.
-[aitextgen](https://github.com/minimaxir/aitextgen) is a Python package that leverages PyTorch, Hugging Face Transformers and pytorch-lightning with specific optimizations for text generation using GPT-2, plus many added features.
-
-Keywords: Training, Generation
-
-## [diffgram](https://github.com/diffgram/diffgram)
-
-Diffgram aims to integrate human supervision into platforms. We support your team programmatically changing the UI (Schema, layout, etc.) like in Streamlit. This means that you can collect and annotate timely data from users. In other words, we are the platform behind your platform, an integrated part of your application, to ship new & better AI products faster.
-
-Keywords: Human supervision, Platform
-
-## [ecco](https://github.com/jalammar/ecco)
-
-Explain, analyze, and visualize NLP language models. Ecco creates interactive visualizations directly in Jupyter notebooks explaining the behavior of Transformer-based language models (like GPT2, BERT, RoBERTA, T5, and T0).
-
-Keywords: Model explainability
-
-## [s3prl](https://github.com/s3prl/s3prl)
-
-[s3prl](https://github.com/s3prl/s3prl) stands for Self-Supervised Speech Pre-training and Representation Learning. Self-supervised speech pre-trained models are called upstream in this toolkit, and are utilized in various downstream tasks.
-
-Keywords: Speech, Training
-
-## [ru-dalle](https://github.com/ai-forever/ru-dalle)
-
-RuDALL-E aims to be similar to DALL-E, targeted to Russian.
-
-Keywords: DALL-E, Russian
-
-## [DeepKE](https://github.com/zjunlp/DeepKE)
-
-[DeepKE](https://github.com/zjunlp/DeepKE) is a knowledge extraction toolkit for knowledge graph construction supporting cnSchema，low-resource, document-level and multimodal scenarios for entity, relation and attribute extraction.
-
-Keywords: Knowledge Extraction, Knowledge Graphs
-
-## [Nebuly](https://github.com/nebuly-ai/nebuly)
-
-Nebuly is the next-generation platform to monitor and optimize your AI costs in one place. The platform connects to all your AI cost sources (compute, API providers, AI software licenses, etc) and centralizes them in one place to give you full visibility on a model basis. The platform also provides optimization recommendations and a co-pilot model that can guide during the optimization process. The platform builds on top of the open-source tools allowing you to optimize the different steps of your AI stack to squeeze out the best possible cost performances.
-
-Keywords: Optimization, Performance, Monitoring
-
-## [imaginAIry](https://github.com/brycedrennan/imaginAIry)
-
-Offers a CLI and a Python API to generate images with Stable Diffusion. It has support for many tools, like image structure control (controlnet), instruction-based image edits (InstructPix2Pix), prompt-based masking (clipseg), among others.
-
-Keywords: Stable Diffusion, CLI, Python API
-
-## [sparseml](https://github.com/neuralmagic/sparseml)
-
-SparseML is an open-source model optimization toolkit that enables you to create inference-optimized sparse models using pruning, quantization, and distillation algorithms. Models optimized with SparseML can then be exported to the ONNX and deployed with DeepSparse for GPU-class performance on CPU hardware.
-
-Keywords: Model optimization, Pruning, Quantization, Distillation
-
-## [opacus](https://github.com/pytorch/opacus)
-
-Opacus is a library that enables training PyTorch models with differential privacy. It supports training with minimal code changes required on the client, has little impact on training performance, and allows the client to online track the privacy budget expended at any given moment.
-
-Keywords: Differential privacy
-
-## [LAVIS](https://github.com/salesforce/LAVIS)
-
-[LAVIS](https://github.com/salesforce/LAVIS) is a Python deep learning library for LAnguage-and-VISion intelligence research and applications. This library aims to provide engineers and researchers with a one-stop solution to rapidly develop models for their specific multimodal scenarios, and benchmark them across standard and customized datasets. It features a unified interface design to access
-
-Keywords: Multimodal, NLP, Vision
-
-## [buzz](https://github.com/chidiwilliams/buzz)
-
-Buzz transcribes and translates audio offline on your personal computer. Powered by OpenAI's Whisper.
-
-Keywords: Audio transcription, Translation
-
-## [rust-bert](https://github.com/guillaume-be/rust-bert)
-
-Rust-native state-of-the-art Natural Language Processing models and pipelines. Port of Hugging Face's Transformers library, using the tch-rs crate and pre-processing from rust-tokenizers. Supports multi-threaded tokenization and GPU inference. This repository exposes the model base architecture, task-specific heads and ready-to-use pipelines.
-
-Keywords: Rust, BERT, Inference
-
-## [EasyNLP](https://github.com/alibaba/EasyNLP)
-
-[EasyNLP](https://github.com/alibaba/EasyNLP) is an easy-to-use NLP development and application toolkit in PyTorch, first released inside Alibaba in 2021. It is built with scalable distributed training strategies and supports a comprehensive suite of NLP algorithms for various NLP applications. [EasyNLP](https://github.com/alibaba/EasyNLP) integrates knowledge distillation and few-shot learning for landing large pre-trained models, together with various popular multi-modality pre-trained models. It provides a unified framework of model training, inference, and deployment for real-world applications.
-
-Keywords: NLP, Knowledge distillation, Few-shot learning, Multi-modality, Training, Inference, Deployment
-
-## [TurboTransformers](https://github.com/Tencent/TurboTransformers)
-
-A fast and user-friendly runtime for transformer inference (Bert, Albert, GPT2, Decoders, etc) on CPU and GPU.
-
-Keywords: Optimization, Performance
-
-## [hivemind](https://github.com/learning-at-home/hivemind)
-
-Hivemind is a PyTorch library for decentralized deep learning across the Internet. Its intended usage is training one large model on hundreds of computers from different universities, companies, and volunteers.
-
-Keywords: Decentralized training
-
-## [docquery](https://github.com/impira/docquery)
-
-DocQuery is a library and command-line tool that makes it easy to analyze semi-structured and unstructured documents (PDFs, scanned images, etc.) using large language models (LLMs). You simply point DocQuery at one or more documents and specify a question you want to ask. DocQuery is created by the team at Impira.
-
-Keywords: Semi-structured documents, Unstructured documents, LLM, Document Question Answering
-
-## [CodeGeeX](https://github.com/THUDM/CodeGeeX)
-
-[CodeGeeX](https://github.com/THUDM/CodeGeeX) is a large-scale multilingual code generation model with 13 billion parameters, pre-trained on a large code corpus of more than 20 programming languages. It has several unique features:
-- Multilingual code generation
-- Crosslingual code translation
-- Is a customizable programming assistant
-
-Keywords: Code Generation Model
-
-## [ktrain](https://github.com/amaiya/ktrain)
-
-[ktrain](https://github.com/amaiya/ktrain) is a lightweight wrapper for the deep learning library TensorFlow Keras (and other libraries) to help build, train, and deploy neural networks and other machine learning models. Inspired by ML framework extensions like fastai and ludwig, [ktrain](https://github.com/amaiya/ktrain) is designed to make deep learning and AI more accessible and easier to apply for both newcomers and experienced practitioners.
-
-Keywords: Keras wrapper, Model building, Training, Deployment
-
-## [FastDeploy](https://github.com/PaddlePaddle/FastDeploy)
-
-[FastDeploy](https://github.com/PaddlePaddle/FastDeploy) is an Easy-to-use and High Performance AI model deployment toolkit for Cloud, Mobile and Edge with packageout-of-the-box and unified experience, endend-to-end optimization for over fire160+ Text, Vision, Speech and Cross-modal AI models. Including image classification, object detection, OCR, face detection, matting, pp-tracking, NLP, stable diffusion, TTS and other tasks to meet developers' industrial deployment needs for multi-scenario, multi-hardware and multi-platform.
-
-Keywords: Model deployment, CLoud, Mobile, Edge
-
-## [underthesea](https://github.com/undertheseanlp/underthesea)
-
-[underthesea](https://github.com/undertheseanlp/underthesea) is a Vietnamese NLP toolkit. Underthesea is a suite of open source Python modules data sets and tutorials supporting research and development in Vietnamese Natural Language Processing. We provides extremely easy API to quickly apply pretrained NLP models to your Vietnamese text, such as word segmentation, part-of-speech tagging (PoS), named entity recognition (NER), text classification and dependency parsing.
-
-Keywords: Vietnamese, NLP
-
-## [hasktorch](https://github.com/hasktorch/hasktorch)
-
-Hasktorch is a library for tensors and neural networks in Haskell. It is an independent open source community project which leverages the core C++ libraries shared by PyTorch.
-
-Keywords: Haskell, Neural Networks
-
-## [donut](https://github.com/clovaai/donut)
-
-Donut, or Document understanding transformer, is a new method of document understanding that utilizes an OCR-free end-to-end Transformer model.
-
-Donut does not require off-the-shelf OCR engines/APIs, yet it shows state-of-the-art performances on various visual document understanding tasks, such as visual document classification or information extraction (a.k.a. document parsing).
-
-Keywords: Document Understanding
-
-## [transformers-interpret](https://github.com/cdpierse/transformers-interpret)
-
-Transformers Interpret is a model explainability tool designed to work exclusively with the transformers package.
-
-In line with the philosophy of the Transformers package Transformers Interpret allows any transformers model to be explained in just two lines. Explainers are available for both text and computer vision models. Visualizations are also available in notebooks and as savable png and html files
-
-Keywords: Model interpretation, Visualization
-
-## [mlrun](https://github.com/mlrun/mlrun)
-
-MLRun is an open MLOps platform for quickly building and managing continuous ML applications across their lifecycle. MLRun integrates into your development and CI/CD environment and automates the delivery of production data, ML pipelines, and online applications, significantly reducing engineering efforts, time to production, and computation resources. With MLRun, you can choose any IDE on your local machine or on the cloud. MLRun breaks the silos between data, ML, software, and DevOps/MLOps teams, enabling collaboration and fast continuous improvements.
-
-Keywords: MLOps
-
-## [FederatedScope](https://github.com/alibaba/FederatedScope)
-
-[FederatedScope](https://github.com/alibaba/FederatedScope) is a comprehensive federated learning platform that provides convenient usage and flexible customization for various federated learning tasks in both academia and industry. Based on an event-driven architecture, [FederatedScope](https://github.com/alibaba/FederatedScope) integrates rich collections of functionalities to satisfy the burgeoning demands from federated learning, and aims to build up an easy-to-use platform for promoting learning safely and effectively.
-
-Keywords: Federated learning, Event-driven
-
-## [pythainlp](https://github.com/PyThaiNLP/pythainlp)
-
-PyThaiNLP is a Python package for text processing and linguistic analysis, similar to NLTK with focus on Thai language.
-
-Keywords: Thai, NLP, NLTK
-
-## [FlagAI](https://github.com/FlagAI-Open/FlagAI)
-
-[FlagAI](https://github.com/FlagAI-Open/FlagAI) (Fast LArge-scale General AI models) is a fast, easy-to-use and extensible toolkit for large-scale model. Our goal is to support training, fine-tuning, and deployment of large-scale models on various downstream tasks with multi-modality.
-
-Keywords: Large models, Training, Fine-tuning, Deployment, Multi-modal
-
-## [pyserini](https://github.com/castorini/pyserini)
-
-[pyserini](https://github.com/castorini/pyserini) is a Python toolkit for reproducible information retrieval research with sparse and dense representations. Retrieval using sparse representations is provided via integration with the group's Anserini IR toolkit. Retrieval using dense representations is provided via integration with Facebook's Faiss library.
-
-Keywords: IR, Information Retrieval, Dense, Sparse
-
-## [baal](https://github.com/baal-org/baal)
-
-[baal](https://github.com/baal-org/baal) is an active learning library that supports both industrial applications and research usecases. [baal](https://github.com/baal-org/baal) currently supports Monte-Carlo Dropout, MCDropConnect, deep ensembles, and semi-supervised learning.
-
-Keywords: Active Learning, Research, Labeling
-
-## [cleanlab](https://github.com/cleanlab/cleanlab)
-
-[cleanlab](https://github.com/cleanlab/cleanlab) is the standard data-centric AI package for data quality and machine learning with messy, real-world data and labels. For text, image, tabular, audio (among others) datasets, you can use cleanlab to automatically: detect data issues (outliers, label errors, near duplicates, etc), train robust ML models, infer consensus + annotator-quality for multi-annotator data, suggest data to (re)label next (active learning).
-
-Keywords: Data-Centric AI, Data Quality, Noisy Labels, Outlier Detection, Active Learning  
-
-## [BentoML](https://github.com/bentoml/BentoML)
-
-[BentoML](https://github.com/bentoml) is the unified framework for building, shipping, and scaling production-ready AI applications incorporating traditional ML, pre-trained AI models, Generative and Large Language Models. 
-All Hugging Face models and pipelines can be seamlessly integrated into BentoML applications, enabling the running of models on the most suitable hardware and independent scaling based on usage.
-
-Keywords: BentoML, Framework, Deployment, AI Applications
-
-## [LLaMA Factory](https://github.com/hiyouga/LLaMA-Factory)
-
-[LLaMA Factory](https://github.com/hiyouga/LLaMA-Factory) offers a user-friendly fine-tuning framework that incorporates PEFT. The repository includes training(fine-tuning) and inference examples for LLaMA-2, BLOOM, Falcon, Baichuan, Qwen, and other LLMs. A ChatGLM version is also available in [ChatGLM-Efficient-Tuning](https://github.com/hiyouga/ChatGLM-Efficient-Tuning).
-
-Keywords: PEFT, fine-tuning, LLaMA-2, ChatGLM, Qwen
-

benchmark/benchmark.py

@@ -1,326 +0,0 @@
-# Copyright 2024 The HuggingFace Team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-"""
-Run benchmark using the `optimum-benchmark` library with some customization in `transformers`.
-
-Assume we are under `transformers` root directory: (make sure the commits are valid commits)
-```bash
-python benchmark/benchmark.py --config-dir benchmark/config --config-name generation --commit=9b9c7f03da625b13643e99205c691fe046461724 --metrics=decode.latency.mean,per_token.latency.mean,per_token.throughput.value backend.model=google/gemma-2b benchmark.input_shapes.sequence_length=5,7 benchmark.input_shapes.batch_size=1,2 --multirun
-```
-"""
-
-import argparse
-import glob
-import json
-import os.path
-import re
-import tempfile
-from contextlib import contextmanager
-from pathlib import Path
-
-from git import Repo
-
-from huggingface_hub import HfApi
-
-from optimum_benchmark import Benchmark
-from optimum_benchmark_wrapper import main
-
-
-PATH_TO_REPO = Path(__file__).parent.parent.resolve()
-
-
-@contextmanager
-def checkout_commit(repo: Repo, commit_id: str):
-    """
-    Context manager that checks out a given commit when entered, but gets back to the reference it was at on exit.
-    Args:
-        repo (`git.Repo`): A git repository (for instance the Transformers repo).
-        commit_id (`str`): The commit reference to checkout inside the context manager.
-    """
-    current_head = repo.head.commit if repo.head.is_detached else repo.head.ref
-
-    try:
-        repo.git.checkout(commit_id)
-        yield
-
-    finally:
-        repo.git.checkout(current_head)
-
-
-def summarize(run_dir, metrics, expand_metrics=False):
-    """Produce a summary for each optimum-benchmark launched job's output directory found in `run_dir`.
-
-    Each summary's format is as follows (for `expand_metrics=False`):
-    ```
-    {
-        "model": "google/gemma-2b",
-        "commit": "3cd6ed22e4d49219f300f5055e71e3929aba20d7",
-        "config": "benchmark.input_shapes.batch_size=1,benchmark.input_shapes.sequence_length=5",
-        "metrics": {
-            "decode.latency.mean": 1.624666809082031,
-            "per_token.latency.mean": 0.012843788806628804,
-            "per_token.throughput.value": 77.85864553330948
-        }
-    }
-    ```
-    """
-    reports = glob.glob(os.path.join(run_dir, "**/benchmark_report.json"), recursive=True)
-    report_dirs = [str(Path(report).parent) for report in reports]
-
-    summaries = []
-    for report_dir in report_dirs:
-        commit = re.search(r"/commit=([^/]+)", report_dir).groups()[0]
-
-        if not os.path.isfile(os.path.join(report_dir, "benchmark.json")):
-            continue
-        benchmark = Benchmark.from_json(os.path.join(report_dir, "benchmark.json"))
-        report = benchmark.report
-
-        model = benchmark.config.backend["model"]
-
-        # Ths looks like `benchmark.input_shapes.batch_size=1,benchmark.input_shapes.sequence_length=5`.
-        # (we rely on the usage of hydra's `${hydra.job.override_dirname}`.)
-        benchmark_name = re.sub(f"backend.model={model},*", "", report_dir)
-        benchmark_name = str(Path(benchmark_name).parts[-1])
-        if benchmark_name.startswith("commit="):
-            benchmark_name = benchmark.config.name
-
-        metrics_values = {}
-        # post-processing of report: show a few selected/important metric
-        for metric in metrics:
-            keys = metric.split(".")
-            value = report
-            current = metrics_values
-            for key in keys:
-                # Avoid KeyError when a user's specified metric has typo.
-                # TODO: Give warnings.
-                if key not in value:
-                    continue
-                value = value[key]
-
-                if expand_metrics:
-                    if isinstance(value, dict):
-                        if key not in current:
-                            current[key] = {}
-                            current = current[key]
-                    else:
-                        current[key] = value
-
-            if not expand_metrics:
-                metrics_values[metric] = value
-
-        # show some config information
-        print(f"model: {model}")
-        print(f"commit: {commit}")
-        print(f"config: {benchmark_name}")
-        if len(metrics_values) > 0:
-            print("metrics:")
-            if expand_metrics:
-                print(metrics_values)
-            else:
-                for metric, value in metrics_values.items():
-                    print(f"  - {metric}: {value}")
-        print("-" * 80)
-
-        summary = {
-            "model": model,
-            "commit": commit,
-            "config": benchmark_name,
-            "metrics": metrics_values,
-        }
-        summaries.append(summary)
-
-        with open(os.path.join(report_dir, "summary.json"), "w") as fp:
-            json.dump(summary, fp, indent=4)
-
-    return summaries
-
-
-def combine_summaries(summaries):
-    """Combine a list of summary obtained from the function `summarize`.
-
-    The combined summary's format is as follows:
-    ```
-    "google/gemma-2b": {
-        "benchmark.input_shapes.batch_size=1,benchmark.input_shapes.sequence_length=5": {
-            "3cd6ed22e4d49219f300f5055e71e3929aba20d7": {
-                "metrics": {"decode.latency.mean": 1.624666809082031}
-            },
-            "c97ee28b117c0abe8e08891f402065e4df6d72aa": {
-                "metrics": {"decode.latency.mean": 1.6278163452148438}
-            }
-        },
-        "benchmark.input_shapes.batch_size=2,benchmark.input_shapes.sequence_length=5": {
-            "3cd6ed22e4d49219f300f5055e71e3929aba20d7": {
-                "metrics": {"decode.latency.mean": 1.6947791748046876}
-            },
-            "c97ee28b117c0abe8e08891f402065e4df6d72aa": {
-                "metrics": {
-                    "decode.latency.mean": 1.6980519409179688}
-            }
-        }
-    }
-    ```
-    """
-    combined = {}
-    for summary in summaries:
-        model = summary["model"]
-        config = summary["config"]
-        commit = summary["commit"]
-
-        if model not in combined:
-            combined[model] = {}
-
-        if config not in combined[model]:
-            combined[model][config] = {}
-
-        if commit not in combined[model][config]:
-            combined[model][config][commit] = {"metrics": summary["metrics"]}
-
-    with open(os.path.join(exp_run_dir, "summary.json"), "w") as fp:
-        json.dump(combined, fp, indent=4)
-
-    print(json.dumps(combined, indent=4))
-
-    return combined
-
-
-if __name__ == "__main__":
-
-    def list_str(values):
-        return values.split(",")
-
-    parser = argparse.ArgumentParser()
-
-    parser.add_argument("--config-dir", type=str, required=True, help="The path to the config directory.")
-    parser.add_argument("--config-name", type=str, required=True, help="The config name.")
-
-    # arguments specific to this wrapper for our own customization
-    parser.add_argument("--ensure_empty", type=bool, default=True, help="If to create a temporary directory.")
-    parser.add_argument(
-        "--commit",
-        type=list_str,
-        default="",
-        help="Comma-separated list of branch names and/or commit sha values on which the benchmark will run. If `diff` is specified, it will run on both the current head and the `main` branch.",
-    )
-    parser.add_argument("--metrics", type=str, help="The metrics to be included in the summary.")
-
-    parser.add_argument("--repo_id", type=str, default=None, help="The repository to which the file will be uploaded.")
-    parser.add_argument("--path_in_repo", type=str, default=None, help="Relative filepath in the repo.")
-    parser.add_argument("--token", type=str, default=None, help="A valid user access token (string).")
-
-    args, optimum_benchmark_args = parser.parse_known_args()
-
-    repo = Repo(PATH_TO_REPO)
-
-    metrics = [
-        "prefill.latency.mean",
-        "prefill.throughput.value",
-        "decode.latency.mean",
-        "decode.throughput.value",
-        "per_token.latency.mean",
-        "per_token.throughput.value",
-    ]
-    if args.metrics is not None:
-        metrics = args.metrics.split(",")
-
-    # Get `backend.model` in a hacky way: We want to control the experiment flow manually.
-    models = [""]
-    for idx, arg in enumerate(optimum_benchmark_args):
-        if arg.startswith("backend.model="):
-            models = arg[len("backend.model=") :]
-            models = models.split(",")
-            break
-    optimum_benchmark_args = [arg for arg in optimum_benchmark_args if not arg.startswith("backend.model=")]
-
-    # Get the commit(s)
-    current_head = str(repo.head.commit) if repo.head.is_detached else str(repo.head.ref)
-    commits = [x for x in args.commit if x != ""]
-    if len(commits) == 0:
-        commits = [current_head]
-    elif len(commits) == 1 and commits[0] == "diff":
-        # compare to `main`
-        commits = ["main", current_head]
-
-    # Get the specified run directory
-    run_dir_arg_idx, run_dir = -1, None
-    sweep_dir_arg_idx, sweep_dir = -1, None
-    for idx, arg in enumerate(optimum_benchmark_args):
-        if arg.startswith("hydra.run.dir="):
-            run_dir = arg[len("hydra.run.dir=") :]
-            run_dir_arg_idx = idx
-        elif arg.startswith("hydra.sweep.dir="):
-            sweep_dir = arg[len("hydra.sweep.dir=") :]
-            sweep_dir_arg_idx = idx
-    exp_run_dir, arg_dix, arg_name = (
-        (sweep_dir, sweep_dir_arg_idx, "hydra.sweep.dir")
-        if "--multirun" in optimum_benchmark_args
-        else (run_dir, run_dir_arg_idx, "hydra.run.dir")
-    )
-
-    # TODO: not hardcoded
-    if exp_run_dir is None and args.ensure_empty:
-        exp_run_dir = "_benchmark"
-
-    if args.ensure_empty:
-        os.makedirs(exp_run_dir, exist_ok=True)
-        exp_run_dir = tempfile.mkdtemp(dir=exp_run_dir)
-
-    run_summaries = []
-    for commit in commits:
-        with checkout_commit(repo, commit):
-            commit = str(repo.head.commit)
-
-            commit_run_dir = exp_run_dir
-            if exp_run_dir is not None:
-                commit_run_dir = os.path.join(exp_run_dir, rf"commit\={commit}")
-
-            print(f"Run benchmark on commit: {commit}")
-
-            for model in models:
-                model_arg = [f"backend.model={model}"] if model != "" else []
-                dir_args = []
-                if commit_run_dir is not None:
-                    if arg_dix > -1:
-                        optimum_benchmark_args[arg_dix] = f"{arg_name}={commit_run_dir}"
-                    else:
-                        dir_args = [
-                            f"hydra.sweep.dir={commit_run_dir}",
-                            f"hydra.run.dir={commit_run_dir}/" + "${hydra.job.override_dirname}",
-                        ]
-                main(args.config_dir, args.config_name, model_arg + dir_args + optimum_benchmark_args)
-
-            if commit_run_dir is not None:
-                # Need to remove the `\` character
-                summaries = summarize(commit_run_dir.replace("\\", ""), metrics)
-                run_summaries.extend(summaries)
-
-    # aggregate the information across the commits
-    if exp_run_dir is not None:
-        with open(os.path.join(exp_run_dir, "summaries.json"), "w") as fp:
-            json.dump(run_summaries, fp, indent=4)
-
-        combined_summary = combine_summaries(run_summaries)
-
-        if args.repo_id is not None and args.path_in_repo is not None:
-            # Upload to Hub
-            api = HfApi()
-            api.upload_folder(
-                folder_path=exp_run_dir,
-                path_in_repo=args.path_in_repo,
-                repo_id=args.repo_id,
-                repo_type="dataset",
-                token=args.token,
-            )

benchmark/config/generation.yaml

@@ -1,57 +0,0 @@
-defaults:
-  - benchmark # inheriting benchmark schema
-  - scenario: inference
-  - launcher: process
-  - backend: pytorch
-  - _self_ # for hydra 1.1 compatibility
-
-name: pytorch_generate
-
-launcher:
-  start_method: spawn
-  device_isolation: true
-  device_isolation_action: warn
-
-backend:
-  device: cuda
-  device_ids: 0
-  no_weights: true
-  model: meta-llama/Llama-2-7b-hf
-  cache_implementation: static
-  torch_compile: true
-  torch_dtype: float16
-  torch_compile_config:
-    backend: inductor
-    mode: reduce-overhead
-    fullgraph: true
-
-scenario:
-  input_shapes:
-    batch_size: 1
-    sequence_length: 7
-  generate_kwargs:
-    max_new_tokens: 128
-    min_new_tokens: 128
-    do_sample: false
-  memory: true
-  latency: true
-  iterations: 2
-  duration: 0
-
-
-# hydra/cli specific settings
-hydra:
-  run:
-    # where to store run results
-    dir: runs/${name}
-  job:
-    # change working directory to the run directory
-    chdir: true
-    env_set:
-      # set environment variable OVERRIDE_BENCHMARKS to 1
-      # to not skip benchmarks that have been run before
-      OVERRIDE_BENCHMARKS: 1
-      LOG_LEVEL: WARN
-  sweep:
-    dir: multirun
-    subdir: ${hydra.job.override_dirname}

benchmark/optimum_benchmark_wrapper.py

@@ -1,16 +0,0 @@
-import argparse
-import subprocess
-
-
-def main(config_dir, config_name, args):
-    subprocess.run(["optimum-benchmark", "--config-dir", f"{config_dir}", "--config-name", f"{config_name}"] + ["hydra/job_logging=disabled", "hydra/hydra_logging=disabled"] + args)
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser()
-
-    parser.add_argument("--config-dir", type=str, required=True, help="The path to the config directory.")
-    parser.add_argument("--config-name", type=str, required=True, help="The config name.")
-    args, unknown = parser.parse_known_args()
-
-    main(args.config_dir, args.config_name, unknown)

conftest.py

@@ -20,62 +20,9 @@ import sys
 import warnings
 from os.path import abspath, dirname, join
 
-import _pytest
-import pytest
-
-from transformers.testing_utils import HfDoctestModule, HfDocTestParser
-
-
-NOT_DEVICE_TESTS = {
-    "test_tokenization",
-    "test_processor",
-    "test_processing",
-    "test_beam_constraints",
-    "test_configuration_utils",
-    "test_data_collator",
-    "test_trainer_callback",
-    "test_trainer_utils",
-    "test_feature_extraction",
-    "test_image_processing",
-    "test_image_processor",
-    "test_image_transforms",
-    "test_optimization",
-    "test_retrieval",
-    "test_config",
-    "test_from_pretrained_no_checkpoint",
-    "test_keep_in_fp32_modules",
-    "test_gradient_checkpointing_backward_compatibility",
-    "test_gradient_checkpointing_enable_disable",
-    "test_save_load_fast_init_from_base",
-    "test_fast_init_context_manager",
-    "test_fast_init_tied_embeddings",
-    "test_save_load_fast_init_to_base",
-    "test_torch_save_load",
-    "test_initialization",
-    "test_forward_signature",
-    "test_model_get_set_embeddings",
-    "test_model_main_input_name",
-    "test_correct_missing_keys",
-    "test_tie_model_weights",
-    "test_can_use_safetensors",
-    "test_load_save_without_tied_weights",
-    "test_tied_weights_keys",
-    "test_model_weights_reload_no_missing_tied_weights",
-    "test_pt_tf_model_equivalence",
-    "test_mismatched_shapes_have_properly_initialized_weights",
-    "test_matched_shapes_have_loaded_weights_when_some_mismatched_shapes_exist",
-    "test_model_is_small",
-    "test_tf_from_pt_safetensors",
-    "test_flax_from_pt_safetensors",
-    "ModelTest::test_pipeline_",  # None of the pipeline tests from PipelineTesterMixin (of which XxxModelTest inherits from) are running on device
-    "ModelTester::test_pipeline_",
-    "/repo_utils/",
-    "/utils/",
-    "/agents/",
-}
 
 # allow having multiple repository checkouts and not needing to remember to rerun
-# `pip install -e '.[dev]'` when switching between checkouts and running tests.
+# 'pip install -e .[dev]' when switching between checkouts and running tests.
 git_repo_path = abspath(join(dirname(__file__), "src"))
 sys.path.insert(1, git_repo_path)
 
@@ -85,23 +32,14 @@ warnings.simplefilter(action="ignore", category=FutureWarning)
 
 
 def pytest_configure(config):
+    config.addinivalue_line("markers", "is_pipeline_test: mark test to run only when pipeline are tested")
     config.addinivalue_line(
         "markers", "is_pt_tf_cross_test: mark test to run only when PT and TF interactions are tested"
     )
     config.addinivalue_line(
         "markers", "is_pt_flax_cross_test: mark test to run only when PT and FLAX interactions are tested"
     )
-    config.addinivalue_line("markers", "is_pipeline_test: mark test to run only when pipelines are tested")
     config.addinivalue_line("markers", "is_staging_test: mark test to run only in the staging environment")
-    config.addinivalue_line("markers", "accelerate_tests: mark test that require accelerate")
-    config.addinivalue_line("markers", "agent_tests: mark the agent tests that are run on their specific schedule")
-    config.addinivalue_line("markers", "not_device_test: mark the tests always running on cpu")
-
-
-def pytest_collection_modifyitems(items):
-    for item in items:
-        if any(test_name in item.nodeid for test_name in NOT_DEVICE_TESTS):
-            item.add_marker(pytest.mark.not_device_test)
 
 
 def pytest_addoption(parser):
@@ -125,7 +63,7 @@ def pytest_sessionfinish(session, exitstatus):
 
 
 # Doctest custom flag to ignore output.
-IGNORE_RESULT = doctest.register_optionflag("IGNORE_RESULT")
+IGNORE_RESULT = doctest.register_optionflag('IGNORE_RESULT')
 
 OutputChecker = doctest.OutputChecker
 
@@ -138,5 +76,3 @@ class CustomOutputChecker(OutputChecker):
 
 
 doctest.OutputChecker = CustomOutputChecker
-_pytest.doctest.DoctestModule = HfDoctestModule
-doctest.DocTestParser = HfDocTestParser

docker/consistency.dockerfile

@@ -1,15 +0,0 @@
-FROM python:3.10-slim
-ENV PYTHONDONTWRITEBYTECODE=1
-USER root
-ARG REF=main
-RUN apt-get update && apt-get install -y time git pkg-config make git-lfs
-ENV UV_PYTHON=/usr/local/bin/python
-RUN pip install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools GitPython
-RUN uv pip install --no-cache-dir --upgrade 'torch' --index-url https://download.pytorch.org/whl/cpu
-# tensorflow pin matching setup.py
-RUN uv pip install --no-cache-dir "tensorflow-cpu<2.16" "tf-keras<2.16"
-RUN uv pip install --no-cache-dir "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[flax,quality,torch-speech,vision,testing]"
-RUN git lfs install
-
-RUN pip uninstall -y transformers
-RUN apt-get clean && rm -rf /var/lib/apt/lists/* && apt-get autoremove && apt-get autoclean

docker/custom-tokenizers.dockerfile

@@ -1,26 +0,0 @@
-FROM python:3.10-slim
-ENV PYTHONDONTWRITEBYTECODE=1
-USER root
-RUN apt-get update && apt-get install -y libsndfile1-dev espeak-ng time git cmake wget xz-utils build-essential g++5 libprotobuf-dev protobuf-compiler
-ENV UV_PYTHON=/usr/local/bin/python
-RUN pip --no-cache-dir install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools
-
-RUN wget https://github.com/ku-nlp/jumanpp/releases/download/v2.0.0-rc3/jumanpp-2.0.0-rc3.tar.xz
-RUN tar xvf jumanpp-2.0.0-rc3.tar.xz
-RUN mkdir jumanpp-2.0.0-rc3/bld
-WORKDIR ./jumanpp-2.0.0-rc3/bld
-RUN wget -LO catch.hpp https://github.com/catchorg/Catch2/releases/download/v2.13.8/catch.hpp
-RUN mv catch.hpp ../libs/
-RUN cmake .. -DCMAKE_INSTALL_PREFIX=/usr/local
-RUN make install -j 10
-
-
-RUN uv pip install --no-cache --upgrade 'torch' --index-url https://download.pytorch.org/whl/cpu
-RUN uv pip install --no-cache-dir  --no-deps accelerate --extra-index-url https://download.pytorch.org/whl/cpu 
-RUN uv pip install  --no-cache-dir "transformers[ja,testing,sentencepiece,jieba,spacy,ftfy,rjieba]" unidic unidic-lite
-# spacy is not used so not tested. Causes to failures. TODO fix later
-RUN python3 -m unidic download
-RUN pip uninstall -y transformers
-
-RUN apt-get clean && rm -rf /var/lib/apt/lists/*
-RUN apt remove -y g++ cmake  xz-utils libprotobuf-dev protobuf-compiler

docker/examples-tf.dockerfile

@@ -1,12 +0,0 @@
-FROM python:3.10-slim
-ENV PYTHONDONTWRITEBYTECODE=1
-USER root
-RUN apt-get update && apt-get install -y libsndfile1-dev espeak-ng time git
-RUN apt-get install -y g++ cmake
-ENV UV_PYTHON=/usr/local/bin/python
-RUN pip --no-cache-dir install uv && uv venv
-RUN uv pip install --no-cache-dir -U pip setuptools albumentations seqeval
-RUN pip install  --upgrade --no-cache-dir "transformers[tf-cpu,sklearn,testing,sentencepiece,tf-speech,vision]"
-RUN uv pip install --no-cache-dir  "protobuf==3.20.3" 
-RUN pip uninstall -y transformers
-RUN apt-get clean && rm -rf /var/lib/apt/lists/*

docker/examples-torch.dockerfile

@@ -1,11 +0,0 @@
-FROM python:3.10-slim
-ENV PYTHONDONTWRITEBYTECODE=1
-USER root
-RUN apt-get update &&  apt-get install -y --no-install-recommends libsndfile1-dev espeak-ng time git g++ cmake pkg-config openssh-client git
-ENV UV_PYTHON=/usr/local/bin/python
-RUN pip --no-cache-dir install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools
-RUN pip install --no-cache-dir 'torch' 'torchvision' 'torchaudio' --index-url https://download.pytorch.org/whl/cpu
-RUN uv pip install --no-deps timm accelerate --extra-index-url https://download.pytorch.org/whl/cpu 
-RUN uv pip install --no-cache-dir librosa "transformers[sklearn,sentencepiece,vision,testing]" seqeval albumentations jiwer
-RUN pip uninstall -y transformers
-RUN apt-get clean && rm -rf /var/lib/apt/lists/*

docker/exotic-models.dockerfile

@@ -1,17 +0,0 @@
-FROM python:3.10-slim
-ENV PYTHONDONTWRITEBYTECODE=1
-ARG REF=main
-USER root
-RUN apt-get update && apt-get install -y libsndfile1-dev espeak-ng time git libgl1-mesa-glx libgl1 g++ tesseract-ocr
-ENV UV_PYTHON=/usr/local/bin/python
-RUN pip --no-cache-dir install uv &&  uv venv && uv pip install --no-cache-dir -U pip setuptools
-RUN pip install --no-cache-dir 'torch' 'torchvision' 'torchaudio' --index-url https://download.pytorch.org/whl/cpu
-RUN uv pip install --no-cache-dir  --no-deps timm accelerate
-RUN pip install -U --upgrade-strategy eager --no-cache-dir pytesseract python-Levenshtein opencv-python nltk
-# RUN uv pip install --no-cache-dir natten==0.15.1+torch210cpu -f https://shi-labs.com/natten/wheels
-RUN pip install  --no-cache-dir "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[testing, vision]" 'scikit-learn' 'torch-stft' 'nose'  'dataset'
-# RUN git clone https://github.com/facebookresearch/detectron2.git
-# RUN python3 -m pip install --no-cache-dir -e detectron2
-RUN pip install 'git+https://github.com/facebookresearch/detectron2.git@92ae9f0b92aba5867824b4f12aa06a22a60a45d3'
-RUN pip uninstall -y transformers
-RUN apt-get clean && rm -rf /var/lib/apt/lists/*

docker/jax-light.dockerfile

@@ -1,10 +0,0 @@
-FROM python:3.10-slim
-ENV PYTHONDONTWRITEBYTECODE=1
-ARG REF=main
-USER root
-RUN apt-get update && apt-get install -y libsndfile1-dev espeak-ng time git g++ cmake
-ENV UV_PYTHON=/usr/local/bin/python
-RUN pip --no-cache-dir install uv &&  uv venv && uv pip install --no-cache-dir -U pip setuptools
-RUN pip install --no-cache-dir "scipy<1.13" "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[flax,testing,sentencepiece,flax-speech,vision]"
-RUN pip uninstall -y transformers
-RUN apt-get clean && rm -rf /var/lib/apt/lists/* && apt-get autoremove && apt-get autoclean

docker/pipeline-tf.dockerfile

@@ -1,10 +0,0 @@
-FROM python:3.10-slim
-ENV PYTHONDONTWRITEBYTECODE=1
-ARG REF=main
-USER root
-RUN apt-get update && apt-get install -y libsndfile1-dev espeak-ng time git cmake g++
-ENV UV_PYTHON=/usr/local/bin/python
-RUN pip --no-cache-dir install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools
-RUN pip install --no-cache-dir "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[sklearn,tf-cpu,testing,sentencepiece,tf-speech,vision]"
-RUN uv pip install --no-cache-dir  "protobuf==3.20.3" tensorflow_probability
-RUN apt-get clean && rm -rf /var/lib/apt/lists/*

docker/pipeline-torch.dockerfile

@@ -1,11 +0,0 @@
-FROM python:3.10-slim
-ENV PYTHONDONTWRITEBYTECODE=1
-ARG REF=main
-USER root
-RUN apt-get update &&  apt-get install -y --no-install-recommends libsndfile1-dev espeak-ng time git pkg-config openssh-client git
-ENV UV_PYTHON=/usr/local/bin/python
-RUN pip --no-cache-dir install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools
-RUN pip install --no-cache-dir 'torch' 'torchvision' 'torchaudio' --index-url https://download.pytorch.org/whl/cpu
-RUN uv pip install --no-deps timm accelerate --extra-index-url https://download.pytorch.org/whl/cpu 
-RUN uv pip install --no-cache-dir librosa "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[sklearn,sentencepiece,vision,testing]"
-RUN pip uninstall -y transformers

docker/quality.dockerfile

@@ -1,9 +0,0 @@
-FROM python:3.10-slim
-ENV PYTHONDONTWRITEBYTECODE=1
-ARG REF=main
-USER root
-RUN apt-get update && apt-get install -y time git 
-ENV UV_PYTHON=/usr/local/bin/python
-RUN pip install uv &&  uv venv
-RUN uv pip install --no-cache-dir -U pip setuptools GitPython "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[ruff]" urllib3
-RUN apt-get install -y jq curl && apt-get clean && rm -rf /var/lib/apt/lists/*

docker/tf-light.dockerfile

@@ -1,12 +0,0 @@
-FROM python:3.10-slim
-ENV PYTHONDONTWRITEBYTECODE=1
-ARG REF=main
-USER root
-RUN apt-get update &&  apt-get install -y --no-install-recommends libsndfile1-dev espeak-ng time git g++ pkg-config openssh-client git
-RUN apt-get install -y  cmake
-ENV UV_PYTHON=/usr/local/bin/python
-RUN pip --no-cache-dir install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools
-RUN pip install  --upgrade --no-cache-dir "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[tf-cpu,sklearn,testing,sentencepiece,tf-speech,vision]"
-RUN uv pip install --no-cache-dir  "protobuf==3.20.3" 
-RUN pip uninstall -y transformers
-RUN apt-get clean && rm -rf /var/lib/apt/lists/* && apt-get autoremove && apt-get autoclean

docker/torch-jax-light.dockerfile

@@ -1,16 +0,0 @@
-FROM python:3.10-slim
-ENV PYTHONDONTWRITEBYTECODE=1
-ARG REF=main
-USER root
-RUN apt-get update &&  apt-get install -y libsndfile1-dev espeak-ng time git g++ cmake pkg-config openssh-client git
-ENV UV_PYTHON=/usr/local/bin/python
-RUN pip --no-cache-dir install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools
-RUN uv pip install --no-deps accelerate
-RUN pip install --no-cache-dir 'torch' 'torchvision' 'torchaudio' --index-url https://download.pytorch.org/whl/cpu
-RUN pip install --no-cache-dir "scipy<1.13" "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[flax,audio,sklearn,sentencepiece,vision,testing]"
-
-
-# RUN pip install --no-cache-dir "scipy<1.13" "transformers[flax,testing,sentencepiece,flax-speech,vision]"
-
-RUN pip uninstall -y transformers
-RUN apt-get clean && rm -rf /var/lib/apt/lists/* && apt-get autoremove && apt-get autoclean

docker/torch-light.dockerfile

@@ -1,11 +0,0 @@
-FROM python:3.10-slim
-ENV PYTHONDONTWRITEBYTECODE=1
-ARG REF=main
-USER root
-RUN apt-get update &&  apt-get install -y --no-install-recommends libsndfile1-dev espeak-ng time git g++ cmake pkg-config openssh-client git git-lfs
-ENV UV_PYTHON=/usr/local/bin/python
-RUN pip --no-cache-dir install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools
-RUN pip install --no-cache-dir 'torch' 'torchvision' 'torchaudio' --index-url https://download.pytorch.org/whl/cpu
-RUN uv pip install --no-deps timm accelerate --extra-index-url https://download.pytorch.org/whl/cpu 
-RUN uv pip install --no-cache-dir librosa "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[sklearn,sentencepiece,vision,testing]"
-RUN pip uninstall -y transformers

docker/torch-tf-light.dockerfile

@@ -1,19 +0,0 @@
-FROM python:3.10-slim
-ENV PYTHONDONTWRITEBYTECODE=1
-ARG REF=main
-RUN echo ${REF}
-USER root
-RUN apt-get update &&  apt-get install -y --no-install-recommends libsndfile1-dev espeak-ng time git g++ cmake pkg-config openssh-client git git-lfs
-ENV UV_PYTHON=/usr/local/bin/python
-RUN pip --no-cache-dir install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools
-RUN uv pip install --no-cache-dir  --no-deps accelerate --extra-index-url https://download.pytorch.org/whl/cpu 
-RUN pip install --no-cache-dir 'torch' 'torchvision' 'torchaudio' --index-url https://download.pytorch.org/whl/cpu
-RUN git lfs install
-
-RUN uv pip install --no-cache-dir pypi-kenlm
-RUN pip install --no-cache-dir  "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[tf-cpu,sklearn,sentencepiece,vision,testing]"
-RUN uv pip install --no-cache-dir  "protobuf==3.20.3" librosa
-
-
-RUN pip uninstall -y transformers
-RUN apt-get clean && rm -rf /var/lib/apt/lists/* && apt-get autoremove && apt-get autoclean

docker/transformers-all-latest-gpu/Dockerfile

@@ -1,4 +1,4 @@
-FROM nvidia/cuda:12.1.0-cudnn8-devel-ubuntu20.04
+FROM nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04
 LABEL maintainer="Hugging Face"
 
 ARG DEBIAN_FRONTEND=noninteractive
@@ -9,11 +9,11 @@ SHELL ["sh", "-lc"]
 # The following `ARG` are mainly used to specify the versions explicitly & directly in this docker file, and not meant
 # to be used as arguments for docker build (so far).
 
-ARG PYTORCH='2.4.0'
+ARG PYTORCH='1.12.1'
 # (not always a valid torch version)
-ARG INTEL_TORCH_EXT='2.3.0'
+ARG INTEL_TORCH_EXT='1.11.0'
 # Example: `cu102`, `cu113`, etc.
-ARG CUDA='cu121'
+ARG CUDA='cu113'
 
 RUN apt update
 RUN apt install -y git libsndfile1-dev tesseract-ocr espeak-ng python3 python3-pip ffmpeg git-lfs
@@ -22,48 +22,33 @@ RUN python3 -m pip install --no-cache-dir --upgrade pip
 
 ARG REF=main
 RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
+RUN python3 -m pip install --no-cache-dir -e ./transformers[dev,onnxruntime]
 
-# 1. Put several commands in a single `RUN` to avoid image/layer exporting issue. Could be revised in the future.
-# 2. Regarding `torch` part, We might need to specify proper versions for `torchvision` and `torchaudio`.
-#    Currently, let's not bother to specify their versions explicitly (so installed with their latest release versions).
-RUN python3 -m pip install --no-cache-dir -U tensorflow==2.13 protobuf==3.20.3 tensorflow_text tensorflow_probability && python3 -m pip install --no-cache-dir -e ./transformers[dev,onnxruntime] && [ ${#PYTORCH} -gt 0 -a "$PYTORCH" != "pre" ] && VERSION='torch=='$PYTORCH'.*' ||  VERSION='torch'; echo "export VERSION='$VERSION'" >> ~/.profile && echo torch=$VERSION && [ "$PYTORCH" != "pre" ] && python3 -m pip install --no-cache-dir -U $VERSION torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/$CUDA || python3 -m pip install --no-cache-dir -U --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/$CUDA
+# TODO: Handle these in a python utility script
+RUN [ ${#PYTORCH} -gt 0 -a "$PYTORCH" != "pre" ] && VERSION='torch=='$PYTORCH'.*' ||  VERSION='torch'; echo "export VERSION='$VERSION'" >> ~/.profile
+RUN echo torch=$VERSION
+# `torchvision` and `torchaudio` should be installed along with `torch`, especially for nightly build.
+# Currently, let's just use their latest releases (when `torch` is installed with a release version)
+# TODO: We might need to specify proper versions that work with a specific torch version (especially for past CI).
+RUN [ "$PYTORCH" != "pre" ] && python3 -m pip install --no-cache-dir -U $VERSION torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/$CUDA || python3 -m pip install --no-cache-dir -U --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/$CUDA
 
+RUN python3 -m pip install --no-cache-dir -U tensorflow
 RUN python3 -m pip uninstall -y flax jax
 
-RUN python3 -m pip install --no-cache-dir intel_extension_for_pytorch==$INTEL_TORCH_EXT -f https://developer.intel.com/ipex-whl-stable-cpu
+# Use installed torch version for `torch-scatter` to avid to deal with PYTORCH='pre'.
+# If torch is nightly version, the link is likely to be invalid, but the installation falls back to the latest torch-scatter
+RUN python3 -m pip install --no-cache-dir torch-scatter -f https://data.pyg.org/whl/torch-$(python3 -c "from torch import version; print(version.__version__.split('+')[0])")+$CUDA.html
+RUN python3 -m pip install --no-cache-dir intel_extension_for_pytorch==$INTEL_TORCH_EXT+cpu -f https://software.intel.com/ipex-whl-stable
 
-RUN python3 -m pip install --no-cache-dir git+https://github.com/facebookresearch/detectron2.git pytesseract
+RUN python3 -m pip install --no-cache-dir git+https://github.com/facebookresearch/detectron2.git pytesseract https://github.com/kpu/kenlm/archive/master.zip
 RUN python3 -m pip install -U "itsdangerous<2.1.0"
 
 RUN python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
 
-RUN python3 -m pip install --no-cache-dir git+https://github.com/huggingface/peft@main#egg=peft
-
-# For bettertransformer
-RUN python3 -m pip install --no-cache-dir git+https://github.com/huggingface/optimum@main#egg=optimum
-
-# For video model testing
-RUN python3 -m pip install --no-cache-dir decord av==9.2.0
-
-# Some slow tests require bnb
+# Add bitsandbytes for mixed int8 testing
 RUN python3 -m pip install --no-cache-dir bitsandbytes
 
-# Some tests require quanto
-RUN python3 -m pip install --no-cache-dir quanto
-
-# `quanto` will install `ninja` which leads to many `CUDA error: an illegal memory access ...` in some model tests
-# (`deformable_detr`, `rwkv`, `mra`)
-RUN python3 -m pip uninstall -y ninja
-
-# For `dinat` model
-# The `XXX` part in `torchXXX` needs to match `PYTORCH` (to some extent)
-RUN python3 -m pip install --no-cache-dir natten==0.15.1+torch220$CUDA -f https://shi-labs.com/natten/wheels
-
-# For `nougat` tokenizer
-RUN python3 -m pip install --no-cache-dir python-Levenshtein
-
-# For `FastSpeech2ConformerTokenizer` tokenizer
-RUN python3 -m pip install --no-cache-dir g2p-en
+RUN python3 -m pip install --no-cache-dir decord
 
 # When installing in editable mode, `transformers` is not recognized as a package.
 # this line must be added in order for python to be aware of transformers.

docker/transformers-cpu/Dockerfile

@@ -0,0 +1,26 @@
+FROM ubuntu:18.04
+LABEL maintainer="Hugging Face"
+LABEL repository="transformers"
+
+RUN apt update && \
+    apt install -y bash \
+                   build-essential \
+                   git \
+                   curl \
+                   ca-certificates \
+                   python3 \
+                   python3-pip && \
+    rm -rf /var/lib/apt/lists
+
+RUN python3 -m pip install --no-cache-dir --upgrade pip && \
+    python3 -m pip install --no-cache-dir \
+    jupyter \
+    tensorflow-cpu \
+    torch
+
+WORKDIR /workspace
+COPY . transformers/
+RUN cd transformers/ && \
+    python3 -m pip install --no-cache-dir .
+
+CMD ["/bin/bash"]

docker/transformers-doc-builder/Dockerfile

@@ -1,4 +1,4 @@
-FROM python:3.10
+FROM python:3.8
 LABEL maintainer="Hugging Face"
 
 RUN apt update
@@ -10,7 +10,9 @@ RUN apt-get -y update && apt-get install -y libsndfile1-dev && apt install -y te
 # Torch needs to be installed before deepspeed
 RUN python3 -m pip install --no-cache-dir ./transformers[deepspeed]
 
-RUN python3 -m pip install --no-cache-dir torchvision git+https://github.com/facebookresearch/detectron2.git pytesseract
+RUN python3 -m pip install --no-cache-dir torch-scatter -f https://data.pyg.org/whl/torch-$(python -c "from torch import version; print(version.__version__.split('+')[0])")+cpu.html
+RUN python3 -m pip install --no-cache-dir torchvision git+https://github.com/facebookresearch/detectron2.git pytesseract https://github.com/kpu/kenlm/archive/master.zip
+RUN python3 -m pip install --no-cache-dir pytorch-quantization --extra-index-url https://pypi.ngc.nvidia.com
 RUN python3 -m pip install -U "itsdangerous<2.1.0"
 
 # Test if the image could successfully build the doc. before publishing the image

docker/transformers-past-gpu/Dockerfile

@@ -1,4 +1,4 @@
-ARG BASE_DOCKER_IMAGE
+ARG BASE_DOCKER_IMAGE="nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04"
 FROM $BASE_DOCKER_IMAGE
 LABEL maintainer="Hugging Face"
 
@@ -8,7 +8,7 @@ ARG DEBIAN_FRONTEND=noninteractive
 SHELL ["sh", "-lc"]
 
 RUN apt update
-RUN apt install -y git libsndfile1-dev tesseract-ocr espeak-ng python3 python3-pip ffmpeg git-lfs libaio-dev
+RUN apt install -y git libsndfile1-dev tesseract-ocr espeak-ng python3 python3-pip ffmpeg git-lfs
 RUN git lfs install
 RUN python3 -m pip install --no-cache-dir --upgrade pip
 
@@ -23,11 +23,9 @@ RUN cd transformers && python3 setup.py develop
 ARG FRAMEWORK
 ARG VERSION
 
-# Control `setuptools` version to avoid some issues
-RUN [ "$VERSION" != "1.10" ] && python3 -m pip install -U setuptools || python3 -m pip install -U "setuptools<=59.5"
-
 # Remove all frameworks
-RUN python3 -m pip uninstall -y torch torchvision torchaudio tensorflow jax flax
+# (`accelerate` requires `torch`, and this causes import issues for TF-only testing)
+RUN python3 -m pip uninstall -y torch torchvision torchaudio accelerate tensorflow jax flax
 
 # Get the libraries and their versions to install, and write installation command to `~/.profile`.
 RUN python3 ./transformers/utils/past_ci_versions.py --framework $FRAMEWORK --version $VERSION
@@ -36,24 +34,10 @@ RUN python3 ./transformers/utils/past_ci_versions.py --framework $FRAMEWORK --ve
 RUN echo "INSTALL_CMD = $INSTALL_CMD"
 RUN $INSTALL_CMD
 
-RUN [ "$FRAMEWORK" != "pytorch" ] && echo "`deepspeed-testing` installation is skipped" || python3 -m pip install --no-cache-dir ./transformers[deepspeed-testing]
-
-# Remove `accelerate`: it requires `torch`, and this causes import issues for TF-only testing
-# We will install `accelerate@main` in Past CI workflow file
-RUN python3 -m pip uninstall -y accelerate
-
-# Uninstall `torch-tensorrt` and `apex` shipped with the base image
-RUN python3 -m pip uninstall -y torch-tensorrt apex
-
-# Pre-build **nightly** release of DeepSpeed, so it would be ready for testing (otherwise, the 1st deepspeed test will timeout)
-RUN python3 -m pip uninstall -y deepspeed
-# This has to be run inside the GPU VMs running the tests. (So far, it fails here due to GPU checks during compilation.)
-# Issue: https://github.com/microsoft/DeepSpeed/issues/2010
-# RUN git clone https://github.com/microsoft/DeepSpeed && cd DeepSpeed && rm -rf build && \
-#    DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 DS_BUILD_UTILS=1 python3 -m pip install . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check 2>&1
+# Having installation problems for torch-scatter with torch <= 1.6. Disable so we have the same set of tests.
+# (This part will be removed once the logic of using `past_ci_versions.py` is used in other Dockerfile files.)
+# # Use installed torch version for `torch-scatter`.
+# # (The env. variable $CUDA is defined in `past_ci_versions.py`)
+# RUN [ "$FRAMEWORK" = "pytorch" ] && python3 -m pip install --no-cache-dir torch-scatter -f https://data.pyg.org/whl/torch-$(python3 -c "from torch import version; print(version.__version__.split('+')[0])")+$CUDA.html || echo "torch-scatter not to be installed"
 
 RUN python3 -m pip install -U "itsdangerous<2.1.0"
-
-# When installing in editable mode, `transformers` is not recognized as a package.
-# this line must be added in order for python to be aware of transformers.
-RUN cd transformers && python3 setup.py develop

docker/transformers-pytorch-amd-gpu/Dockerfile

@@ -1,34 +0,0 @@
-FROM rocm/dev-ubuntu-22.04:6.0.2
-# rocm/pytorch has no version with 2.1.0
-LABEL maintainer="Hugging Face"
-
-ARG DEBIAN_FRONTEND=noninteractive
-
-RUN apt update && \
-    apt install -y --no-install-recommends git libsndfile1-dev tesseract-ocr espeak-ng python3 python3-dev python3-pip python3-dev ffmpeg && \
-    apt clean && \
-    rm -rf /var/lib/apt/lists/*
-
-RUN python3 -m pip install --no-cache-dir --upgrade pip numpy
-
-RUN python3 -m pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm6.0
-
-RUN python3 -m pip install --no-cache-dir --upgrade importlib-metadata setuptools ninja git+https://github.com/facebookresearch/detectron2.git pytesseract "itsdangerous<2.1.0"
-
-ARG REF=main
-WORKDIR /
-
-# Invalidate docker cache from here if new commit is available.
-ADD https://api.github.com/repos/huggingface/transformers/git/refs/heads/main version.json
-RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
-
-RUN python3 -m pip install --no-cache-dir -e ./transformers[dev-torch,testing,video]
-
-RUN python3 -m pip uninstall -y tensorflow flax
-
-# When installing in editable mode, `transformers` is not recognized as a package.
-# this line must be added in order for python to be aware of transformers.
-RUN cd transformers && python3 setup.py develop
-
-# Remove nvml as it is not compatible with ROCm. apex is not tested on NVIDIA either.
-RUN python3 -m pip uninstall py3nvml pynvml apex -y

docker/transformers-pytorch-cpu/Dockerfile

@@ -0,0 +1,25 @@
+FROM ubuntu:18.04
+LABEL maintainer="Hugging Face"
+LABEL repository="transformers"
+
+RUN apt update && \
+    apt install -y bash \
+                   build-essential \
+                   git \
+                   curl \
+                   ca-certificates \
+                   python3 \
+                   python3-pip && \
+    rm -rf /var/lib/apt/lists
+
+RUN python3 -m pip install --no-cache-dir --upgrade pip && \
+    python3 -m pip install --no-cache-dir \
+    jupyter \
+    torch
+
+WORKDIR /workspace
+COPY . transformers/
+RUN cd transformers/ && \
+    python3 -m pip install --no-cache-dir .
+
+CMD ["/bin/bash"]

docker/transformers-pytorch-deepspeed-amd-gpu/Dockerfile

@@ -1,48 +0,0 @@
-FROM rocm/dev-ubuntu-22.04:5.6
-LABEL maintainer="Hugging Face"
-
-ARG DEBIAN_FRONTEND=noninteractive
-ARG PYTORCH='2.1.1'
-ARG TORCH_VISION='0.16.1'
-ARG TORCH_AUDIO='2.1.1'
-ARG ROCM='5.6'
-
-RUN apt update && \
-    apt install -y --no-install-recommends \
-    libaio-dev \
-    git \
-    # These are required to build deepspeed.
-    python3-dev \
-    python-is-python3 \
-    rocrand-dev \
-    rocthrust-dev \
-    hipsparse-dev \
-    hipblas-dev \
-    rocblas-dev && \
-    apt clean && \
-    rm -rf /var/lib/apt/lists/*
-
-RUN python3 -m pip install --no-cache-dir --upgrade pip ninja "pydantic<2"
-RUN python3 -m pip uninstall -y apex torch torchvision torchaudio
-RUN python3 -m pip install torch==$PYTORCH torchvision==$TORCH_VISION torchaudio==$TORCH_AUDIO --index-url https://download.pytorch.org/whl/rocm$ROCM --no-cache-dir
-
-# Pre-build DeepSpeed, so it's be ready for testing (to avoid timeout)
-RUN DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 python3 -m pip install deepspeed --global-option="build_ext" --global-option="-j8" --no-cache-dir -v --disable-pip-version-check 2>&1
-
-ARG REF=main
-WORKDIR /
-
-# Invalidate docker cache from here if new commit is available.
-ADD https://api.github.com/repos/huggingface/transformers/git/refs/heads/main version.json
-RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
-
-RUN python3 -m pip install --no-cache-dir ./transformers[accelerate,testing,sentencepiece,sklearn]
-
-# When installing in editable mode, `transformers` is not recognized as a package.
-# this line must be added in order for python to be aware of transformers.
-RUN cd transformers && python3 setup.py develop
-
-RUN python3 -c "from deepspeed.launcher.runner import main"
-
-# Remove nvml as it is not compatible with ROCm
-RUN python3 -m pip uninstall py3nvml pynvml -y

docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile

@@ -1,12 +1,11 @@
-# https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-23-11.html#rel-23-11
-FROM nvcr.io/nvidia/pytorch:23.04-py3
+FROM nvcr.io/nvidia/pytorch:21.03-py3
 LABEL maintainer="Hugging Face"
 
 ARG DEBIAN_FRONTEND=noninteractive
 
-ARG PYTORCH='2.2.0'
+ARG PYTORCH='1.12.1'
 # Example: `cu102`, `cu113`, etc.
-ARG CUDA='cu121'
+ARG CUDA='cu113'
 
 RUN apt -y update
 RUN apt install -y libaio-dev
@@ -15,39 +14,22 @@ RUN python3 -m pip install --no-cache-dir --upgrade pip
 ARG REF=main
 RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
 
-RUN python3 -m pip install --no-cache-dir ./transformers[deepspeed-testing]
-
 # Install latest release PyTorch
 # (PyTorch must be installed before pre-compiling any DeepSpeed c++/cuda ops.)
 # (https://www.deepspeed.ai/tutorials/advanced-install/#pre-install-deepspeed-ops)
-RUN python3 -m pip uninstall -y torch torchvision torchaudio && python3 -m pip install --no-cache-dir -U torch==$PYTORCH torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/$CUDA
+RUN python3 -m pip install --no-cache-dir -U torch==$PYTORCH torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/$CUDA
 
-RUN python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
-
-# Uninstall `transformer-engine` shipped with the base image
-RUN python3 -m pip uninstall -y transformer-engine
-
-# Uninstall `torch-tensorrt` shipped with the base image
-RUN python3 -m pip uninstall -y torch-tensorrt
-
-# recompile apex
-RUN python3 -m pip uninstall -y apex
-# RUN git clone https://github.com/NVIDIA/apex
-#  `MAX_JOBS=1` disables parallel building to avoid cpu memory OOM when building image on GitHub Action (standard) runners
-# TODO: check if there is alternative way to install latest apex
-# RUN cd apex && MAX_JOBS=1 python3 -m pip install --global-option="--cpp_ext" --global-option="--cuda_ext" --no-cache -v --disable-pip-version-check .
+RUN python3 -m pip install --no-cache-dir ./transformers[deepspeed-testing]
 
 # Pre-build **latest** DeepSpeed, so it would be ready for testing (otherwise, the 1st deepspeed test will timeout)
 RUN python3 -m pip uninstall -y deepspeed
 # This has to be run (again) inside the GPU VMs running the tests.
 # The installation works here, but some tests fail, if we don't pre-build deepspeed again in the VMs running the tests.
 # TODO: Find out why test fail.
-RUN DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 python3 -m pip install "deepspeed<=0.14.0" --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check 2>&1
+RUN DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 DS_BUILD_AIO=1 DS_BUILD_UTILS=1 python3 -m pip install deepspeed --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check 2>&1
 
 # When installing in editable mode, `transformers` is not recognized as a package.
 # this line must be added in order for python to be aware of transformers.
 RUN cd transformers && python3 setup.py develop
 
-# The base image ships with `pydantic==1.8.2` which is not working - i.e. the next command fails
-RUN python3 -m pip install -U --no-cache-dir "pydantic<2"
 RUN python3 -c "from deepspeed.launcher.runner import main"

docker/transformers-pytorch-deepspeed-nightly-gpu/Dockerfile

@@ -1,11 +1,10 @@
-# https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-23-11.html#rel-23-11
-FROM nvcr.io/nvidia/pytorch:23.11-py3
+FROM nvcr.io/nvidia/pytorch:21.03-py3
 LABEL maintainer="Hugging Face"
 
 ARG DEBIAN_FRONTEND=noninteractive
 
 # Example: `cu102`, `cu113`, etc.
-ARG CUDA='cu121'
+ARG CUDA='cu113'
 
 RUN apt -y update
 RUN apt install -y libaio-dev
@@ -14,8 +13,6 @@ RUN python3 -m pip install --no-cache-dir --upgrade pip
 ARG REF=main
 RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
 
-RUN python3 -m pip uninstall -y torch torchvision torchaudio
-
 # Install **nightly** release PyTorch (flag `--pre`)
 # (PyTorch must be installed before pre-compiling any DeepSpeed c++/cuda ops.)
 # (https://www.deepspeed.ai/tutorials/advanced-install/#pre-install-deepspeed-ops)
@@ -23,38 +20,12 @@ RUN python3 -m pip install --no-cache-dir -U --pre torch torchvision torchaudio
 
 RUN python3 -m pip install --no-cache-dir ./transformers[deepspeed-testing]
 
-RUN python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
-
-# Uninstall `transformer-engine` shipped with the base image
-RUN python3 -m pip uninstall -y transformer-engine
-
-# Uninstall `torch-tensorrt` and `apex` shipped with the base image
-RUN python3 -m pip uninstall -y torch-tensorrt apex
-
 # Pre-build **nightly** release of DeepSpeed, so it would be ready for testing (otherwise, the 1st deepspeed test will timeout)
 RUN python3 -m pip uninstall -y deepspeed
 # This has to be run inside the GPU VMs running the tests. (So far, it fails here due to GPU checks during compilation.)
 # Issue: https://github.com/microsoft/DeepSpeed/issues/2010
 # RUN git clone https://github.com/microsoft/DeepSpeed && cd DeepSpeed && rm -rf build && \
-#    DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 DS_BUILD_UTILS=1 python3 -m pip install . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check 2>&1
-
-## For `torchdynamo` tests
-## (see https://github.com/huggingface/transformers/pull/17765)
-#RUN git clone https://github.com/pytorch/functorch
-#RUN python3 -m pip install --no-cache-dir ./functorch[aot]
-#RUN cd functorch && python3 setup.py develop
-#
-#RUN git clone https://github.com/pytorch/torchdynamo
-#RUN python3 -m pip install -r ./torchdynamo/requirements.txt
-#RUN cd torchdynamo && python3 setup.py develop
-#
-## install TensorRT
-#RUN python3 -m pip install --no-cache-dir -U nvidia-pyindex
-#RUN python3 -m pip install --no-cache-dir -U nvidia-tensorrt==8.2.4.2
-#
-## install torch_tensorrt (fx path)
-#RUN git clone https://github.com/pytorch/TensorRT.git
-#RUN cd TensorRT/py && python3 setup.py install --fx-only
+#    DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 DS_BUILD_AIO=1 DS_BUILD_UTILS=1 python3 -m pip install . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check 2>&1
 
 # When installing in editable mode, `transformers` is not recognized as a package.
 # this line must be added in order for python to be aware of transformers.

docker/transformers-pytorch-gpu/Dockerfile

@@ -1,4 +1,4 @@
-FROM nvidia/cuda:12.1.0-cudnn8-devel-ubuntu20.04
+FROM nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04
 LABEL maintainer="Hugging Face"
 
 ARG DEBIAN_FRONTEND=noninteractive
@@ -9,23 +9,21 @@ RUN python3 -m pip install --no-cache-dir --upgrade pip
 
 ARG REF=main
 RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
+RUN python3 -m pip install --no-cache-dir -e ./transformers[dev-torch,testing]
 
 # If set to nothing, will install the latest version
-ARG PYTORCH='2.4.0'
+ARG PYTORCH='1.12.1'
 ARG TORCH_VISION=''
 ARG TORCH_AUDIO=''
-# Example: `cu102`, `cu113`, etc.
-ARG CUDA='cu121'
 
-RUN [ ${#PYTORCH} -gt 0 ] && VERSION='torch=='$PYTORCH'.*' ||  VERSION='torch'; python3 -m pip install --no-cache-dir -U $VERSION --extra-index-url https://download.pytorch.org/whl/$CUDA
-RUN [ ${#TORCH_VISION} -gt 0 ] && VERSION='torchvision=='TORCH_VISION'.*' ||  VERSION='torchvision'; python3 -m pip install --no-cache-dir -U $VERSION --extra-index-url https://download.pytorch.org/whl/$CUDA
-RUN [ ${#TORCH_AUDIO} -gt 0 ] && VERSION='torchaudio=='TORCH_AUDIO'.*' ||  VERSION='torchaudio'; python3 -m pip install --no-cache-dir -U $VERSION --extra-index-url https://download.pytorch.org/whl/$CUDA
-
-RUN python3 -m pip install --no-cache-dir -e ./transformers[dev-torch,testing,video]
+RUN [ ${#PYTORCH} -gt 0 ] && VERSION='torch=='$PYTORCH'.*' ||  VERSION='torch'; python3 -m pip install --no-cache-dir -U $VERSION --extra-index-url https://download.pytorch.org/whl/cu113
+RUN [ ${#TORCH_VISION} -gt 0 ] && VERSION='torchvision=='TORCH_VISION'.*' ||  VERSION='torchvision'; python3 -m pip install --no-cache-dir -U $VERSION --extra-index-url https://download.pytorch.org/whl/cu113
+RUN [ ${#TORCH_AUDIO} -gt 0 ] && VERSION='torchaudio=='TORCH_AUDIO'.*' ||  VERSION='torchaudio'; python3 -m pip install --no-cache-dir -U $VERSION --extra-index-url https://download.pytorch.org/whl/cu113
 
 RUN python3 -m pip uninstall -y tensorflow flax
 
-RUN python3 -m pip install --no-cache-dir git+https://github.com/facebookresearch/detectron2.git pytesseract
+RUN python3 -m pip install --no-cache-dir torch-scatter -f https://data.pyg.org/whl/torch-$(python3 -c "from torch import version; print(version.__version__.split('+')[0])")+cu113.html
+RUN python3 -m pip install --no-cache-dir git+https://github.com/facebookresearch/detectron2.git pytesseract https://github.com/kpu/kenlm/archive/master.zip
 RUN python3 -m pip install -U "itsdangerous<2.1.0"
 
 # When installing in editable mode, `transformers` is not recognized as a package.

docker/transformers-quantization-latest-gpu/Dockerfile

@@ -1,66 +0,0 @@
-FROM nvidia/cuda:11.8.0-cudnn8-devel-ubuntu20.04
-LABEL maintainer="Hugging Face"
-
-ARG DEBIAN_FRONTEND=noninteractive
-
-# Use login shell to read variables from `~/.profile` (to pass dynamic created variables between RUN commands)
-SHELL ["sh", "-lc"]
-
-# The following `ARG` are mainly used to specify the versions explicitly & directly in this docker file, and not meant
-# to be used as arguments for docker build (so far).
-
-ARG PYTORCH='2.2.1'
-# Example: `cu102`, `cu113`, etc.
-ARG CUDA='cu118'
-
-RUN apt update
-RUN apt install -y git libsndfile1-dev tesseract-ocr espeak-ng python python3-pip ffmpeg
-RUN python3 -m pip install --no-cache-dir --upgrade pip
-
-ARG REF=main
-RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
-
-RUN [ ${#PYTORCH} -gt 0 ] && VERSION='torch=='$PYTORCH'.*' ||  VERSION='torch'; echo "export VERSION='$VERSION'" >> ~/.profile
-RUN echo torch=$VERSION
-# `torchvision` and `torchaudio` should be installed along with `torch`, especially for nightly build.
-# Currently, let's just use their latest releases (when `torch` is installed with a release version)
-RUN python3 -m pip install --no-cache-dir -U $VERSION torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/$CUDA
-
-RUN python3 -m pip install --no-cache-dir -e ./transformers[dev-torch]
-
-RUN python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
-
-# needed in bnb and awq
-RUN python3 -m pip install --no-cache-dir einops
-
-# Add bitsandbytes for mixed int8 testing
-RUN python3 -m pip install --no-cache-dir bitsandbytes
-
-# Add auto-gptq for gtpq quantization testing
-RUN python3 -m pip install --no-cache-dir auto-gptq --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/
-
-# Add optimum for gptq quantization testing
-RUN python3 -m pip install --no-cache-dir git+https://github.com/huggingface/optimum@main#egg=optimum
-
-# Add aqlm for quantization testing
-RUN python3 -m pip install --no-cache-dir aqlm[gpu]==1.0.2
-
-# Add hqq for quantization testing
-RUN python3 -m pip install --no-cache-dir hqq
-
-# For GGUF tests
-RUN python3 -m pip install --no-cache-dir gguf
-
-# Add autoawq for quantization testing
-# >=v0.2.3 needed for compatibility with torch 2.2.1
-RUN python3 -m pip install --no-cache-dir https://github.com/casper-hansen/AutoAWQ/releases/download/v0.2.3/autoawq-0.2.3+cu118-cp38-cp38-linux_x86_64.whl
-
-# Add quanto for quantization testing
-RUN python3 -m pip install --no-cache-dir quanto
-
-# Add eetq for quantization testing
-RUN python3 -m pip install git+https://github.com/NetEase-FuXi/EETQ.git
-
-# When installing in editable mode, `transformers` is not recognized as a package.
-# this line must be added in order for python to be aware of transformers.
-RUN cd transformers && python3 setup.py develop

docker/transformers-tensorflow-cpu/Dockerfile

@@ -0,0 +1,25 @@
+FROM ubuntu:18.04
+LABEL maintainer="Hugging Face"
+LABEL repository="transformers"
+
+RUN apt update && \
+    apt install -y bash \
+                   build-essential \
+                   git \
+                   curl \
+                   ca-certificates \
+                   python3 \
+                   python3-pip && \
+    rm -rf /var/lib/apt/lists
+
+RUN python3 -m pip install --no-cache-dir --upgrade pip && \
+    python3 -m pip install --no-cache-dir \
+    mkl \
+    tensorflow-cpu
+
+WORKDIR /workspace
+COPY . transformers/
+RUN cd transformers/ && \
+    python3 -m pip install --no-cache-dir .
+
+CMD ["/bin/bash"]

docker/transformers-tensorflow-gpu/Dockerfile

@@ -1,4 +1,4 @@
-FROM nvidia/cuda:12.1.0-cudnn8-devel-ubuntu20.04
+FROM nvidia/cuda:11.2.2-cudnn8-devel-ubuntu20.04
 LABEL maintainer="Hugging Face"
 
 ARG DEBIAN_FRONTEND=noninteractive
@@ -12,14 +12,12 @@ RUN git clone https://github.com/huggingface/transformers && cd transformers &&
 RUN python3 -m pip install --no-cache-dir -e ./transformers[dev-tensorflow,testing]
 
 # If set to nothing, will install the latest version
-ARG TENSORFLOW='2.13'
+ARG TENSORFLOW=''
 
 RUN [ ${#TENSORFLOW} -gt 0 ] && VERSION='tensorflow=='$TENSORFLOW'.*' ||  VERSION='tensorflow'; python3 -m pip install --no-cache-dir -U $VERSION
 RUN python3 -m pip uninstall -y torch flax
 RUN python3 -m pip install -U "itsdangerous<2.1.0"
 
-RUN python3 -m pip install --no-cache-dir -U tensorflow_probability
-
 # When installing in editable mode, `transformers` is not recognized as a package.
 # this line must be added in order for python to be aware of transformers.
 RUN cd transformers && python3 setup.py develop

docs/README.md

@@ -16,7 +16,7 @@ limitations under the License.
 
 # Generating the documentation
 
-To generate the documentation, you first have to build it. Several packages are necessary to build the doc, 
+To generate the documentation, you first have to build it. Several packages are necessary to build the doc,
 you can install them with the following command, at the root of the code repository:
 
 ```bash
@@ -33,7 +33,7 @@ pip install git+https://github.com/huggingface/doc-builder
 **NOTE**
 
 You only need to generate the documentation to inspect it locally (if you're planning changes and want to
-check how they look before committing for instance). You don't have to commit the built documentation.
+check how they look like before committing for instance). You don't have to commit the built documentation.
 
 ---
 
@@ -81,16 +81,16 @@ The `preview` command only works with existing doc files. When you add a complet
 
 ## Adding a new element to the navigation bar
 
-Accepted files are Markdown (.md).
+Accepted files are Markdown (.md or .mdx).
 
 Create a file with its extension and put it in the source directory. You can then link it to the toc-tree by putting
-the filename without the extension in the [`_toctree.yml`](https://github.com/huggingface/transformers/blob/main/docs/source/en/_toctree.yml) file.
+the filename without the extension in the [`_toctree.yml`](https://github.com/huggingface/transformers/blob/main/docs/source/_toctree.yml) file.
 
 ## Renaming section headers and moving sections
 
-It helps to keep the old links working when renaming the section header and/or moving sections from one document to another. This is because the old links are likely to be used in Issues, Forums, and Social media and it'd make for a much more superior user experience if users reading those months later could still easily navigate to the originally intended information.
+It helps to keep the old links working when renaming section header and/or moving sections from one document to another. This is because the old links are likely to be used in Issues, Forums and Social media and it'd be make for a much more superior user experience if users reading those months later could still easily navigate to the originally intended information.
 
-Therefore, we simply keep a little map of moved sections at the end of the document where the original section was. The key is to preserve the original anchor.
+Therefore we simply keep a little map of moved sections at the end of the document where the original section was. The key is to preserve the original anchor.
 
 So if you renamed a section from: "Section A" to "Section B", then you can add at the end of the file:
 
@@ -99,7 +99,7 @@ Sections that were moved:
 
 [ <a href="#section-b">Section A</a><a id="section-a"></a> ]
 ```
-and of course, if you moved it to another file, then:
+and of course if you moved it to another file, then:
 
 ```
 Sections that were moved:
@@ -109,7 +109,7 @@ Sections that were moved:
 
 Use the relative style to link to the new file so that the versioned docs continue to work.
 
-For an example of a rich moved section set please see the very end of [the Trainer doc](https://github.com/huggingface/transformers/blob/main/docs/source/en/main_classes/trainer.md).
+For an example of a rich moved sections set please see the very end of [the Trainer doc](https://github.com/huggingface/transformers/blob/main/docs/source/main_classes/trainer.mdx).
 
 
 ## Writing Documentation - Specification
@@ -126,7 +126,7 @@ Adding a new tutorial or section is done in two steps:
 - Link that file in `./source/_toctree.yml` on the correct toc-tree.
 
 Make sure to put your new file under the proper section. It's unlikely to go in the first section (*Get Started*), so
-depending on the intended targets (beginners, more advanced users, or researchers) it should go in sections two, three, or
+depending on the intended targets (beginners, more advanced users or researchers) it should go in section two, three or
 four.
 
 ### Translating
@@ -138,7 +138,7 @@ When translating, refer to the guide at [./TRANSLATING.md](https://github.com/hu
 
 When adding a new model:
 
-- Create a file `xxx.md` or under `./source/model_doc` (don't hesitate to copy an existing file as template).
+- Create a file `xxx.mdx` or under `./source/model_doc` (don't hesitate to copy an existing file as template).
 - Link that file in `./source/_toctree.yml`.
 - Write a short overview of the model:
     - Overview with paper & authors
@@ -147,7 +147,7 @@ When adding a new model:
 - Add the classes that should be linked in the model. This generally includes the configuration, the tokenizer, and
   every model of that class (the base model, alongside models with additional heads), both in PyTorch and TensorFlow.
   The order is generally:
-    - Configuration
+    - Configuration,
     - Tokenizer
     - PyTorch base model
     - PyTorch head models
@@ -177,8 +177,8 @@ not to be displayed in the documentation, you can do so by specifying which meth
     - save_vocabulary
 ```
 
-If you just want to add a method that is not documented (for instance magic methods like `__call__` are not documented
-by default) you can put the list of methods to add in a list that contains `all`:
+If you just want to add a method that is not documented (for instance magic method like `__call__` are not documented
+byt default) you can put the list of methods to add in a list that contains `all`:
 
 ```
 ## XXXTokenizer
@@ -191,9 +191,9 @@ by default) you can put the list of methods to add in a list that contains `all`
 ### Writing source documentation
 
 Values that should be put in `code` should either be surrounded by backticks: \`like so\`. Note that argument names
-and objects like True, None, or any strings should usually be put in `code`.
+and objects like True, None or any strings should usually be put in `code`.
 
-When mentioning a class, function, or method, it is recommended to use our syntax for internal links so that our tool
+When mentioning a class, function or method, it is recommended to use our syntax for internal links so that our tool
 adds a link to its documentation with this syntax: \[\`XXXClass\`\] or \[\`function\`\]. This requires the class or 
 function to be in the main package.
 
@@ -202,12 +202,12 @@ provide its path. For instance: \[\`utils.ModelOutput\`\]. This will be converte
 `utils.ModelOutput` in the description. To get rid of the path and only keep the name of the object you are
 linking to in the description, add a ~: \[\`~utils.ModelOutput\`\] will generate a link with `ModelOutput` in the description.
 
-The same works for methods so you can either use \[\`XXXClass.method\`\] or \[\`~XXXClass.method\`\].
+The same works for methods so you can either use \[\`XXXClass.method\`\] or \[~\`XXXClass.method\`\].
 
 #### Defining arguments in a method
 
 Arguments should be defined with the `Args:` (or `Arguments:` or `Parameters:`) prefix, followed by a line return and
-an indentation. The argument should be followed by its type, with its shape if it is a tensor, a colon, and its
+an indentation. The argument should be followed by its type, with its shape if it is a tensor, a colon and its
 description:
 
 ```
@@ -216,7 +216,7 @@ description:
 ```
 
 If the description is too long to fit in one line, another indentation is necessary before writing the description
-after the argument.
+after th argument.
 
 Here's an example showcasing everything so far:
 
@@ -250,7 +250,7 @@ then its documentation should look like this:
 
 Note that we always omit the "defaults to \`None\`" when None is the default for any argument. Also note that even
 if the first line describing your argument type and its default gets long, you can't break it on several lines. You can
-however, write as many lines as you want in the indented description (see the example above with `input_ids`).
+however write as many lines as you want in the indented description (see the example above with `input_ids`).
 
 #### Writing a multi-line code block
 
@@ -266,7 +266,7 @@ Multi-line code blocks can be useful for displaying examples. They are done betw
 ````
 
 We follow the [doctest](https://docs.python.org/3/library/doctest.html) syntax for the examples to automatically test
-the results to stay consistent with the library.
+the results stay consistent with the library.
 
 #### Writing a return block
 
@@ -274,27 +274,27 @@ The return block should be introduced with the `Returns:` prefix, followed by a
 The first line should be the type of the return, followed by a line return. No need to indent further for the elements
 building the return.
 
-Here's an example of a single value return:
+Here's an example for a single value return:
 
 ```
     Returns:
         `List[int]`: A list of integers in the range [0, 1] --- 1 for a special token, 0 for a sequence token.
 ```
 
-Here's an example of a tuple return, comprising several objects:
+Here's an example for tuple return, comprising several objects:
 
 ```
     Returns:
         `tuple(torch.FloatTensor)` comprising various elements depending on the configuration ([`BertConfig`]) and inputs:
         - ** loss** (*optional*, returned when `masked_lm_labels` is provided) `torch.FloatTensor` of shape `(1,)` --
-          Total loss is the sum of the masked language modeling loss and the next sequence prediction (classification) loss.
+          Total loss as the sum of the masked language modeling loss and the next sequence prediction (classification) loss.
         - **prediction_scores** (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`) --
           Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
 ```
 
 #### Adding an image
 
-Due to the rapidly growing repository, it is important to make sure that no files that would significantly weigh down the repository are added. This includes images, videos, and other non-text files. We prefer to leverage a hf.co hosted `dataset` like
+Due to the rapidly growing repository, it is important to make sure that no files that would significantly weigh down the repository are added. This includes images, videos and other non-text files. We prefer to leverage a hf.co hosted `dataset` like
 the ones hosted on [`hf-internal-testing`](https://huggingface.co/hf-internal-testing) in which to place these files and reference
 them by URL. We recommend putting them in the following dataset: [huggingface/documentation-images](https://huggingface.co/datasets/huggingface/documentation-images).
 If an external contribution, feel free to add the images to your PR and ask a Hugging Face member to migrate your images
@@ -312,13 +312,13 @@ easily.
 
 # Testing documentation examples
 
-Good documentation often comes with an example of how a specific function or class should be used. 
+Good documentation oftens comes with an example of how a specific function or class should be used. 
 Each model class should contain at least one example showcasing
 how to use this model class in inference. *E.g.* the class [Wav2Vec2ForCTC](https://huggingface.co/docs/transformers/model_doc/wav2vec2#transformers.Wav2Vec2ForCTC) 
 includes an example of how to transcribe speech to text in the 
 [docstring of its forward function](https://huggingface.co/docs/transformers/model_doc/wav2vec2#transformers.Wav2Vec2ForCTC.forward).
 
-## Writing documentation examples
+## Writing documenation examples
 
 The syntax for Example docstrings can look as follows:
 
@@ -354,7 +354,7 @@ The docstring should give a minimal, clear example of how the respective model
 is to be used in inference and also include the expected (ideally sensible)
 output.
 Often, readers will try out the example before even going through the function 
-or class definitions. Therefore, it is of utmost importance that the example 
+or class definitions. Therefore it is of utmost importance that the example 
 works as expected.
 
 ## Docstring testing
@@ -364,9 +364,25 @@ We use pytests' [doctest integration](https://docs.pytest.org/doctest.html) to v
 For Transformers, the doctests are run on a daily basis via GitHub Actions as can be 
 seen [here](https://github.com/huggingface/transformers/actions/workflows/doctests.yml).
 
+To include your example in the daily doctests, you need add the filename that
+contains the example docstring to the [documentation_tests.txt](../utils/documentation_tests.txt).
+
 ### For Python files
 
-Run all the tests in the docstrings of a given file with the following command, here is how we test the modeling file of Wav2Vec2 for instance:
+You will first need to run the following command (from the root of the repository) to prepare the doc file (doc-testing needs to add additional lines that we don't include in the doc source files):
+
+```bash
+python utils/prepare_for_doc_test.py src docs
+```
+
+If you work on a specific python module, say `modeling_wav2vec2.py`, you can run the command as follows (to avoid the unnecessary temporary changes in irrelevant files):
+
+```bash
+python utils/prepare_for_doc_test.py src/transformers/utils/doc.py src/transformers/models/wav2vec2/modeling_wav2vec2.py
+```
+(`utils/doc.py` should always be included)
+
+Then you can run all the tests in the docstrings of a given file with the following command, here is how we test the modeling file of Wav2Vec2 for instance:
 
 ```bash
 pytest --doctest-modules src/transformers/models/wav2vec2/modeling_wav2vec2.py -sv --doctest-continue-on-failure
@@ -378,12 +394,30 @@ If you want to isolate a specific docstring, just add `::` after the file name t
 pytest --doctest-modules src/transformers/models/wav2vec2/modeling_wav2vec2.py::transformers.models.wav2vec2.modeling_wav2vec2.Wav2Vec2ForCTC.forward -sv --doctest-continue-on-failure
 ```
 
-### For Markdown files
-
-You can test locally a given file with this command (here testing the quicktour):
+Once you're done, you can run the following command (still from the root of the repository) to undo the changes made by the first command before committing:
 
 ```bash
-pytest --doctest-modules docs/source/quicktour.md -sv --doctest-continue-on-failure --doctest-glob="*.md"
+python utils/prepare_for_doc_test.py src docs --remove_new_line
+```
+
+### For Markdown files
+
+You will first need to run the following command (from the root of the repository) to prepare the doc file (doc-testing needs to add additional lines that we don't include in the doc source files):
+
+```bash
+python utils/prepare_for_doc_test.py src docs
+```
+
+Then you can test locally a given file with this command (here testing the quicktour):
+
+```bash
+pytest --doctest-modules docs/source/quicktour.mdx -sv --doctest-continue-on-failure --doctest-glob="*.mdx"
+```
+
+Once you're done, you can run the following command (still from the root of the repository) to undo the changes made by the first command before committing:
+
+```bash
+python utils/prepare_for_doc_test.py src docs --remove_new_line
 ```
 
 ### Writing doctests
@@ -392,6 +426,6 @@ Here are a few tips to help you debug the doctests and make them pass:
 
 - The outputs of the code need to match the expected output **exactly**, so make sure you have the same outputs. In particular doctest will see a difference between single quotes and double quotes, or a missing parenthesis. The only exceptions to that rule are:
   * whitespace: one give whitespace (space, tabulation, new line) is equivalent to any number of whitespace, so you can add new lines where there are spaces to make your output more readable.
-  * numerical values: you should never put more than 4 or 5 digits to expected results as different setups or library versions might get you slightly different results. `doctest` is configured to ignore any difference lower than the precision to which you wrote (so 1e-4 if you write 4 digits).
+  * numerical values: you should never put more than 4 or 5 digits to expected results as different setups or library versions might get you slightly different results. `doctest` is configure to ignore any difference lower than the precision to which you wrote (so 1e-4 if you write 4 digits).
 - Don't leave a block of code that is very long to execute. If you can't make it fast, you can either not use the doctest syntax on it (so that it's ignored), or if you want to use the doctest syntax to show the results, you can add a comment `# doctest: +SKIP` at the end of the lines of code too long to execute
 - Each line of code that produces a result needs to have that result written below. You can ignore an output if you don't want to show it in your code example by adding a comment ` # doctest: +IGNORE_RESULT` at the end of the line of code producing it.

docs/TRANSLATING.md

@@ -54,4 +54,4 @@ The fields you should add are `local` (with the name of the file containing the
 
 Once you have translated the `_toctree.yml` file, you can start translating the [MDX](https://mdxjs.com/) files associated with your docs chapter.
 
-> 🙋 If you'd like others to help you with the translation, you should [open an issue](https://github.com/huggingface/transformers/issues) and tag @stevhliu and @MKhalusova.
+> 🙋 If you'd like others to help you with the translation, you should [open an issue](https://github.com/huggingface/transformers/issues) and tag @sgugger.

docs/source/_config.py

@@ -1,7 +1,7 @@
 # docstyle-ignore
 INSTALL_CONTENT = """
 # Transformers installation
-! pip install transformers datasets evaluate accelerate
+! pip install transformers datasets
 # To install from source instead of the last release, comment the command above and uncomment the following one.
 # ! pip install git+https://github.com/huggingface/transformers.git
 """
@@ -10,5 +10,5 @@ notebook_first_cells = [{"type": "code", "content": INSTALL_CONTENT}]
 black_avoid_patterns = {
     "{processor_class}": "FakeProcessorClass",
     "{model_class}": "FakeModelClass",
-    "{object_class}": "FakeObjectClass",
+    "{object_class}": "FakeObjectClass",    
 }

docs/source/de/_config.py

@@ -1,7 +1,7 @@
 # docstyle-ignore
 INSTALL_CONTENT = """
 # Transformers installation
-! pip install transformers datasets evaluate accelerate
+! pip install transformers datasets
 # To install from source instead of the last release, comment the command above and uncomment the following one.
 # ! pip install git+https://github.com/huggingface/transformers.git
 """

docs/source/de/_toctree.yml

@@ -9,34 +9,4 @@
 - sections:
   - local: pipeline_tutorial
     title: Pipelines für Inferenzen
-  - local: autoclass_tutorial
-    title: Laden von vortrainierten Instanzen mit einer AutoClass
-  - local: preprocessing
-    title: Vorverarbeiten
-  - local: training
-    title: Optimierung eines vortrainierten Modells
-  - local: run_scripts
-    title: Trainieren mit einem Skript
-  - local: accelerate
-    title: Verteiltes Training mit 🤗 Accelerate
-  - local: peft
-    title: Laden und Trainieren von Adaptern mit 🤗 PEFT
-  - local: model_sharing
-    title: Ein Modell teilen
-  - local: transformers_agents
-    title: Agents
-  - local: llm_tutorial
-    title: Generation with LLMs
   title: Tutorials
-- sections:
-  - local: contributing
-    title: Wie kann man zu 🤗 Transformers beitragen?
-  - local: add_new_model
-    title: Wie fügt man ein Modell zu 🤗 Transformers hinzu?
-  - local: add_new_pipeline
-    title: Wie fügt man eine Pipeline zu 🤗 Transformers hinzu?
-  - local: testing
-    title: Testen
-  - local: pr_checks
-    title: Überprüfung einer Pull Request
-  title: Contribute

docs/source/de/accelerate.md

@@ -1,136 +0,0 @@
-<!--Copyright 2022 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
-rendered properly in your Markdown viewer.
-
--->
-
-# Verteiltes Training mit 🤗 Accelerate
-
-Da die Modelle immer größer werden, hat sich die Parallelität als Strategie zum Trainieren größerer Modelle auf begrenzter Hardware und zur Beschleunigung der Trainingsgeschwindigkeit um mehrere Größenordnungen erwiesen. Bei Hugging Face haben wir die Bibliothek [🤗 Accelerate](https://huggingface.co/docs/accelerate) entwickelt, um Nutzern zu helfen, ein 🤗 Transformers-Modell auf jeder Art von verteiltem Setup zu trainieren, egal ob es sich um mehrere GPUs auf einer Maschine oder mehrere GPUs auf mehreren Maschinen handelt. In diesem Tutorial lernen Sie, wie Sie Ihre native PyTorch-Trainingsschleife anpassen, um das Training in einer verteilten Umgebung zu ermöglichen.
-
-## Einrichtung
-
-Beginnen Sie mit der Installation von 🤗 Accelerate:
-
-```bash
-pip install accelerate
-```
-
-Dann importieren und erstellen Sie ein [`~accelerate.Accelerator`]-Objekt. Der [`~accelerate.Accelerator`] wird automatisch Ihre Art der verteilten Einrichtung erkennen und alle notwendigen Komponenten für das Training initialisieren. Sie müssen Ihr Modell nicht explizit auf einem Gerät platzieren.
-
-```py
->>> from accelerate import Accelerator
-
->>> accelerator = Accelerator()
-```
-
-## Vorbereiten auf die Beschleunigung
-
-Der nächste Schritt ist die Übergabe aller relevanten Trainingsobjekte an die Methode [`~accelerate.Accelerator.prepare`]. Dazu gehören Ihre Trainings- und Evaluierungs-DataLoader, ein Modell und ein Optimierer:
-
-```py
->>> train_dataloader, eval_dataloader, model, optimizer = accelerator.prepare(
-...     train_dataloader, eval_dataloader, model, optimizer
-... )
-```
-
-## Rückwärts
-
-Die letzte Ergänzung besteht darin, das typische `loss.backward()` in der Trainingsschleife durch die 🤗 Accelerate-Methode [`~accelerate.Accelerator.backward`] zu ersetzen:
-
-```py
->>> for epoch in range(num_epochs):
-...     for batch in train_dataloader:
-...         outputs = model(**batch)
-...         loss = outputs.loss
-...         accelerator.backward(loss)
-
-...         optimizer.step()
-...         lr_scheduler.step()
-...         optimizer.zero_grad()
-...         progress_bar.update(1)
-```
-
-Wie Sie im folgenden Code sehen können, müssen Sie nur vier zusätzliche Codezeilen zu Ihrer Trainingsschleife hinzufügen, um verteiltes Training zu ermöglichen!
-
-```diff
-+ from accelerate import Accelerator
-  from transformers import AdamW, AutoModelForSequenceClassification, get_scheduler
-
-+ accelerator = Accelerator()
-
-  model = AutoModelForSequenceClassification.from_pretrained(checkpoint, num_labels=2)
-  optimizer = AdamW(model.parameters(), lr=3e-5)
-
-- device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
-- model.to(device)
-
-+ train_dataloader, eval_dataloader, model, optimizer = accelerator.prepare(
-+     train_dataloader, eval_dataloader, model, optimizer
-+ )
-
-  num_epochs = 3
-  num_training_steps = num_epochs * len(train_dataloader)
-  lr_scheduler = get_scheduler(
-      "linear",
-      optimizer=optimizer,
-      num_warmup_steps=0,
-      num_training_steps=num_training_steps
-  )
-
-  progress_bar = tqdm(range(num_training_steps))
-
-  model.train()
-  for epoch in range(num_epochs):
-      for batch in train_dataloader:
--         batch = {k: v.to(device) for k, v in batch.items()}
-          outputs = model(**batch)
-          loss = outputs.loss
--         loss.backward()
-+         accelerator.backward(loss)
-
-          optimizer.step()
-          lr_scheduler.step()
-          optimizer.zero_grad()
-          progress_bar.update(1)
-```
-
-## Trainieren
-
-Sobald Sie die entsprechenden Codezeilen hinzugefügt haben, starten Sie Ihr Training in einem Skript oder einem Notebook wie Colaboratory.
-
-### Trainieren mit einem Skript
-
-Wenn Sie Ihr Training mit einem Skript durchführen, führen Sie den folgenden Befehl aus, um eine Konfigurationsdatei zu erstellen und zu speichern:
-
-```bash
-accelerate config
-```
-
-Dann starten Sie Ihr Training mit:
-
-```bash
-accelerate launch train.py
-```
-
-### Trainieren mit einem Notebook
-
-🤗 Accelerate kann auch in einem Notebook laufen, wenn Sie planen, die TPUs von Colaboratory zu verwenden. Verpacken Sie den gesamten Code, der für das Training verantwortlich ist, in eine Funktion und übergeben Sie diese an [`~accelerate.notebook_launcher`]:
-
-```py
->>> from accelerate import notebook_launcher
-
->>> notebook_launcher(training_function)
-```
-
-Weitere Informationen über 🤗 Accelerate und seine umfangreichen Funktionen finden Sie in der [Dokumentation](https://huggingface.co/docs/accelerate).

docs/source/de/add_new_model.md

@@ -1,891 +0,0 @@
-<!--Copyright 2020 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
-rendered properly in your Markdown viewer.
-
--->
-
-# Wie kann ich ein Modell zu 🤗 Transformers hinzufügen?
-
-Die 🤗 Transformers-Bibliothek ist dank der Beiträge der Community oft in der Lage, neue Modelle anzubieten. Aber das kann ein anspruchsvolles Projekt sein und erfordert eine eingehende Kenntnis der 🤗 Transformers-Bibliothek und des zu implementierenden Modells. Bei Hugging Face versuchen wir, mehr Mitgliedern der Community die Möglichkeit zu geben, aktiv Modelle hinzuzufügen, und wir haben diese Anleitung zusammengestellt, die Sie durch den Prozess des Hinzufügens eines PyTorch-Modells führt (stellen Sie sicher, dass Sie [PyTorch installiert haben](https://pytorch.org/get-started/locally/)).
-
-Auf dem Weg dorthin, werden Sie:
-
-- Einblicke in bewährte Open-Source-Verfahren erhalten
-- die Konstruktionsprinzipien hinter einer der beliebtesten Deep-Learning-Bibliotheken verstehen
-- lernen Sie, wie Sie große Modelle effizient testen können
-- lernen Sie, wie Sie Python-Hilfsprogramme wie `black`, `ruff` und `make fix-copies` integrieren, um sauberen und lesbaren Code zu gewährleisten
-
-Ein Mitglied des Hugging Face-Teams wird Ihnen dabei zur Seite stehen, damit Sie nicht alleine sind. 🤗 ❤️
-
-Um loszulegen, öffnen Sie eine [New model addition](https://github.com/huggingface/transformers/issues/new?assignees=&labels=New+model&template=new-model-addition.yml) Ausgabe für das Modell, das Sie in 🤗 Transformers sehen möchten. Wenn Sie nicht besonders wählerisch sind, wenn es darum geht, ein bestimmtes Modell beizusteuern, können Sie nach dem [New model label](https://github.com/huggingface/transformers/labels/New%20model) filtern, um zu sehen, ob es noch unbeanspruchte Modellanfragen gibt, und daran arbeiten.
-
-Sobald Sie eine neue Modellanfrage eröffnet haben, sollten Sie sich zunächst mit 🤗 Transformers vertraut machen, falls Sie das noch nicht sind!
-
-## Allgemeiner Überblick über 🤗 Transformers
-
-Zunächst sollten Sie sich einen allgemeinen Überblick über 🤗 Transformers verschaffen. 🤗 Transformers ist eine sehr meinungsfreudige Bibliothek, es ist also möglich, dass
-Es besteht also die Möglichkeit, dass Sie mit einigen der Philosophien oder Designentscheidungen der Bibliothek nicht einverstanden sind. Aus unserer Erfahrung heraus haben wir jedoch
-dass die grundlegenden Designentscheidungen und Philosophien der Bibliothek entscheidend sind, um 🤗 Transformers effizient zu skalieren.
-Transformatoren zu skalieren und gleichzeitig die Wartungskosten auf einem vernünftigen Niveau zu halten.
-
-Ein guter erster Ansatzpunkt, um die Bibliothek besser zu verstehen, ist die Lektüre der [Dokumentation unserer Philosophie](Philosophie). Als Ergebnis unserer Arbeitsweise gibt es einige Entscheidungen, die wir versuchen, auf alle Modelle anzuwenden:
-
-- Komposition wird im Allgemeinen gegenüber Abstraktion bevorzugt
-- Die Duplizierung von Code ist nicht immer schlecht, wenn sie die Lesbarkeit oder Zugänglichkeit eines Modells stark verbessert
-- Modelldateien sind so in sich geschlossen wie möglich, so dass Sie, wenn Sie den Code eines bestimmten Modells lesen, idealerweise nur
-  in die entsprechende Datei `modeling_....py` schauen müssen.
-
-Unserer Meinung nach ist der Code der Bibliothek nicht nur ein Mittel, um ein Produkt bereitzustellen, *z.B.* die Möglichkeit, BERT für
-Inferenz zu verwenden, sondern auch als das Produkt selbst, das wir verbessern wollen. Wenn Sie also ein Modell hinzufügen, ist der Benutzer nicht nur die
-Person, die Ihr Modell verwenden wird, sondern auch jeder, der Ihren Code liest, zu verstehen versucht und ihn möglicherweise verbessert.
-
-Lassen Sie uns daher ein wenig tiefer in das allgemeine Design der Bibliothek einsteigen.
-
-### Überblick über die Modelle
-
-Um ein Modell erfolgreich hinzuzufügen, ist es wichtig, die Interaktion zwischen Ihrem Modell und seiner Konfiguration zu verstehen,
-[`PreTrainedModel`] und [`PretrainedConfig`]. Als Beispiel werden wir
-das Modell, das zu 🤗 Transformers hinzugefügt werden soll, `BrandNewBert` nennen.
-
-Schauen wir uns das mal an:
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers_overview.png"/>
-
-Wie Sie sehen, machen wir in 🤗 Transformers von der Vererbung Gebrauch, aber wir beschränken die Abstraktionsebene auf ein absolutes Minimum.
-Minimum. Es gibt nie mehr als zwei Abstraktionsebenen für ein Modell in der Bibliothek. `BrandNewBertModel`
-erbt von `BrandNewBertPreTrainedModel`, das wiederum von [`PreTrainedModel`] erbt und
-das war's. In der Regel wollen wir sicherstellen, dass ein neues Modell nur von
-[`PreTrainedModel`] abhängt. Die wichtigen Funktionalitäten, die jedem neuen Modell automatisch zur Verfügung gestellt werden, sind
-Modell automatisch bereitgestellt werden, sind [`~PreTrainedModel.from_pretrained`] und
-[`~PreTrainedModel.save_pretrained`], die für die Serialisierung und Deserialisierung verwendet werden. Alle
-anderen wichtigen Funktionalitäten, wie `BrandNewBertModel.forward` sollten vollständig in der neuen
-Skript `modeling_brand_new_bert.py` definiert werden. Als nächstes wollen wir sicherstellen, dass ein Modell mit einer bestimmten Kopfebene, wie z.B.
-`BrandNewBertForMaskedLM` nicht von `BrandNewBertModel` erbt, sondern `BrandNewBertModel` verwendet
-als Komponente, die im Forward Pass aufgerufen werden kann, um die Abstraktionsebene niedrig zu halten. Jedes neue Modell erfordert eine
-Konfigurationsklasse, genannt `BrandNewBertConfig`. Diese Konfiguration wird immer als ein Attribut in
-[PreTrainedModel] gespeichert und kann daher über das Attribut `config` für alle Klassen aufgerufen werden
-die von `BrandNewBertPreTrainedModel` erben:
-
-```python
-model = BrandNewBertModel.from_pretrained("brandy/brand_new_bert")
-model.config  # model has access to its config
-```
-
-Ähnlich wie das Modell erbt die Konfiguration grundlegende Serialisierungs- und Deserialisierungsfunktionalitäten von
-[`PretrainedConfig`]. Beachten Sie, dass die Konfiguration und das Modell immer in zwei verschiedene Formate serialisiert werden
-unterschiedliche Formate serialisiert werden - das Modell in eine *pytorch_model.bin* Datei und die Konfiguration in eine *config.json* Datei. Aufruf von
-[`~PreTrainedModel.save_pretrained`] wird automatisch
-[`~PretrainedConfig.save_pretrained`] auf, so dass sowohl das Modell als auch die Konfiguration gespeichert werden.
-
-
-### Code-Stil
-
-Wenn Sie Ihr neues Modell kodieren, sollten Sie daran denken, dass Transformers eine Bibliothek mit vielen Meinungen ist und dass wir selbst ein paar Macken haben
-wie der Code geschrieben werden sollte :-)
-
-1. Der Vorwärtsdurchlauf Ihres Modells sollte vollständig in die Modellierungsdatei geschrieben werden und dabei völlig unabhängig von anderen
-   Modellen in der Bibliothek. Wenn Sie einen Block aus einem anderen Modell wiederverwenden möchten, kopieren Sie den Code und fügen ihn mit einem
-   `# Kopiert von` ein (siehe [hier](https://github.com/huggingface/transformers/blob/v4.17.0/src/transformers/models/roberta/modeling_roberta.py#L160)
-   für ein gutes Beispiel und [hier](pr_checks#check-copies) für weitere Dokumentation zu Copied from). 
-2. Der Code sollte vollständig verständlich sein, auch für einen Nicht-Muttersprachler. Das heißt, Sie sollten
-   beschreibende Variablennamen wählen und Abkürzungen vermeiden. Ein Beispiel: `activation` ist `act` vorzuziehen.
-   Von Variablennamen mit nur einem Buchstaben wird dringend abgeraten, es sei denn, es handelt sich um einen Index in einer for-Schleife.
-3. Generell ziehen wir längeren expliziten Code einem kurzen magischen Code vor.
-4. Vermeiden Sie die Unterklassifizierung von `nn.Sequential` in PyTorch, sondern unterklassifizieren Sie `nn.Module` und schreiben Sie den Vorwärtspass, so dass jeder
-   so dass jeder, der Ihren Code verwendet, ihn schnell debuggen kann, indem er Druckanweisungen oder Haltepunkte hinzufügt.
-5. Ihre Funktionssignatur sollte mit einer Typ-Annotation versehen sein. Im Übrigen sind gute Variablennamen viel lesbarer und verständlicher
-   verständlicher als Typ-Anmerkungen.
-
-### Übersicht der Tokenizer
-
-Noch nicht ganz fertig :-( Dieser Abschnitt wird bald hinzugefügt!
-
-## Schritt-für-Schritt-Rezept zum Hinzufügen eines Modells zu 🤗 Transformers
-
-Jeder hat andere Vorlieben, was die Portierung eines Modells angeht. Daher kann es sehr hilfreich sein, wenn Sie sich Zusammenfassungen ansehen
-wie andere Mitwirkende Modelle auf Hugging Face portiert haben. Hier ist eine Liste von Blogbeiträgen aus der Community, wie man ein Modell portiert:
-
-1. [Portierung eines GPT2-Modells](https://medium.com/huggingface/from-tensorflow-to-pytorch-265f40ef2a28) von [Thomas](https://huggingface.co/thomwolf)
-2. [Portierung des WMT19 MT-Modells](https://huggingface.co/blog/porting-fsmt) von [Stas](https://huggingface.co/stas)
-
-Aus Erfahrung können wir Ihnen sagen, dass die wichtigsten Dinge, die Sie beim Hinzufügen eines Modells beachten müssen, sind:
-
-- Erfinden Sie das Rad nicht neu! Die meisten Teile des Codes, den Sie für das neue 🤗 Transformers-Modell hinzufügen werden, existieren bereits
-  irgendwo in 🤗 Transformers. Nehmen Sie sich etwas Zeit, um ähnliche, bereits vorhandene Modelle und Tokenizer zu finden, die Sie kopieren können
-  von. [grep](https://www.gnu.org/software/grep/) und [rg](https://github.com/BurntSushi/ripgrep) sind Ihre
-  Freunde. Beachten Sie, dass es sehr gut möglich ist, dass der Tokenizer Ihres Modells auf einer Modellimplementierung basiert und
-  und der Modellierungscode Ihres Modells auf einer anderen. *Z.B.* Der Modellierungscode von FSMT basiert auf BART, während der Tokenizer-Code von FSMT
-  auf XLM basiert.
-- Es handelt sich eher um eine technische als um eine wissenschaftliche Herausforderung. Sie sollten mehr Zeit auf die Schaffung einer
-  eine effiziente Debugging-Umgebung zu schaffen, als zu versuchen, alle theoretischen Aspekte des Modells in dem Papier zu verstehen.
-- Bitten Sie um Hilfe, wenn Sie nicht weiterkommen! Modelle sind der Kernbestandteil von 🤗 Transformers, so dass wir bei Hugging Face mehr als
-  mehr als glücklich, Ihnen bei jedem Schritt zu helfen, um Ihr Modell hinzuzufügen. Zögern Sie nicht zu fragen, wenn Sie merken, dass Sie nicht weiterkommen.
-  Fortschritte machen.
-
-Im Folgenden versuchen wir, Ihnen ein allgemeines Rezept an die Hand zu geben, das uns bei der Portierung eines Modells auf 🤗 Transformers am nützlichsten erschien.
-
-Die folgende Liste ist eine Zusammenfassung all dessen, was getan werden muss, um ein Modell hinzuzufügen und kann von Ihnen als To-Do verwendet werden
-Liste verwenden:
-
-☐ (Optional) Verstehen der theoretischen Aspekte des Modells<br>
-☐ Vorbereiten der 🤗 Transformers-Entwicklungsumgebung<br>
-☐ Debugging-Umgebung des ursprünglichen Repositorys eingerichtet<br>
-☐ Skript erstellt, das den Durchlauf `forward()` unter Verwendung des ursprünglichen Repositorys und des Checkpoints erfolgreich durchführt<br>
-☐ Erfolgreich das Modellskelett zu 🤗 Transformers hinzugefügt<br>
-☐ Erfolgreiche Umwandlung des ursprünglichen Prüfpunkts in den 🤗 Transformers-Prüfpunkt<br>
-☐ Erfolgreich den Durchlauf `forward()` in 🤗 Transformers ausgeführt, der eine identische Ausgabe wie der ursprüngliche Prüfpunkt liefert<br>
-☐ Modell-Tests in 🤗 Transformers abgeschlossen<br>
-☐ Erfolgreich Tokenizer in 🤗 Transformers hinzugefügt<br>
-☐ End-to-End-Integrationstests ausgeführt<br>
-☐ Docs fertiggestellt<br>
-☐ Modellgewichte in den Hub hochgeladen<br>
-☐ Die Pull-Anfrage eingereicht<br>
-☐ (Optional) Hinzufügen eines Demo-Notizbuchs
-
-Für den Anfang empfehlen wir in der Regel, mit einem guten theoretischen Verständnis von `BrandNewBert` zu beginnen. Wie auch immer,
-wenn Sie es vorziehen, die theoretischen Aspekte des Modells *on-the-job* zu verstehen, dann ist es völlig in Ordnung, direkt in die
-in die Code-Basis von `BrandNewBert` einzutauchen. Diese Option könnte für Sie besser geeignet sein, wenn Ihre technischen Fähigkeiten besser sind als
-als Ihre theoretischen Fähigkeiten, wenn Sie Schwierigkeiten haben, die Arbeit von `BrandNewBert` zu verstehen, oder wenn Sie einfach Spaß am Programmieren
-mehr Spaß am Programmieren haben als am Lesen wissenschaftlicher Abhandlungen.
-
-### 1. (Optional) Theoretische Aspekte von BrandNewBert
-
-Sie sollten sich etwas Zeit nehmen, um die Abhandlung von *BrandNewBert* zu lesen, falls eine solche Beschreibung existiert. Möglicherweise gibt es große
-Abschnitte des Papiers, die schwer zu verstehen sind. Wenn das der Fall ist, ist das in Ordnung - machen Sie sich keine Sorgen! Das Ziel ist
-ist es nicht, ein tiefes theoretisches Verständnis des Papiers zu erlangen, sondern die notwendigen Informationen zu extrahieren, um
-das Modell effektiv in 🤗 Transformers zu implementieren. Das heißt, Sie müssen nicht zu viel Zeit auf die
-theoretischen Aspekten verbringen, sondern sich lieber auf die praktischen Aspekte konzentrieren, nämlich:
-
-- Welche Art von Modell ist *brand_new_bert*? BERT-ähnliches Modell nur für den Encoder? GPT2-ähnliches reines Decoder-Modell? BART-ähnliches
-  Encoder-Decoder-Modell? Sehen Sie sich die [model_summary](model_summary) an, wenn Sie mit den Unterschieden zwischen diesen Modellen nicht vertraut sind.
-- Was sind die Anwendungen von *brand_new_bert*? Textklassifizierung? Texterzeugung? Seq2Seq-Aufgaben, *z.B.,*
-  Zusammenfassungen?
-- Was ist die neue Eigenschaft des Modells, die es von BERT/GPT-2/BART unterscheidet?
-- Welches der bereits existierenden [🤗 Transformers-Modelle](https://huggingface.co/transformers/#contents) ist am ähnlichsten
-  ähnlich wie *brand_new_bert*?
-- Welche Art von Tokenizer wird verwendet? Ein Satzteil-Tokenisierer? Ein Wortstück-Tokenisierer? Ist es derselbe Tokenisierer, der für
-  für BERT oder BART?
-
-Nachdem Sie das Gefühl haben, einen guten Überblick über die Architektur des Modells erhalten zu haben, können Sie dem
-Hugging Face Team schreiben und Ihre Fragen stellen. Dazu können Fragen zur Architektur des Modells gehören,
-seiner Aufmerksamkeitsebene usw. Wir werden Ihnen gerne weiterhelfen.
-
-### 2. Bereiten Sie als nächstes Ihre Umgebung vor
-
-1. Forken Sie das [Repository](https://github.com/huggingface/transformers), indem Sie auf der Seite des Repositorys auf die Schaltfläche 'Fork' klicken.
-   Seite des Repositorys klicken. Dadurch wird eine Kopie des Codes unter Ihrem GitHub-Benutzerkonto erstellt.
-
-2. Klonen Sie Ihren `transformers` Fork auf Ihre lokale Festplatte und fügen Sie das Basis-Repository als Remote hinzu:
-
-```bash
-git clone https://github.com/[your Github handle]/transformers.git
-cd transformers
-git remote add upstream https://github.com/huggingface/transformers.git
-```
-
-3. Richten Sie eine Entwicklungsumgebung ein, indem Sie z.B. den folgenden Befehl ausführen:
-
-```bash
-python -m venv .env
-source .env/bin/activate
-pip install -e ".[dev]"
-```
-
-Abhängig von Ihrem Betriebssystem und da die Anzahl der optionalen Abhängigkeiten von Transformers wächst, kann es sein, dass Sie bei diesem Befehl einen
-Fehler mit diesem Befehl. Stellen Sie in diesem Fall sicher, dass Sie das Deep Learning Framework, mit dem Sie arbeiten, installieren
-(PyTorch, TensorFlow und/oder Flax) und führen Sie es aus:
-
-```bash
-pip install -e ".[quality]"
-```
-
-was für die meisten Anwendungsfälle ausreichend sein sollte. Sie können dann zum übergeordneten Verzeichnis zurückkehren
-
-```bash
-cd ..
-```
-
-4. Wir empfehlen, die PyTorch-Version von *brand_new_bert* zu Transformers hinzuzufügen. Um PyTorch zu installieren, folgen Sie bitte den
-   Anweisungen auf https://pytorch.org/get-started/locally/.
-
-**Anmerkung:** Sie müssen CUDA nicht installiert haben. Es reicht aus, das neue Modell auf der CPU zum Laufen zu bringen.
-
-5. Um *brand_new_bert* zu portieren, benötigen Sie außerdem Zugriff auf das Original-Repository:
-
-```bash
-git clone https://github.com/org_that_created_brand_new_bert_org/brand_new_bert.git
-cd brand_new_bert
-pip install -e .
-```
-
-Jetzt haben Sie eine Entwicklungsumgebung eingerichtet, um *brand_new_bert* auf 🤗 Transformers zu portieren.
-
-### 3.-4. Führen Sie einen Pre-Training-Checkpoint mit dem Original-Repository durch
-
-Zunächst werden Sie mit dem ursprünglichen *brand_new_bert* Repository arbeiten. Oft ist die ursprüngliche Implementierung sehr
-"forschungslastig". Das bedeutet, dass es an Dokumentation mangeln kann und der Code schwer zu verstehen sein kann. Aber das sollte
-genau Ihre Motivation sein, *brand_new_bert* neu zu implementieren. Eines unserer Hauptziele bei Hugging Face ist es, *die Menschen dazu zu bringen
-auf den Schultern von Giganten zu stehen*, was sich hier sehr gut darin ausdrückt, dass wir ein funktionierendes Modell nehmen und es umschreiben, um es so
-es so **zugänglich, benutzerfreundlich und schön** wie möglich zu machen. Dies ist die wichtigste Motivation für die Neuimplementierung von
-Modelle in 🤗 Transformers umzuwandeln - der Versuch, komplexe neue NLP-Technologie für **jeden** zugänglich zu machen.
-
-Sie sollten damit beginnen, indem Sie in das Original-Repository eintauchen.
-
-Die erfolgreiche Ausführung des offiziellen Pre-Trainingsmodells im Original-Repository ist oft **der schwierigste** Schritt.
-Unserer Erfahrung nach ist es sehr wichtig, dass Sie einige Zeit damit verbringen, sich mit der ursprünglichen Code-Basis vertraut zu machen. Sie müssen
-das Folgende herausfinden:
-
-- Wo finden Sie die vortrainierten Gewichte?
-- Wie lädt man die vorab trainierten Gewichte in das entsprechende Modell?
-- Wie kann der Tokenizer unabhängig vom Modell ausgeführt werden?
-- Verfolgen Sie einen Forward Pass, damit Sie wissen, welche Klassen und Funktionen für einen einfachen Forward Pass erforderlich sind. Normalerweise,
-  müssen Sie nur diese Funktionen reimplementieren.
-- Sie müssen in der Lage sein, die wichtigen Komponenten des Modells zu finden: Wo befindet sich die Klasse des Modells? Gibt es Unterklassen des Modells,
-  *z.B.* EncoderModel, DecoderModel? Wo befindet sich die Selbstaufmerksamkeitsschicht? Gibt es mehrere verschiedene Aufmerksamkeitsebenen,
-  *z.B.* *Selbstaufmerksamkeit*, *Kreuzaufmerksamkeit*...?
-- Wie können Sie das Modell in der ursprünglichen Umgebung des Repo debuggen? Müssen Sie *print* Anweisungen hinzufügen, können Sie
-  mit einem interaktiven Debugger wie *ipdb* arbeiten oder sollten Sie eine effiziente IDE zum Debuggen des Modells verwenden, wie z.B. PyCharm?
-
-Es ist sehr wichtig, dass Sie, bevor Sie mit der Portierung beginnen, den Code im Original-Repository **effizient** debuggen können
-Repository können! Denken Sie auch daran, dass Sie mit einer Open-Source-Bibliothek arbeiten, also zögern Sie nicht, ein Problem oder
-oder sogar eine Pull-Anfrage im Original-Repository zu stellen. Die Betreuer dieses Repositorys sind wahrscheinlich sehr froh darüber
-dass jemand in ihren Code schaut!
-
-An diesem Punkt liegt es wirklich an Ihnen, welche Debugging-Umgebung und Strategie Sie zum Debuggen des ursprünglichen
-Modell zu debuggen. Wir raten dringend davon ab, eine kostspielige GPU-Umgebung einzurichten, sondern arbeiten Sie einfach auf einer CPU, sowohl wenn Sie mit dem
-in das ursprüngliche Repository einzutauchen und auch, wenn Sie beginnen, die 🤗 Transformers-Implementierung des Modells zu schreiben. Nur
-ganz am Ende, wenn das Modell bereits erfolgreich auf 🤗 Transformers portiert wurde, sollte man überprüfen, ob das
-Modell auch auf der GPU wie erwartet funktioniert.
-
-Im Allgemeinen gibt es zwei mögliche Debugging-Umgebungen für die Ausführung des Originalmodells
-
-- [Jupyter notebooks](https://jupyter.org/) / [google colab](https://colab.research.google.com/notebooks/intro.ipynb)
-- Lokale Python-Skripte.
-
-Jupyter-Notebooks haben den Vorteil, dass sie eine zellenweise Ausführung ermöglichen, was hilfreich sein kann, um logische Komponenten besser voneinander zu trennen und
-logische Komponenten voneinander zu trennen und schnellere Debugging-Zyklen zu haben, da Zwischenergebnisse gespeichert werden können. Außerdem,
-Außerdem lassen sich Notebooks oft leichter mit anderen Mitwirkenden teilen, was sehr hilfreich sein kann, wenn Sie das Hugging Face Team um Hilfe bitten möchten.
-Face Team um Hilfe bitten. Wenn Sie mit Jupyter-Notizbüchern vertraut sind, empfehlen wir Ihnen dringend, mit ihnen zu arbeiten.
-
-Der offensichtliche Nachteil von Jupyter-Notizbüchern ist, dass Sie, wenn Sie nicht daran gewöhnt sind, mit ihnen zu arbeiten, einige Zeit damit verbringen müssen
-einige Zeit damit verbringen müssen, sich an die neue Programmierumgebung zu gewöhnen, und dass Sie möglicherweise Ihre bekannten Debugging-Tools nicht mehr verwenden können
-wie z.B. `ipdb` nicht mehr verwenden können.
-
-Für jede Codebasis ist es immer ein guter erster Schritt, einen **kleinen** vortrainierten Checkpoint zu laden und in der Lage zu sein, einen
-einzelnen Vorwärtsdurchlauf mit einem Dummy-Integer-Vektor von Eingabe-IDs als Eingabe zu reproduzieren. Ein solches Skript könnte wie folgt aussehen (in
-Pseudocode):
-
-```python
-model = BrandNewBertModel.load_pretrained_checkpoint("/path/to/checkpoint/")
-input_ids = [0, 4, 5, 2, 3, 7, 9]  # vector of input ids
-original_output = model.predict(input_ids)
-```
-
-Was die Debugging-Strategie anbelangt, so können Sie im Allgemeinen aus mehreren Strategien wählen:
-
-- Zerlegen Sie das ursprüngliche Modell in viele kleine testbare Komponenten und führen Sie für jede dieser Komponenten einen Vorwärtsdurchlauf zur
-  Überprüfung
-- Zerlegen Sie das ursprüngliche Modell nur in den ursprünglichen *Tokenizer* und das ursprüngliche *Modell*, führen Sie einen Vorwärtsdurchlauf für diese Komponenten durch
-  und verwenden Sie dazwischenliegende Druckanweisungen oder Haltepunkte zur Überprüfung.
-
-Auch hier bleibt es Ihnen überlassen, welche Strategie Sie wählen. Oft ist die eine oder die andere Strategie vorteilhaft, je nach der ursprünglichen Codebasis
-Basis.
-
-Wenn die ursprüngliche Codebasis es Ihnen erlaubt, das Modell in kleinere Teilkomponenten zu zerlegen, *z.B.* wenn die ursprüngliche
-Code-Basis problemlos im Eager-Modus ausgeführt werden kann, lohnt es sich in der Regel, dies zu tun. Es gibt einige wichtige Vorteile
-am Anfang den schwierigeren Weg zu gehen:
-
-- Wenn Sie später das ursprüngliche Modell mit der Hugging Face-Implementierung vergleichen, können Sie automatisch überprüfen, ob
-  für jede Komponente einzeln überprüfen, ob die entsprechende Komponente der 🤗 Transformers-Implementierung übereinstimmt, anstatt sich auf
-  anstatt sich auf den visuellen Vergleich über Druckanweisungen zu verlassen
-- können Sie das große Problem der Portierung eines Modells in kleinere Probleme der Portierung einzelner Komponenten zerlegen
-  einzelnen Komponenten zu zerlegen und so Ihre Arbeit besser zu strukturieren
-- Die Aufteilung des Modells in logisch sinnvolle Komponenten hilft Ihnen, einen besseren Überblick über das Design des Modells zu bekommen
-  und somit das Modell besser zu verstehen
-- In einem späteren Stadium helfen Ihnen diese komponentenweisen Tests dabei, sicherzustellen, dass keine Regressionen auftreten, während Sie fortfahren
-  Ihren Code ändern
-
-[Lysandre's](https://gist.github.com/LysandreJik/db4c948f6b4483960de5cbac598ad4ed) Integrationstests für ELECTRA
-gibt ein schönes Beispiel dafür, wie dies geschehen kann.
-
-Wenn die ursprüngliche Codebasis jedoch sehr komplex ist oder nur die Ausführung von Zwischenkomponenten in einem kompilierten Modus erlaubt,
-könnte es zu zeitaufwändig oder sogar unmöglich sein, das Modell in kleinere testbare Teilkomponenten zu zerlegen. Ein gutes
-Beispiel ist die [T5's MeshTensorFlow](https://github.com/tensorflow/mesh/tree/master/mesh_tensorflow) Bibliothek, die sehr komplex ist
-sehr komplex ist und keine einfache Möglichkeit bietet, das Modell in seine Unterkomponenten zu zerlegen. Bei solchen Bibliotheken ist man
-oft auf die Überprüfung von Druckanweisungen angewiesen.
-
-Unabhängig davon, welche Strategie Sie wählen, ist die empfohlene Vorgehensweise oft die gleiche, nämlich dass Sie mit der Fehlersuche in den
-die Anfangsebenen zuerst und die Endebenen zuletzt debuggen.
-
-Es wird empfohlen, dass Sie die Ausgaben der folgenden Ebenen abrufen, entweder durch Druckanweisungen oder Unterkomponentenfunktionen
-Schichten in der folgenden Reihenfolge abrufen:
-
-1. Rufen Sie die Eingabe-IDs ab, die an das Modell übergeben wurden
-2. Rufen Sie die Worteinbettungen ab
-3. Rufen Sie die Eingabe der ersten Transformer-Schicht ab
-4. Rufen Sie die Ausgabe der ersten Transformer-Schicht ab
-5. Rufen Sie die Ausgabe der folgenden n - 1 Transformer-Schichten ab
-6. Rufen Sie die Ausgabe des gesamten BrandNewBert Modells ab
-
-Die Eingabe-IDs sollten dabei aus einem Array von Ganzzahlen bestehen, *z.B.* `input_ids = [0, 4, 4, 3, 2, 4, 1, 7, 19]`
-
-Die Ausgaben der folgenden Schichten bestehen oft aus mehrdimensionalen Float-Arrays und können wie folgt aussehen:
-
-```
-[[
- [-0.1465, -0.6501,  0.1993,  ...,  0.1451,  0.3430,  0.6024],
- [-0.4417, -0.5920,  0.3450,  ..., -0.3062,  0.6182,  0.7132],
- [-0.5009, -0.7122,  0.4548,  ..., -0.3662,  0.6091,  0.7648],
- ...,
- [-0.5613, -0.6332,  0.4324,  ..., -0.3792,  0.7372,  0.9288],
- [-0.5416, -0.6345,  0.4180,  ..., -0.3564,  0.6992,  0.9191],
- [-0.5334, -0.6403,  0.4271,  ..., -0.3339,  0.6533,  0.8694]]],
-```
-
-Wir erwarten, dass jedes zu 🤗 Transformers hinzugefügte Modell eine Reihe von Integrationstests besteht, was bedeutet, dass das ursprüngliche
-Modell und die neu implementierte Version in 🤗 Transformers exakt dieselbe Ausgabe liefern müssen, und zwar mit einer Genauigkeit von 0,001!
-Da es normal ist, dass das exakt gleiche Modell, das in verschiedenen Bibliotheken geschrieben wurde, je nach Bibliotheksrahmen eine leicht unterschiedliche Ausgabe liefern kann
-eine leicht unterschiedliche Ausgabe liefern kann, akzeptieren wir eine Fehlertoleranz von 1e-3 (0,001). Es reicht nicht aus, wenn das Modell
-fast das gleiche Ergebnis liefert, sie müssen fast identisch sein. Daher werden Sie sicherlich die Zwischenergebnisse
-Zwischenergebnisse der 🤗 Transformers-Version mehrfach mit den Zwischenergebnissen der ursprünglichen Implementierung von
-*brand_new_bert* vergleichen. In diesem Fall ist eine **effiziente** Debugging-Umgebung des ursprünglichen Repositorys absolut
-wichtig ist. Hier sind einige Ratschläge, um Ihre Debugging-Umgebung so effizient wie möglich zu gestalten.
-
-- Finden Sie den besten Weg, um Zwischenergebnisse zu debuggen. Ist das ursprüngliche Repository in PyTorch geschrieben? Dann sollten Sie
-  dann sollten Sie sich wahrscheinlich die Zeit nehmen, ein längeres Skript zu schreiben, das das ursprüngliche Modell in kleinere Unterkomponenten zerlegt, um
-  Zwischenwerte abzurufen. Ist das ursprüngliche Repository in Tensorflow 1 geschrieben? Dann müssen Sie sich möglicherweise auf die
-  TensorFlow Druckoperationen wie [tf.print](https://www.tensorflow.org/api_docs/python/tf/print) verlassen, um die
-  Zwischenwerte auszugeben. Ist das ursprüngliche Repository in Jax geschrieben? Dann stellen Sie sicher, dass das Modell **nicht jitted** ist, wenn
-  wenn Sie den Vorwärtsdurchlauf ausführen, *z.B.* schauen Sie sich [dieser Link](https://github.com/google/jax/issues/196) an.
-- Verwenden Sie den kleinsten vortrainierten Prüfpunkt, den Sie finden können. Je kleiner der Prüfpunkt ist, desto schneller wird Ihr Debugging-Zyklus
-  wird. Es ist nicht effizient, wenn Ihr vorab trainiertes Modell so groß ist, dass Ihr Vorwärtsdurchlauf mehr als 10 Sekunden dauert.
-  Falls nur sehr große Checkpoints verfügbar sind, kann es sinnvoller sein, ein Dummy-Modell in der neuen
-  Umgebung mit zufällig initialisierten Gewichten zu erstellen und diese Gewichte zum Vergleich mit der 🤗 Transformers-Version
-  Ihres Modells
-- Vergewissern Sie sich, dass Sie den einfachsten Weg wählen, um einen Forward Pass im ursprünglichen Repository aufzurufen. Idealerweise sollten Sie
-  die Funktion im originalen Repository finden, die **nur** einen einzigen Vorwärtspass aufruft, *d.h.* die oft aufgerufen wird
-  Vorhersagen", "Auswerten", "Vorwärts" oder "Aufruf" genannt wird. Sie wollen keine Funktion debuggen, die `forward` aufruft
-  mehrfach aufruft, *z.B.* um Text zu erzeugen, wie `autoregressive_sample`, `generate`.
-- Versuchen Sie, die Tokenisierung vom *Forward*-Pass des Modells zu trennen. Wenn das Original-Repository Beispiele zeigt, bei denen
-  Sie eine Zeichenkette eingeben müssen, dann versuchen Sie herauszufinden, an welcher Stelle im Vorwärtsaufruf die Zeichenketteneingabe in Eingabe-IDs geändert wird
-  geändert wird und beginnen Sie an dieser Stelle. Das könnte bedeuten, dass Sie möglicherweise selbst ein kleines Skript schreiben oder den
-  Originalcode so ändern müssen, dass Sie die ids direkt eingeben können, anstatt eine Zeichenkette einzugeben.
-- Vergewissern Sie sich, dass sich das Modell in Ihrem Debugging-Setup **nicht** im Trainingsmodus befindet, der oft dazu führt, dass das Modell
-  Dies führt häufig zu zufälligen Ergebnissen, da das Modell mehrere Dropout-Schichten enthält. Stellen Sie sicher, dass der Vorwärtsdurchlauf in Ihrer Debugging
-  Umgebung **deterministisch** ist, damit die Dropout-Schichten nicht verwendet werden. Oder verwenden Sie *transformers.utils.set_seed*.
-  wenn sich die alte und die neue Implementierung im selben Framework befinden.
-
-Im folgenden Abschnitt finden Sie genauere Details/Tipps, wie Sie dies für *brand_new_bert* tun können.
-
-### 5.-14. Portierung von BrandNewBert auf 🤗 Transformatoren
-
-Als nächstes können Sie endlich damit beginnen, neuen Code zu 🤗 Transformers hinzuzufügen. Gehen Sie in den Klon Ihres 🤗 Transformers Forks:
-
-```bash
-cd transformers
-```
-
-In dem speziellen Fall, dass Sie ein Modell hinzufügen, dessen Architektur genau mit der Modellarchitektur eines
-Modells übereinstimmt, müssen Sie nur ein Konvertierungsskript hinzufügen, wie in [diesem Abschnitt](#write-a-conversion-script) beschrieben.
-In diesem Fall können Sie einfach die gesamte Modellarchitektur des bereits vorhandenen Modells wiederverwenden.
-
-Andernfalls beginnen wir mit der Erstellung eines neuen Modells. Wir empfehlen die Verwendung des folgenden Skripts, um ein Modell hinzuzufügen
-ein bestehendes Modell:
-
-```bash
-transformers-cli add-new-model-like
-```
-
-Sie werden mit einem Fragebogen aufgefordert, die grundlegenden Informationen Ihres Modells einzugeben.
-
-**Eröffnen Sie einen Pull Request auf dem Haupt-Repositorium huggingface/transformers**
-
-Bevor Sie mit der Anpassung des automatisch generierten Codes beginnen, ist es nun an der Zeit, einen "Work in progress (WIP)" Pull
-Anfrage, *z.B.* "[WIP] Add *brand_new_bert*", in 🤗 Transformers zu öffnen, damit Sie und das Hugging Face Team
-Seite an Seite an der Integration des Modells in 🤗 Transformers arbeiten können.
-
-Sie sollten Folgendes tun:
-
-1. Erstellen Sie eine Verzweigung mit einem beschreibenden Namen von Ihrer Hauptverzweigung
-
-```bash
-git checkout -b add_brand_new_bert
-```
-
-2. Bestätigen Sie den automatisch generierten Code:
-
-```bash
-git add .
-git commit
-```
-
-3. Abrufen und zurücksetzen auf die aktuelle Haupt
-
-```bash
-git fetch upstream
-git rebase upstream/main
-```
-
-4. Übertragen Sie die Änderungen auf Ihr Konto mit:
-
-```bash
-git push -u origin a-descriptive-name-for-my-changes
-```
-
-5. Wenn Sie zufrieden sind, gehen Sie auf die Webseite Ihrer Abspaltung auf GitHub. Klicken Sie auf "Pull request". Stellen Sie sicher, dass Sie das
-   GitHub-Handle einiger Mitglieder des Hugging Face-Teams als Reviewer hinzuzufügen, damit das Hugging Face-Team über zukünftige Änderungen informiert wird.
-   zukünftige Änderungen benachrichtigt wird.
-
-6. Ändern Sie den PR in einen Entwurf, indem Sie auf der rechten Seite der GitHub-Pull-Request-Webseite auf "In Entwurf umwandeln" klicken.
-
-Vergessen Sie im Folgenden nicht, wenn Sie Fortschritte gemacht haben, Ihre Arbeit zu committen und in Ihr Konto zu pushen, damit sie in der Pull-Anfrage erscheint.
-damit sie in der Pull-Anfrage angezeigt wird. Außerdem sollten Sie darauf achten, dass Sie Ihre Arbeit von Zeit zu Zeit mit dem aktuellen main
-von Zeit zu Zeit zu aktualisieren, indem Sie dies tun:
-
-```bash
-git fetch upstream
-git merge upstream/main
-```
-
-Generell sollten Sie alle Fragen, die Sie in Bezug auf das Modell oder Ihre Implementierung haben, in Ihrem PR stellen und
-in der PR diskutiert/gelöst werden. Auf diese Weise wird das Hugging Face Team immer benachrichtigt, wenn Sie neuen Code einreichen oder
-wenn Sie eine Frage haben. Es ist oft sehr hilfreich, das Hugging Face-Team auf Ihren hinzugefügten Code hinzuweisen, damit das Hugging Face-Team Ihr Problem oder Ihre Frage besser verstehen kann.
-Face-Team Ihr Problem oder Ihre Frage besser verstehen kann.
-
-Gehen Sie dazu auf die Registerkarte "Geänderte Dateien", auf der Sie alle Ihre Änderungen sehen, gehen Sie zu einer Zeile, zu der Sie eine Frage stellen möchten
-eine Frage stellen möchten, und klicken Sie auf das "+"-Symbol, um einen Kommentar hinzuzufügen. Wenn eine Frage oder ein Problem gelöst wurde,
-können Sie auf die Schaltfläche "Lösen" des erstellten Kommentars klicken.
-
-Auf dieselbe Weise wird das Hugging Face-Team Kommentare öffnen, wenn es Ihren Code überprüft. Wir empfehlen, die meisten Fragen
-auf GitHub in Ihrem PR zu stellen. Für einige sehr allgemeine Fragen, die für die Öffentlichkeit nicht sehr nützlich sind, können Sie das
-Hugging Face Team per Slack oder E-Mail zu stellen.
-
-**5. Passen Sie den Code der generierten Modelle für brand_new_bert** an.
-
-Zunächst werden wir uns nur auf das Modell selbst konzentrieren und uns nicht um den Tokenizer kümmern. Den gesamten relevanten Code sollten Sie
-finden Sie in den generierten Dateien `src/transformers/models/brand_new_bert/modeling_brand_new_bert.py` und
-`src/transformers/models/brand_new_bert/configuration_brand_new_bert.py`.
-
-Jetzt können Sie endlich mit dem Programmieren beginnen :). Der generierte Code in
-`src/transformers/models/brand_new_bert/modeling_brand_new_bert.py` wird entweder die gleiche Architektur wie BERT haben, wenn
-wenn es sich um ein reines Encoder-Modell handelt oder BART, wenn es sich um ein Encoder-Decoder-Modell handelt. An diesem Punkt sollten Sie sich daran erinnern, was
-was Sie am Anfang über die theoretischen Aspekte des Modells gelernt haben: *Wie unterscheidet sich das Modell von BERT oder
-BART?*". Implementieren Sie diese Änderungen, was oft bedeutet, dass Sie die *Selbstaufmerksamkeitsschicht*, die Reihenfolge der Normalisierungsschicht usw. ändern müssen.
-Schicht usw... Auch hier ist es oft nützlich, sich die ähnliche Architektur bereits bestehender Modelle in Transformers anzusehen, um ein besseres Gefühl dafür zu bekommen
-ein besseres Gefühl dafür zu bekommen, wie Ihr Modell implementiert werden sollte.
-
-**Beachten Sie**, dass Sie an diesem Punkt nicht sehr sicher sein müssen, dass Ihr Code völlig korrekt oder sauber ist. Vielmehr ist es
-Sie sollten vielmehr eine erste *unbereinigte*, kopierte Version des ursprünglichen Codes in
-src/transformers/models/brand_new_bert/modeling_brand_new_bert.py" hinzuzufügen, bis Sie das Gefühl haben, dass der gesamte notwendige Code
-hinzugefügt wurde. Unserer Erfahrung nach ist es viel effizienter, schnell eine erste Version des erforderlichen Codes hinzuzufügen und
-den Code iterativ mit dem Konvertierungsskript zu verbessern/korrigieren, wie im nächsten Abschnitt beschrieben. Das einzige, was
-zu diesem Zeitpunkt funktionieren muss, ist, dass Sie die 🤗 Transformers-Implementierung von *brand_new_bert* instanziieren können, *d.h.* der
-folgende Befehl sollte funktionieren:
-
-```python
-from transformers import BrandNewBertModel, BrandNewBertConfig
-
-model = BrandNewBertModel(BrandNewBertConfig())
-```
-
-Der obige Befehl erstellt ein Modell gemäß den Standardparametern, die in `BrandNewBertConfig()` definiert sind, mit
-zufälligen Gewichten und stellt damit sicher, dass die `init()` Methoden aller Komponenten funktionieren.
-
-Beachten Sie, dass alle zufälligen Initialisierungen in der Methode `_init_weights` Ihres `BrandnewBertPreTrainedModel` stattfinden sollten.
-Klasse erfolgen sollte. Sie sollte alle Blattmodule in Abhängigkeit von den Variablen der Konfiguration initialisieren. Hier ist ein Beispiel mit der
-BERT `_init_weights` Methode:
-
-```py
-def _init_weights(self, module):
-    """Initialize the weights"""
-    if isinstance(module, nn.Linear):
-        module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
-        if module.bias is not None:
-            module.bias.data.zero_()
-    elif isinstance(module, nn.Embedding):
-        module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
-        if module.padding_idx is not None:
-            module.weight.data[module.padding_idx].zero_()
-    elif isinstance(module, nn.LayerNorm):
-        module.bias.data.zero_()
-        module.weight.data.fill_(1.0)
-```
-
-Sie können weitere benutzerdefinierte Schemata verwenden, wenn Sie eine spezielle Initialisierung für einige Module benötigen. Zum Beispiel in
-`Wav2Vec2ForPreTraining` müssen die letzten beiden linearen Schichten die Initialisierung des regulären PyTorch `nn.Linear` haben.
-aber alle anderen sollten eine Initialisierung wie oben verwenden. Dies ist wie folgt kodiert:
-
-```py
-def _init_weights(self, module):
-    """Initialize the weights"""
-    if isinstance(module, Wav2Vec2ForPreTraining):
-        module.project_hid.reset_parameters()
-        module.project_q.reset_parameters()
-        module.project_hid._is_hf_initialized = True
-        module.project_q._is_hf_initialized = True
-    elif isinstance(module, nn.Linear):
-        module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
-        if module.bias is not None:
-            module.bias.data.zero_()
-```
-
-Das Flag `_is_hf_initialized` wird intern verwendet, um sicherzustellen, dass wir ein Submodul nur einmal initialisieren. Wenn Sie es auf
-`True` für `module.project_q` und `module.project_hid` setzen, stellen wir sicher, dass die benutzerdefinierte Initialisierung, die wir vorgenommen haben, später nicht überschrieben wird,
-die Funktion `_init_weights` nicht auf sie angewendet wird.
-
-**6. Schreiben Sie ein Konvertierungsskript**
-
-Als nächstes sollten Sie ein Konvertierungsskript schreiben, mit dem Sie den Checkpoint, den Sie zum Debuggen von *brand_new_bert* im
-im ursprünglichen Repository in einen Prüfpunkt konvertieren, der mit Ihrer gerade erstellten 🤗 Transformers-Implementierung von
-*brand_new_bert*. Es ist nicht ratsam, das Konvertierungsskript von Grund auf neu zu schreiben, sondern die bereits
-bestehenden Konvertierungsskripten in 🤗 Transformers nach einem Skript zu suchen, das für die Konvertierung eines ähnlichen Modells verwendet wurde, das im
-demselben Framework wie *brand_new_bert* geschrieben wurde. Normalerweise reicht es aus, ein bereits vorhandenes Konvertierungsskript zu kopieren und
-es für Ihren Anwendungsfall leicht anzupassen. Zögern Sie nicht, das Hugging Face Team zu bitten, Sie auf ein ähnliches, bereits vorhandenes
-Konvertierungsskript für Ihr Modell zu finden.
-
-- Wenn Sie ein Modell von TensorFlow nach PyTorch portieren, ist ein guter Ausgangspunkt das Konvertierungsskript von BERT [hier](https://github.com/huggingface/transformers/blob/7acfa95afb8194f8f9c1f4d2c6028224dbed35a2/src/transformers/models/bert/modeling_bert.py#L91)
-- Wenn Sie ein Modell von PyTorch nach PyTorch portieren, ist ein guter Ausgangspunkt das Konvertierungsskript von BART [hier](https://github.com/huggingface/transformers/blob/main/src/transformers/models/bart/convert_bart_original_pytorch_checkpoint_to_pytorch.py)
-
-Im Folgenden werden wir kurz erklären, wie PyTorch-Modelle Ebenengewichte speichern und Ebenennamen definieren. In PyTorch wird der
-Name einer Ebene durch den Namen des Klassenattributs definiert, das Sie der Ebene geben. Lassen Sie uns ein Dummy-Modell in
-PyTorch, das wir `SimpleModel` nennen, wie folgt:
-
-```python
-from torch import nn
-
-
-class SimpleModel(nn.Module):
-    def __init__(self):
-        super().__init__()
-        self.dense = nn.Linear(10, 10)
-        self.intermediate = nn.Linear(10, 10)
-        self.layer_norm = nn.LayerNorm(10)
-```
-
-Jetzt können wir eine Instanz dieser Modelldefinition erstellen, die alle Gewichte ausfüllt: `dense`, `intermediate`,
-`layer_norm` mit zufälligen Gewichten. Wir können das Modell ausdrucken, um seine Architektur zu sehen
-
-```python
-model = SimpleModel()
-
-print(model)
-```
-
-Dies gibt folgendes aus:
-
-```
-SimpleModel(
-  (dense): Linear(in_features=10, out_features=10, bias=True)
-  (intermediate): Linear(in_features=10, out_features=10, bias=True)
-  (layer_norm): LayerNorm((10,), eps=1e-05, elementwise_affine=True)
-)
-```
-
-Wir können sehen, dass die Ebenennamen durch den Namen des Klassenattributs in PyTorch definiert sind. Sie können die Gewichtswerte
-Werte einer bestimmten Ebene anzeigen lassen:
-
-```python
-print(model.dense.weight.data)
-```
-
-um zu sehen, dass die Gewichte zufällig initialisiert wurden
-
-```
-tensor([[-0.0818,  0.2207, -0.0749, -0.0030,  0.0045, -0.1569, -0.1598,  0.0212,
-         -0.2077,  0.2157],
-        [ 0.1044,  0.0201,  0.0990,  0.2482,  0.3116,  0.2509,  0.2866, -0.2190,
-          0.2166, -0.0212],
-        [-0.2000,  0.1107, -0.1999, -0.3119,  0.1559,  0.0993,  0.1776, -0.1950,
-         -0.1023, -0.0447],
-        [-0.0888, -0.1092,  0.2281,  0.0336,  0.1817, -0.0115,  0.2096,  0.1415,
-         -0.1876, -0.2467],
-        [ 0.2208, -0.2352, -0.1426, -0.2636, -0.2889, -0.2061, -0.2849, -0.0465,
-          0.2577,  0.0402],
-        [ 0.1502,  0.2465,  0.2566,  0.0693,  0.2352, -0.0530,  0.1859, -0.0604,
-          0.2132,  0.1680],
-        [ 0.1733, -0.2407, -0.1721,  0.1484,  0.0358, -0.0633, -0.0721, -0.0090,
-          0.2707, -0.2509],
-        [-0.1173,  0.1561,  0.2945,  0.0595, -0.1996,  0.2988, -0.0802,  0.0407,
-          0.1829, -0.1568],
-        [-0.1164, -0.2228, -0.0403,  0.0428,  0.1339,  0.0047,  0.1967,  0.2923,
-          0.0333, -0.0536],
-        [-0.1492, -0.1616,  0.1057,  0.1950, -0.2807, -0.2710, -0.1586,  0.0739,
-          0.2220,  0.2358]]).
-```
-
-Im Konvertierungsskript sollten Sie diese zufällig initialisierten Gewichte mit den genauen Gewichten der
-entsprechenden Ebene im Kontrollpunkt. *Z.B.*
-
-```python
-# retrieve matching layer weights, e.g. by
-# recursive algorithm
-layer_name = "dense"
-pretrained_weight = array_of_dense_layer
-
-model_pointer = getattr(model, "dense")
-
-model_pointer.weight.data = torch.from_numpy(pretrained_weight)
-```
-
-Dabei müssen Sie sicherstellen, dass jedes zufällig initialisierte Gewicht Ihres PyTorch-Modells und sein entsprechendes
-Checkpoint-Gewicht in **Form und Name** genau übereinstimmen. Zu diesem Zweck ist es **notwendig**, assert
-Anweisungen für die Form hinzuzufügen und die Namen der Checkpoint-Gewichte auszugeben. Sie sollten z.B. Anweisungen hinzufügen wie:
-
-```python
-assert (
-    model_pointer.weight.shape == pretrained_weight.shape
-), f"Pointer shape of random weight {model_pointer.shape} and array shape of checkpoint weight {pretrained_weight.shape} mismatched"
-```
-
-Außerdem sollten Sie die Namen der beiden Gewichte ausdrucken, um sicherzustellen, dass sie übereinstimmen, *z.B.*.
-
-```python
-logger.info(f"Initialize PyTorch weight {layer_name} from {pretrained_weight.name}")
-```
-
-Wenn entweder die Form oder der Name nicht übereinstimmt, haben Sie wahrscheinlich das falsche Kontrollpunktgewicht einer zufällig
-Ebene der 🤗 Transformers-Implementierung zugewiesen.
-
-Eine falsche Form ist höchstwahrscheinlich auf eine falsche Einstellung der Konfigurationsparameter in `BrandNewBertConfig()` zurückzuführen, die
-nicht genau mit denen übereinstimmen, die für den zu konvertierenden Prüfpunkt verwendet wurden. Es könnte aber auch sein, dass
-die PyTorch-Implementierung eines Layers erfordert, dass das Gewicht vorher transponiert wird.
-
-Schließlich sollten Sie auch überprüfen, ob **alle** erforderlichen Gewichte initialisiert sind und alle Checkpoint-Gewichte ausgeben, die
-die nicht zur Initialisierung verwendet wurden, um sicherzustellen, dass das Modell korrekt konvertiert wurde. Es ist völlig normal, dass die
-Konvertierungsversuche entweder mit einer falschen Shape-Anweisung oder einer falschen Namenszuweisung fehlschlagen. Das liegt höchstwahrscheinlich daran, dass entweder
-Sie haben falsche Parameter in `BrandNewBertConfig()` verwendet, haben eine falsche Architektur in der 🤗 Transformers
-Implementierung, Sie haben einen Fehler in den `init()` Funktionen einer der Komponenten der 🤗 Transformers
-Implementierung oder Sie müssen eine der Kontrollpunktgewichte transponieren.
-
-Dieser Schritt sollte mit dem vorherigen Schritt wiederholt werden, bis alle Gewichte des Kontrollpunkts korrekt in das
-Transformers-Modell geladen sind. Nachdem Sie den Prüfpunkt korrekt in die 🤗 Transformers-Implementierung geladen haben, können Sie das Modell
-das Modell unter einem Ordner Ihrer Wahl `/path/to/converted/checkpoint/folder` speichern, der dann sowohl ein
-Datei `pytorch_model.bin` und eine Datei `config.json` enthalten sollte:
-
-```python
-model.save_pretrained("/path/to/converted/checkpoint/folder")
-```
-
-**7. Implementieren Sie den Vorwärtspass**
-
-Nachdem es Ihnen gelungen ist, die trainierten Gewichte korrekt in die 🤗 Transformers-Implementierung zu laden, sollten Sie nun dafür sorgen
-sicherstellen, dass der Forward Pass korrekt implementiert ist. In [Machen Sie sich mit dem ursprünglichen Repository vertraut](#3-4-führen-sie-einen-pre-training-checkpoint-mit-dem-original-repository-durch) haben Sie bereits ein Skript erstellt, das einen Forward Pass
-Durchlauf des Modells unter Verwendung des Original-Repositorys durchführt. Jetzt sollten Sie ein analoges Skript schreiben, das die 🤗 Transformers
-Implementierung anstelle der Originalimplementierung verwenden. Es sollte wie folgt aussehen:
-
-```python
-model = BrandNewBertModel.from_pretrained("/path/to/converted/checkpoint/folder")
-input_ids = [0, 4, 4, 3, 2, 4, 1, 7, 19]
-output = model(input_ids).last_hidden_states
-```
-
-Es ist sehr wahrscheinlich, dass die 🤗 Transformers-Implementierung und die ursprüngliche Modell-Implementierung nicht genau die gleiche Ausgabe liefern.
-beim ersten Mal nicht die gleiche Ausgabe liefern oder dass der Vorwärtsdurchlauf einen Fehler auslöst. Seien Sie nicht enttäuscht - das ist zu erwarten! Erstens,
-sollten Sie sicherstellen, dass der Vorwärtsdurchlauf keine Fehler auslöst. Es passiert oft, dass die falschen Dimensionen verwendet werden
-verwendet werden, was zu einem *Dimensionality mismatch* Fehler führt oder dass der falsche Datentyp verwendet wird, *z.B.* `torch.long`
-anstelle von `torch.float32`. Zögern Sie nicht, das Hugging Face Team um Hilfe zu bitten, wenn Sie bestimmte Fehler nicht lösen können.
-bestimmte Fehler nicht lösen können.
-
-Um sicherzustellen, dass die Implementierung von 🤗 Transformers korrekt funktioniert, müssen Sie sicherstellen, dass die Ausgaben
-einer Genauigkeit von `1e-3` entsprechen. Zunächst sollten Sie sicherstellen, dass die Ausgabeformen identisch sind, *d.h.*.
-Die Ausgabeform *outputs.shape* sollte für das Skript der 🤗 Transformers-Implementierung und die ursprüngliche
-Implementierung ergeben. Als nächstes sollten Sie sicherstellen, dass auch die Ausgabewerte identisch sind. Dies ist einer der schwierigsten
-Teile des Hinzufügens eines neuen Modells. Häufige Fehler, warum die Ausgaben nicht identisch sind, sind:
-
-- Einige Ebenen wurden nicht hinzugefügt, *d.h.* eine *Aktivierungsebene* wurde nicht hinzugefügt, oder die Restverbindung wurde vergessen
-- Die Worteinbettungsmatrix wurde nicht gebunden
-- Es werden die falschen Positionseinbettungen verwendet, da die ursprüngliche Implementierung einen Offset verwendet
-- Dropout wird während des Vorwärtsdurchlaufs angewendet. Um dies zu beheben, stellen Sie sicher, dass *model.training auf False* steht und dass keine Dropout
-  Schicht während des Vorwärtsdurchlaufs fälschlicherweise aktiviert wird, *d.h.* übergeben Sie *self.training* an [PyTorch's functional dropout](https://pytorch.org/docs/stable/nn.functional.html?highlight=dropout#torch.nn.functional.dropout)
-
-Der beste Weg, das Problem zu beheben, besteht normalerweise darin, sich den Vorwärtsdurchlauf der ursprünglichen Implementierung und die 🤗
-Transformers-Implementierung nebeneinander zu sehen und zu prüfen, ob es Unterschiede gibt. Idealerweise sollten Sie die
-Zwischenergebnisse beider Implementierungen des Vorwärtsdurchlaufs debuggen/ausdrucken, um die genaue Position im Netzwerk zu finden, an der die 🤗
-Transformers-Implementierung eine andere Ausgabe zeigt als die ursprüngliche Implementierung. Stellen Sie zunächst sicher, dass die
-hartcodierten `input_ids` in beiden Skripten identisch sind. Überprüfen Sie dann, ob die Ausgaben der ersten Transformation von
-der `input_ids` (normalerweise die Worteinbettungen) identisch sind. Und dann arbeiten Sie sich bis zur allerletzten Schicht des
-Netzwerks. Irgendwann werden Sie einen Unterschied zwischen den beiden Implementierungen feststellen, der Sie auf den Fehler
-in der Implementierung von 🤗 Transformers hinweist. Unserer Erfahrung nach ist ein einfacher und effizienter Weg, viele Druckanweisungen hinzuzufügen
-sowohl in der Original-Implementierung als auch in der 🤗 Transformers-Implementierung an den gleichen Stellen im Netzwerk
-hinzuzufügen und nacheinander Druckanweisungen zu entfernen, die dieselben Werte für Zwischenpräsentationen anzeigen.
-
-Wenn Sie sicher sind, dass beide Implementierungen die gleiche Ausgabe liefern, überprüfen Sie die Ausgaben mit
-`torch.allclose(original_output, output, atol=1e-3)` überprüfen, haben Sie den schwierigsten Teil hinter sich! Herzlichen Glückwunsch - die
-Arbeit, die noch zu erledigen ist, sollte ein Kinderspiel sein 😊.
-
-**8. Hinzufügen aller notwendigen Modelltests**
-
-An diesem Punkt haben Sie erfolgreich ein neues Modell hinzugefügt. Es ist jedoch sehr gut möglich, dass das Modell noch nicht
-noch nicht vollständig mit dem erforderlichen Design übereinstimmt. Um sicherzustellen, dass die Implementierung vollständig kompatibel mit 🤗 Transformers ist, sollten alle
-gemeinsamen Tests bestehen. Der Cookiecutter sollte automatisch eine Testdatei für Ihr Modell hinzugefügt haben, wahrscheinlich unter
-demselben `tests/models/brand_new_bert/test_modeling_brand_new_bert.py`. Führen Sie diese Testdatei aus, um zu überprüfen, ob alle gängigen
-Tests bestehen:
-
-```bash
-pytest tests/models/brand_new_bert/test_modeling_brand_new_bert.py
-```
-
-Nachdem Sie alle allgemeinen Tests festgelegt haben, müssen Sie nun sicherstellen, dass all die schöne Arbeit, die Sie geleistet haben, gut getestet ist, damit
-
-- a) die Community Ihre Arbeit leicht nachvollziehen kann, indem sie sich spezifische Tests von *brand_new_bert* ansieht
-- b) zukünftige Änderungen an Ihrem Modell keine wichtigen Funktionen des Modells zerstören.
-
-Als erstes sollten Sie Integrationstests hinzufügen. Diese Integrationstests tun im Wesentlichen dasselbe wie die Debugging-Skripte
-die Sie zuvor zur Implementierung des Modells in 🤗 Transformers verwendet haben. Eine Vorlage für diese Modelltests wurde bereits von dem
-Cookiecutter hinzugefügt, die `BrandNewBertModelIntegrationTests` heißt und nur noch von Ihnen ausgefüllt werden muss. Um sicherzustellen, dass diese
-Tests erfolgreich sind, führen Sie
-
-```bash
-RUN_SLOW=1 pytest -sv tests/models/brand_new_bert/test_modeling_brand_new_bert.py::BrandNewBertModelIntegrationTests
-```
-
-<Tip>
-
-Falls Sie Windows verwenden, sollten Sie `RUN_SLOW=1` durch `SET RUN_SLOW=1` ersetzen.
-
-</Tip>
-
-Zweitens sollten alle Funktionen, die speziell für *brand_new_bert* sind, zusätzlich in einem separaten Test getestet werden unter
-`BrandNewBertModelTester`/`BrandNewBertModelTest`. Dieser Teil wird oft vergessen, ist aber in zweierlei Hinsicht äußerst nützlich
-Weise:
-
-- Er hilft dabei, das Wissen, das Sie während der Modellerweiterung erworben haben, an die Community weiterzugeben, indem er zeigt, wie die
-  speziellen Funktionen von *brand_new_bert* funktionieren sollten.
-- Künftige Mitwirkende können Änderungen am Modell schnell testen, indem sie diese speziellen Tests ausführen.
-
-
-**9. Implementieren Sie den Tokenizer**
-
-Als nächstes sollten wir den Tokenizer von *brand_new_bert* hinzufügen. Normalerweise ist der Tokenizer äquivalent oder sehr ähnlich zu einem
-bereits vorhandenen Tokenizer von 🤗 Transformers.
-
-Es ist sehr wichtig, die ursprüngliche Tokenizer-Datei zu finden/extrahieren und es zu schaffen, diese Datei in die 🤗
-Transformers Implementierung des Tokenizers zu laden.
-
-Um sicherzustellen, dass der Tokenizer korrekt funktioniert, empfiehlt es sich, zunächst ein Skript im ursprünglichen Repository zu erstellen
-zu erstellen, das eine Zeichenkette eingibt und die `input_ids` zurückgibt. Es könnte etwa so aussehen (in Pseudocode):
-
-```python
-input_str = "This is a long example input string containing special characters .$?-, numbers 2872 234 12 and words."
-model = BrandNewBertModel.load_pretrained_checkpoint("/path/to/checkpoint/")
-input_ids = model.tokenize(input_str)
-```
-
-Möglicherweise müssen Sie noch einmal einen Blick in das ursprüngliche Repository werfen, um die richtige Tokenizer-Funktion zu finden, oder Sie müssen
-Sie müssen vielleicht sogar Änderungen an Ihrem Klon des Original-Repositorys vornehmen, um nur die `input_ids` auszugeben. Nach dem Schreiben
-ein funktionierendes Tokenisierungsskript geschrieben, das das ursprüngliche Repository verwendet, sollten Sie ein analoges Skript für 🤗 Transformers
-erstellt werden. Es sollte ähnlich wie dieses aussehen:
-
-```python
-from transformers import BrandNewBertTokenizer
-
-input_str = "This is a long example input string containing special characters .$?-, numbers 2872 234 12 and words."
-
-tokenizer = BrandNewBertTokenizer.from_pretrained("/path/to/tokenizer/folder/")
-
-input_ids = tokenizer(input_str).input_ids
-```
-
-Wenn beide `input_ids` die gleichen Werte ergeben, sollte als letzter Schritt auch eine Tokenizer-Testdatei hinzugefügt werden.
-
-Analog zu den Modellierungstestdateien von *brand_new_bert* sollten auch die Tokenisierungs-Testdateien von *brand_new_bert*
-eine Reihe von fest kodierten Integrationstests enthalten.
-
-**10. Führen Sie End-to-End-Integrationstests aus**
-
-Nachdem Sie den Tokenizer hinzugefügt haben, sollten Sie auch ein paar End-to-End-Integrationstests, die sowohl das Modell als auch den
-Tokenizer zu `tests/models/brand_new_bert/test_modeling_brand_new_bert.py` in 🤗 Transformers.
-Ein solcher Test sollte bei einem aussagekräftigen
-Text-zu-Text-Beispiel zeigen, dass die Implementierung von 🤗 Transformers wie erwartet funktioniert. Ein aussagekräftiges Text-zu-Text-Beispiel kann
-z.B. *ein Quell-zu-Ziel-Übersetzungspaar, ein Artikel-zu-Zusammenfassung-Paar, ein Frage-zu-Antwort-Paar, usw... Wenn keiner der
-der portierten Prüfpunkte in einer nachgelagerten Aufgabe feinabgestimmt wurde, genügt es, sich einfach auf die Modelltests zu verlassen. In einem
-letzten Schritt, um sicherzustellen, dass das Modell voll funktionsfähig ist, sollten Sie alle Tests auch auf der GPU durchführen. Es kann
-Es kann vorkommen, dass Sie vergessen haben, einige `.to(self.device)` Anweisungen zu internen Tensoren des Modells hinzuzufügen, was in einem solchen
-Test zu einem Fehler führen würde. Falls Sie keinen Zugang zu einem Grafikprozessor haben, kann das Hugging Face Team diese Tests für Sie durchführen.
-Tests für Sie übernehmen.
-
-**11. Docstring hinzufügen**
-
-Nun sind alle notwendigen Funktionen für *brand_new_bert* hinzugefügt - Sie sind fast fertig! Das Einzige, was Sie noch hinzufügen müssen, ist
-ein schöner Docstring und eine Doku-Seite. Der Cookiecutter sollte eine Vorlagendatei namens
-`docs/source/model_doc/brand_new_bert.md` hinzugefügt haben, die Sie ausfüllen sollten. Die Benutzer Ihres Modells werden in der Regel zuerst einen Blick auf
-diese Seite ansehen, bevor sie Ihr Modell verwenden. Daher muss die Dokumentation verständlich und prägnant sein. Es ist sehr nützlich für
-die Gemeinschaft, einige *Tipps* hinzuzufügen, um zu zeigen, wie das Modell verwendet werden sollte. Zögern Sie nicht, das Hugging Face-Team anzupingen
-bezüglich der Docstrings.
-
-Stellen Sie als nächstes sicher, dass der zu `src/transformers/models/brand_new_bert/modeling_brand_new_bert.py` hinzugefügte docstring
-korrekt ist und alle erforderlichen Eingaben und Ausgaben enthält. Wir haben eine ausführliche Anleitung zum Schreiben von Dokumentationen und unserem Docstring-Format [hier](writing-documentation). Es ist immer gut, sich daran zu erinnern, dass die Dokumentation
-mindestens so sorgfältig behandelt werden sollte wie der Code in 🤗 Transformers, denn die Dokumentation ist in der Regel der erste Kontaktpunkt der
-Berührungspunkt der Community mit dem Modell ist.
-
-**Code refactor**
-
-Großartig, jetzt haben Sie den gesamten erforderlichen Code für *brand_new_bert* hinzugefügt. An diesem Punkt sollten Sie einige mögliche
-falschen Codestil korrigieren, indem Sie ausführen:
-
-```bash
-make style
-```
-
-und überprüfen Sie, ob Ihr Kodierungsstil die Qualitätsprüfung besteht:
-
-```bash
-make quality
-```
-
-Es gibt noch ein paar andere sehr strenge Designtests in 🤗 Transformers, die möglicherweise noch fehlschlagen, was sich in den
-den Tests Ihres Pull Requests. Dies liegt oft an fehlenden Informationen im Docstring oder an einer falschen
-Benennung. Das Hugging Face Team wird Ihnen sicherlich helfen, wenn Sie hier nicht weiterkommen.
-
-Und schließlich ist es immer eine gute Idee, den eigenen Code zu refaktorisieren, nachdem man sichergestellt hat, dass er korrekt funktioniert. Wenn alle
-Tests bestanden haben, ist es nun an der Zeit, den hinzugefügten Code noch einmal durchzugehen und einige Überarbeitungen vorzunehmen.
-
-Sie haben nun den Codierungsteil abgeschlossen, herzlichen Glückwunsch! 🎉 Sie sind großartig! 😎
-
-**12. Laden Sie die Modelle in den Model Hub hoch**
-
-In diesem letzten Teil sollten Sie alle Checkpoints konvertieren und in den Modell-Hub hochladen und eine Modellkarte für jeden
-hochgeladenen Modell-Kontrollpunkt. Sie können sich mit den Hub-Funktionen vertraut machen, indem Sie unsere [Model sharing and uploading Page](model_sharing) lesen. Hier sollten Sie mit dem Hugging Face-Team zusammenarbeiten, um einen passenden Namen für jeden
-Checkpoint festzulegen und die erforderlichen Zugriffsrechte zu erhalten, um das Modell unter der Organisation des Autors *brand_new_bert* hochladen zu können.
-*brand_new_bert*. Die Methode `push_to_hub`, die in allen Modellen in `transformers` vorhanden ist, ist ein schneller und effizienter Weg, Ihren Checkpoint in den Hub zu pushen. Ein kleines Snippet ist unten eingefügt:
-
-```python
-brand_new_bert.push_to_hub("brand_new_bert")
-# Uncomment the following line to push to an organization.
-# brand_new_bert.push_to_hub("<organization>/brand_new_bert")
-```
-
-Es lohnt sich, etwas Zeit darauf zu verwenden, für jeden Kontrollpunkt passende Musterkarten zu erstellen. Die Modellkarten sollten die
-spezifischen Merkmale dieses bestimmten Prüfpunkts hervorheben, * z.B.* auf welchem Datensatz wurde der Prüfpunkt
-vortrainiert/abgestimmt? Für welche nachgelagerte Aufgabe sollte das Modell verwendet werden? Und fügen Sie auch etwas Code bei, wie Sie
-wie das Modell korrekt verwendet wird.
-
-**13. (Optional) Notizbuch hinzufügen**
-
-Es ist sehr hilfreich, ein Notizbuch hinzuzufügen, in dem im Detail gezeigt wird, wie *brand_new_bert* für Schlussfolgerungen verwendet werden kann und/oder
-bei einer nachgelagerten Aufgabe feinabgestimmt wird. Dies ist nicht zwingend erforderlich, um Ihren PR zusammenzuführen, aber sehr nützlich für die Gemeinschaft.
-
-**14. Reichen Sie Ihren fertigen PR ein**
-
-Sie sind jetzt mit der Programmierung fertig und können zum letzten Schritt übergehen, nämlich der Zusammenführung Ihres PR mit main. Normalerweise hat das
-Hugging Face Team Ihnen an diesem Punkt bereits geholfen haben, aber es lohnt sich, sich etwas Zeit zu nehmen, um Ihrem fertigen
-PR eine schöne Beschreibung zu geben und eventuell Kommentare zu Ihrem Code hinzuzufügen, wenn Sie Ihren Gutachter auf bestimmte Designentscheidungen hinweisen wollen.
-Gutachter hinweisen wollen.
-
-### Teilen Sie Ihre Arbeit!!
-
-Jetzt ist es an der Zeit, von der Community Anerkennung für Ihre Arbeit zu bekommen! Die Fertigstellung einer Modellergänzung ist ein wichtiger
-Beitrag zu Transformers und der gesamten NLP-Gemeinschaft. Ihr Code und die portierten vortrainierten Modelle werden sicherlich
-von Hunderten und vielleicht sogar Tausenden von Entwicklern und Forschern genutzt werden. Sie sollten stolz auf Ihre Arbeit sein und Ihre
-Ihre Leistung mit der Gemeinschaft teilen.
-
-**Sie haben ein weiteres Modell erstellt, das für jeden in der Community super einfach zugänglich ist! 🤯**