unable to download models

[stevenh]

Describe the bug

When using AutoModelForSequenceClassification.from_pretrained to download a model I got the following.

OSError: distilbert-base-uncased does not appear to have a file named pytorch_model.bin but there is a file for TensorFlow weights. Use `from_tf=True` to load this model from those weights.
.venv/lib/python3.12/site-packages/transformers/modeling_utils.

This seems to indicate that I needed to specify from_tf=True but that was already set.

After much digging I found the issue was I had an older version of hf-xet, specifically 1.0.0.

This passes the is_package_available used by is_xet_available but then when xet_get is called it tries to load the dependency and fails, resulting in the above message.

Reproduction

Run the following with hf-xet v1.0.0 installed.

from transformers.models.auto.modeling_auto import AutoModelForSequenceClassification

AutoModelForSequenceClassification.from_pretrained("distilbert-base-uncased", from_tf=True)

Logs

.venv/lib/python3.12/site-packages/transformers/models/auto/auto_factory.py:571: in from_pretrained
    return model_class.from_pretrained(
.venv/lib/python3.12/site-packages/transformers/modeling_utils.py:279: in _wrapper
    return func(*args, **kwargs)
.venv/lib/python3.12/site-packages/transformers/modeling_utils.py:4260: in from_pretrained
    checkpoint_files, sharded_metadata = _get_resolved_checkpoint_files(
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
pretrained_model_name_or_path = 'distilbert-base-uncased', subfolder = ''
variant = None, gguf_file = None, from_tf = True, from_flax = False
use_safetensors = None, cache_dir = None, force_download = False, proxies = None
local_files_only = False, token = None
user_agent = {'file_type': 'model', 'framework': 'pytorch', 'from_auto_class': False}
revision = 'main', commit_hash = '12040accade4e8a0f71eabdb258fecc2e7e948be'
    def _get_resolved_checkpoint_files(
        pretrained_model_name_or_path: Optional[Union[str, os.PathLike]],
        subfolder: str,
        variant: Optional[str],
        gguf_file: Optional[str],
        from_tf: bool,
        from_flax: bool,
        use_safetensors: bool,
        cache_dir: str,
        force_download: bool,
        proxies: Optional[Dict[str, str]],
        local_files_only: bool,
        token: Optional[Union[str, bool]],
        user_agent: dict,
        revision: str,
        commit_hash: Optional[str],
    ) -> Tuple[Optional[List[str]], Optional[Dict]]:
        """Get all the checkpoint filenames based on `pretrained_model_name_or_path`, and optional metadata if the
        checkpoints are sharded.
        This function will download the data if necesary.
        """
        is_sharded = False
    
        if pretrained_model_name_or_path is not None and gguf_file is None:
            pretrained_model_name_or_path = str(pretrained_model_name_or_path)
            is_local = os.path.isdir(pretrained_model_name_or_path)
            if is_local:
                if from_tf and os.path.isfile(
                    os.path.join(pretrained_model_name_or_path, subfolder, TF_WEIGHTS_NAME + ".index")
                ):
                    # Load from a TF 1.0 checkpoint in priority if from_tf
                    archive_file = os.path.join(pretrained_model_name_or_path, subfolder, TF_WEIGHTS_NAME + ".index")
                elif from_tf and os.path.isfile(os.path.join(pretrained_model_name_or_path, subfolder, TF2_WEIGHTS_NAME)):
                    # Load from a TF 2.0 checkpoint in priority if from_tf
                    archive_file = os.path.join(pretrained_model_name_or_path, subfolder, TF2_WEIGHTS_NAME)
                elif from_flax and os.path.isfile(
                    os.path.join(pretrained_model_name_or_path, subfolder, FLAX_WEIGHTS_NAME)
                ):
                    # Load from a Flax checkpoint in priority if from_flax
                    archive_file = os.path.join(pretrained_model_name_or_path, subfolder, FLAX_WEIGHTS_NAME)
                elif use_safetensors is not False and os.path.isfile(
                    os.path.join(pretrained_model_name_or_path, subfolder, _add_variant(SAFE_WEIGHTS_NAME, variant))
                ):
                                cached_file_kwargs = {
                                    "cache_dir": cache_dir,
                                    "force_download": force_download,
                                    "local_files_only": local_files_only,
                                    "user_agent": user_agent,
                                    "subfolder": subfolder,
                                    "_raise_exceptions_for_gated_repo": False,
                                    "_raise_exceptions_for_missing_entries": False,
                                    "_commit_hash": commit_hash,
                                    **has_file_kwargs,
                                }
                                if not has_file(pretrained_model_name_or_path, safe_weights_name, **has_file_kwargs):
                                    Thread(
                                        target=auto_conversion,
                                        args=(pretrained_model_name_or_path,),
                                        kwargs={"ignore_errors_during_conversion": True, **cached_file_kwargs},
                                        name="Thread-auto_conversion",
                                    ).start()
                        else:
                            # Otherwise, no PyTorch file was found, maybe there is a TF or Flax model file.
                            # We try those to give a helpful error message.
                            has_file_kwargs = {
                                "revision": revision,
                                "proxies": proxies,
                                "token": token,
                                "cache_dir": cache_dir,
                                "local_files_only": local_files_only,
                            }
                            if has_file(pretrained_model_name_or_path, TF2_WEIGHTS_NAME, **has_file_kwargs):
>                               raise EnvironmentError(
                                    f"{pretrained_model_name_or_path} does not appear to have a file named"
                                    f" {_add_variant(WEIGHTS_NAME, variant)} but there is a file for TensorFlow weights."
                                    " Use `from_tf=True` to load this model from those weights."
E                                   OSError: distilbert-base-uncased does not appear to have a file named pytorch_model.bin but there is a file for TensorFlow weights. Use `from_tf=True` to load this model from those weights.
.venv/lib/python3.12/site-packages/transformers/modeling_utils.py:1080: OSError

System info

huggingface-cli env

Copy-and-paste the text below in your GitHub issue.

- huggingface_hub version: 0.31.2
- Platform: macOS-15.5-arm64-arm-64bit
- Python version: 3.12.9
- Running in iPython ?: No
- Running in notebook ?: No
- Running in Google Colab ?: No
- Running in Google Colab Enterprise ?: No
- Token path ?: /Users/steve/.cache/huggingface/token
- Has saved token ?: False
- Configured git credential helpers: osxkeychain
- FastAI: N/A
- Tensorflow: N/A
- Torch: 2.7.0
- Jinja2: 3.1.6
- Graphviz: N/A
- keras: N/A
- Pydot: N/A
- Pillow: 10.4.0
- hf_transfer: N/A
- gradio: N/A
- tensorboard: N/A
- numpy: 2.2.5
- pydantic: 2.11.4
- aiohttp: 3.11.18
- hf_xet: 1.0.0
- ENDPOINT: https://huggingface.co
- HF_HUB_CACHE: /Users/steve/.cache/huggingface/hub
- HF_ASSETS_CACHE: /Users/steve/.cache/huggingface/assets
- HF_TOKEN_PATH: /Users/steve/.cache/huggingface/token
- HF_STORED_TOKENS_PATH: /Users/steve/.cache/huggingface/stored_tokens
- HF_HUB_OFFLINE: False
- HF_HUB_DISABLE_TELEMETRY: False
- HF_HUB_DISABLE_PROGRESS_BARS: None
- HF_HUB_DISABLE_SYMLINKS_WARNING: False
- HF_HUB_DISABLE_EXPERIMENTAL_WARNING: False
- HF_HUB_DISABLE_IMPLICIT_TOKEN: False
- HF_HUB_ENABLE_HF_TRANSFER: False
- HF_HUB_ETAG_TIMEOUT: 10
- HF_HUB_DOWNLOAD_TIMEOUT: 10