OSError due to huggingface-hub FutureWarning about resume_download

[albertvillanova]

The hugginface-hub FutureWarning about resume_download makes transformers raise an OSError.

See: https://github.com/huggingface/datasets/actions/runs/8973676702/job/25330013267

____________________ TokenizersHashTest.test_hash_tokenizer ____________________
[gw1] linux -- Python 3.8.18 /opt/hostedtoolcache/Python/3.8.18/x64/bin/python

cls = <class 'transformers.configuration_utils.PretrainedConfig'>
pretrained_model_name_or_path = 'bert-base-uncased'
kwargs = {'name_or_path': 'bert-base-uncased'}, cache_dir = None
force_download = False, resume_download = False, proxies = None, token = None
local_files_only = False, revision = None

    @classmethod
    def _get_config_dict(
        cls, pretrained_model_name_or_path: Union[str, os.PathLike], **kwargs
    ) -> Tuple[Dict[str, Any], Dict[str, Any]]:
        cache_dir = kwargs.pop("cache_dir", None)
        force_download = kwargs.pop("force_download", False)
        resume_download = kwargs.pop("resume_download", False)
        proxies = kwargs.pop("proxies", None)
        token = kwargs.pop("token", None)
        local_files_only = kwargs.pop("local_files_only", False)
        revision = kwargs.pop("revision", None)
        trust_remote_code = kwargs.pop("trust_remote_code", None)
        subfolder = kwargs.pop("subfolder", "")
        from_pipeline = kwargs.pop("_from_pipeline", None)
        from_auto_class = kwargs.pop("_from_auto", False)
        commit_hash = kwargs.pop("_commit_hash", None)
    
        gguf_file = kwargs.get("gguf_file", None)
    
        if trust_remote_code is True:
            logger.warning(
                "The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is"
                " ignored."
            )
    
        user_agent = {"file_type": "config", "from_auto_class": from_auto_class}
        if from_pipeline is not None:
            user_agent["using_pipeline"] = from_pipeline
    
        pretrained_model_name_or_path = str(pretrained_model_name_or_path)
    
        is_local = os.path.isdir(pretrained_model_name_or_path)
        if os.path.isfile(os.path.join(subfolder, pretrained_model_name_or_path)):
            # Special case when pretrained_model_name_or_path is a local file
            resolved_config_file = pretrained_model_name_or_path
            is_local = True
        elif is_remote_url(pretrained_model_name_or_path):
            configuration_file = pretrained_model_name_or_path if gguf_file is None else gguf_file
            resolved_config_file = download_url(pretrained_model_name_or_path)
        else:
            configuration_file = kwargs.pop("_configuration_file", CONFIG_NAME) if gguf_file is None else gguf_file
    
            try:
                # Load from local folder or from cache or download from model Hub and cache
>               resolved_config_file = cached_file(
                    pretrained_model_name_or_path,
                    configuration_file,
                    cache_dir=cache_dir,
                    force_download=force_download,
                    proxies=proxies,
                    resume_download=resume_download,
                    local_files_only=local_files_only,
                    token=token,
                    user_agent=user_agent,
                    revision=revision,
                    subfolder=subfolder,
                    _commit_hash=commit_hash,
                )

/opt/hostedtoolcache/Python/3.8.18/x64/lib/python3.8/site-packages/transformers/configuration_utils.py:689: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
/opt/hostedtoolcache/Python/3.8.18/x64/lib/python3.8/site-packages/transformers/utils/hub.py:399: in cached_file
    resolved_file = hf_hub_download(
/opt/hostedtoolcache/Python/3.8.18/x64/lib/python3.8/site-packages/huggingface_hub/utils/_validators.py:114: in _inner_fn
    return fn(*args, **kwargs)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

repo_id = 'bert-base-uncased', filename = 'config.json'

    @validate_hf_hub_args
    def hf_hub_download(
        repo_id: str,
        filename: str,
        *,
        subfolder: Optional[str] = None,
        repo_type: Optional[str] = None,
        revision: Optional[str] = None,
        library_name: Optional[str] = None,
        library_version: Optional[str] = None,
        cache_dir: Union[str, Path, None] = None,
        local_dir: Union[str, Path, None] = None,
        user_agent: Union[Dict, str, None] = None,
        force_download: bool = False,
        proxies: Optional[Dict] = None,
        etag_timeout: float = DEFAULT_ETAG_TIMEOUT,
        token: Union[bool, str, None] = None,
        local_files_only: bool = False,
        headers: Optional[Dict[str, str]] = None,
        endpoint: Optional[str] = None,
        # Deprecated args
        legacy_cache_layout: bool = False,
        resume_download: Optional[bool] = None,
        force_filename: Optional[str] = None,
        local_dir_use_symlinks: Union[bool, Literal["auto"]] = "auto",
    ) -> str:
        """Download a given file if it's not already present in the local cache.
        ...

        """
        if HF_HUB_ETAG_TIMEOUT != DEFAULT_ETAG_TIMEOUT:
            # Respect environment variable above user value
            etag_timeout = HF_HUB_ETAG_TIMEOUT
    
        if force_filename is not None:
            warnings.warn(
                "The `force_filename` parameter is deprecated as a new caching system, "
                "which keeps the filenames as they are on the Hub, is now in place.",
                FutureWarning,
            )
            legacy_cache_layout = True
        if resume_download is not None:
>           warnings.warn(
                "`resume_download` is deprecated and will be removed in version 1.0.0. "
                "Downloads always resume when possible. "
                "If you want to force a new download, use `force_download=True`.",
                FutureWarning,
            )
E           FutureWarning: `resume_download` is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use `force_download=True`.

/opt/hostedtoolcache/Python/3.8.18/x64/lib/python3.8/site-packages/huggingface_hub/file_download.py:1132: FutureWarning

During handling of the above exception, another exception occurred:

self = <tests.test_fingerprint.TokenizersHashTest testMethod=test_hash_tokenizer>

    @require_transformers
    @pytest.mark.integration
    def test_hash_tokenizer(self):
        from transformers import AutoTokenizer
    
        def encode(x):
            return tokenizer(x)
    
        # TODO: add hash consistency tests across sessions
>       tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")

tests/test_fingerprint.py:90: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
/opt/hostedtoolcache/Python/3.8.18/x64/lib/python3.8/site-packages/transformers/models/auto/tokenization_auto.py:837: in from_pretrained
    config = AutoConfig.from_pretrained(
/opt/hostedtoolcache/Python/3.8.18/x64/lib/python3.8/site-packages/transformers/models/auto/configuration_auto.py:934: in from_pretrained
    config_dict, unused_kwargs = PretrainedConfig.get_config_dict(pretrained_model_name_or_path, **kwargs)
/opt/hostedtoolcache/Python/3.8.18/x64/lib/python3.8/site-packages/transformers/configuration_utils.py:632: in get_config_dict
    config_dict, kwargs = cls._get_config_dict(pretrained_model_name_or_path, **kwargs)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

cls = <class 'transformers.configuration_utils.PretrainedConfig'>
pretrained_model_name_or_path = 'bert-base-uncased'
kwargs = {'name_or_path': 'bert-base-uncased'}, cache_dir = None
force_download = False, resume_download = False, proxies = None, token = None
local_files_only = False, revision = None

    @classmethod
    def _get_config_dict(
        cls, pretrained_model_name_or_path: Union[str, os.PathLike], **kwargs
    ) -> Tuple[Dict[str, Any], Dict[str, Any]]:
        cache_dir = kwargs.pop("cache_dir", None)
        force_download = kwargs.pop("force_download", False)
        resume_download = kwargs.pop("resume_download", False)
        proxies = kwargs.pop("proxies", None)
        token = kwargs.pop("token", None)
        local_files_only = kwargs.pop("local_files_only", False)
        revision = kwargs.pop("revision", None)
        trust_remote_code = kwargs.pop("trust_remote_code", None)
        subfolder = kwargs.pop("subfolder", "")
        from_pipeline = kwargs.pop("_from_pipeline", None)
        from_auto_class = kwargs.pop("_from_auto", False)
        commit_hash = kwargs.pop("_commit_hash", None)
    
        gguf_file = kwargs.get("gguf_file", None)
    
        if trust_remote_code is True:
            logger.warning(
                "The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is"
                " ignored."
            )
    
        user_agent = {"file_type": "config", "from_auto_class": from_auto_class}
        if from_pipeline is not None:
            user_agent["using_pipeline"] = from_pipeline
    
        pretrained_model_name_or_path = str(pretrained_model_name_or_path)
    
        is_local = os.path.isdir(pretrained_model_name_or_path)
        if os.path.isfile(os.path.join(subfolder, pretrained_model_name_or_path)):
            # Special case when pretrained_model_name_or_path is a local file
            resolved_config_file = pretrained_model_name_or_path
            is_local = True
        elif is_remote_url(pretrained_model_name_or_path):
            configuration_file = pretrained_model_name_or_path if gguf_file is None else gguf_file
            resolved_config_file = download_url(pretrained_model_name_or_path)
        else:
            configuration_file = kwargs.pop("_configuration_file", CONFIG_NAME) if gguf_file is None else gguf_file
    
            try:
                # Load from local folder or from cache or download from model Hub and cache
                resolved_config_file = cached_file(
                    pretrained_model_name_or_path,
                    configuration_file,
                    cache_dir=cache_dir,
                    force_download=force_download,
                    proxies=proxies,
                    resume_download=resume_download,
                    local_files_only=local_files_only,
                    token=token,
                    user_agent=user_agent,
                    revision=revision,
                    subfolder=subfolder,
                    _commit_hash=commit_hash,
                )
                commit_hash = extract_commit_hash(resolved_config_file, commit_hash)
            except EnvironmentError:
                # Raise any environment error raise by `cached_file`. It will have a helpful error message adapted to
                # the original exception.
                raise
            except Exception:
                # For any other exception, we throw a generic error.
>               raise EnvironmentError(
                    f"Can't load the configuration of '{pretrained_model_name_or_path}'. If you were trying to load it"
                    " from 'https://huggingface.co/models', make sure you don't have a local directory with the same"
                    f" name. Otherwise, make sure '{pretrained_model_name_or_path}' is the correct path to a directory"
                    f" containing a {configuration_file} file"
                )
E               OSError: Can't load the configuration of 'bert-base-uncased'. If you were trying to load it from 'https://huggingface.co/models', make sure you don't have a local directory with the same name. Otherwise, make sure 'bert-base-uncased' is the correct path to a directory containing a config.json file

/opt/hostedtoolcache/Python/3.8.18/x64/lib/python3.8/site-packages/transformers/configuration_utils.py:710: OSError

The error can be avoided by filtering the warning out. See: huggingface/datasets@889a48d#diff-6d4f609754165d567fd0edfc980788464067245d6220097b4c76a5120238dd2cR91-R93

with warnings.catch_warnings():
    warnings.simplefilter("ignore")
    tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")

Reproduction

To reproduce the bug locally, you must first remove the model from the "hub" cache:

rm -fr ~/.cache/huggingface/hub/models--bert-base-uncased

from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")

Follows:

• FutureWarning about resume_download is raised after huggingface-hub 0.23.0 release #30618

Maybe related to:

• Missing "config.json" when loading Llama-2-7b-chat-hf #31001
• Autotokenizer."from_pretrained" read wrong config file. not "tokenizer_config.json", but "config.json" #30977
• Failed to Download GPT2-large Model from Hub #30715

CC: @Wauplin

[Wauplin]

Opened #31007 to fix this. Looks like an overlook on my side, sorry about that.
Though the OSError is triggered by pytest because of CI config, not happening to all users.