huggingface-hub 0.26.5__py3-none-any.whl → 0.27.0__py3-none-any.whl

huggingface_hub/serialization/_torch.py

@@ -17,10 +17,12 @@ import importlib
 import json
 import os
 import re
-from collections import defaultdict
+from collections import defaultdict, namedtuple
 from functools import lru_cache
 from pathlib import Path
-from typing import TYPE_CHECKING, Any, Dict, List, Optional, Set, Tuple, Union
+from typing import TYPE_CHECKING, Any, Dict, Iterable, List, NamedTuple, Optional, Set, Tuple, Union
+
+from packaging import version
 
 from .. import constants, logging
 from ._base import MAX_SHARD_SIZE, StateDictSplit, split_state_dict_into_shards_factory
@@ -31,6 +33,8 @@ logger = logging.get_logger(__file__)
 if TYPE_CHECKING:
     import torch
 
+# SAVING
+
 
 def save_torch_model(
     model: "torch.nn.Module",
@@ -41,6 +45,7 @@ def save_torch_model(
     max_shard_size: Union[int, str] = MAX_SHARD_SIZE,
     metadata: Optional[Dict[str, str]] = None,
     safe_serialization: bool = True,
+    is_main_process: bool = True,
     shared_tensors_to_discard: Optional[List[str]] = None,
 ):
     """
@@ -95,6 +100,10 @@ def save_torch_model(
             Whether to save as safetensors, which is the default behavior. If `False`, the shards are saved as pickle.
             Safe serialization is recommended for security reasons. Saving as pickle is deprecated and will be removed
             in a future version.
+        is_main_process (`bool`, *optional*):
+            Whether the process calling this is the main process or not. Useful when in distributed training like
+            TPUs and need to call this function from all processes. In this case, set `is_main_process=True` only on
+            the main process to avoid race conditions. Defaults to True.
         shared_tensors_to_discard (`List[str]`, *optional*):
             List of tensor names to drop when saving shared tensors. If not provided and shared tensors are
             detected, it will drop the first name alphabetically.
@@ -122,6 +131,7 @@ def save_torch_model(
         metadata=metadata,
         safe_serialization=safe_serialization,
         save_directory=save_directory,
+        is_main_process=is_main_process,
         shared_tensors_to_discard=shared_tensors_to_discard,
     )
 
@@ -135,6 +145,7 @@ def save_torch_state_dict(
     max_shard_size: Union[int, str] = MAX_SHARD_SIZE,
     metadata: Optional[Dict[str, str]] = None,
     safe_serialization: bool = True,
+    is_main_process: bool = True,
     shared_tensors_to_discard: Optional[List[str]] = None,
 ) -> None:
     """
@@ -189,6 +200,10 @@ def save_torch_state_dict(
             Whether to save as safetensors, which is the default behavior. If `False`, the shards are saved as pickle.
             Safe serialization is recommended for security reasons. Saving as pickle is deprecated and will be removed
             in a future version.
+        is_main_process (`bool`, *optional*):
+            Whether the process calling this is the main process or not. Useful when in distributed training like
+            TPUs and need to call this function from all processes. In this case, set `is_main_process=True` only on
+            the main process to avoid race conditions. Defaults to True.
         shared_tensors_to_discard (`List[str]`, *optional*):
             List of tensor names to drop when saving shared tensors. If not provided and shared tensors are
             detected, it will drop the first name alphabetically.
@@ -243,15 +258,18 @@ def save_torch_state_dict(
         state_dict, filename_pattern=filename_pattern, max_shard_size=max_shard_size
     )
 
-    # Clean the folder from previous save
-    existing_files_regex = re.compile(filename_pattern.format(suffix=r"(-\d{5}-of-\d{5})?") + r"(\.index\.json)?")
-    for filename in os.listdir(save_directory):
-        if existing_files_regex.match(filename):
-            try:
-                logger.debug(f"Removing existing file '{filename}' from folder.")
-                os.remove(os.path.join(save_directory, filename))
-            except Exception as e:
-                logger.warning(f"Error when trying to remove existing '{filename}' from folder: {e}. Continuing...")
+    # Only main process should clean up existing files to avoid race conditions in distributed environment
+    if is_main_process:
+        existing_files_regex = re.compile(filename_pattern.format(suffix=r"(-\d{5}-of-\d{5})?") + r"(\.index\.json)?")
+        for filename in os.listdir(save_directory):
+            if existing_files_regex.match(filename):
+                try:
+                    logger.debug(f"Removing existing file '{filename}' from folder.")
+                    os.remove(os.path.join(save_directory, filename))
+                except Exception as e:
+                    logger.warning(
+                        f"Error when trying to remove existing '{filename}' from folder: {e}. Continuing..."
+                    )
 
     # Save each shard
     per_file_metadata = {"format": "pt"}
@@ -357,6 +375,331 @@ def split_torch_state_dict_into_shards(
     )
 
 
+# LOADING
+
+
+def load_torch_model(
+    model: "torch.nn.Module",
+    checkpoint_path: Union[str, os.PathLike],
+    *,
+    strict: bool = False,
+    safe: bool = True,
+    weights_only: bool = False,
+    map_location: Optional[Union[str, "torch.device"]] = None,
+    mmap: bool = False,
+    filename_pattern: Optional[str] = None,
+) -> NamedTuple:
+    """
+    Load a checkpoint into a model, handling both sharded and non-sharded checkpoints.
+
+    Args:
+        model (`torch.nn.Module`):
+            The model in which to load the checkpoint.
+        checkpoint_path (`str` or `os.PathLike`):
+            Path to either the checkpoint file or directory containing the checkpoint(s).
+        strict (`bool`, *optional*, defaults to `False`):
+            Whether to strictly enforce that the keys in the model state dict match the keys in the checkpoint.
+        safe (`bool`, *optional*, defaults to `True`):
+            If `safe` is True, the safetensors files will be loaded. If `safe` is False, the function
+            will first attempt to load safetensors files if they are available, otherwise it will fall back to loading
+            pickle files. `filename_pattern` parameter takes precedence over `safe` parameter.
+        weights_only (`bool`, *optional*, defaults to `False`):
+            If True, only loads the model weights without optimizer states and other metadata.
+            Only supported in PyTorch >= 1.13.
+        map_location (`str` or `torch.device`, *optional*):
+            A `torch.device` object, string or a dict specifying how to remap storage locations. It
+            indicates the location where all tensors should be loaded.
+        mmap (`bool`, *optional*, defaults to `False`):
+            Whether to use memory-mapped file loading. Memory mapping can improve loading performance
+            for large models in PyTorch >= 2.1.0 with zipfile-based checkpoints.
+        filename_pattern (`str`, *optional*):
+            The pattern to look for the index file. Pattern must be a string that
+            can be formatted with `filename_pattern.format(suffix=...)` and must contain the keyword `suffix`
+            Defaults to `"model{suffix}.safetensors"`.
+    Returns:
+        `NamedTuple`: A named tuple with `missing_keys` and `unexpected_keys` fields.
+            - `missing_keys` is a list of str containing the missing keys, i.e. keys that are in the model but not in the checkpoint.
+            - `unexpected_keys` is a list of str containing the unexpected keys, i.e. keys that are in the checkpoint but not in the model.
+
+    Raises:
+        [`FileNotFoundError`](https://docs.python.org/3/library/exceptions.html#FileNotFoundError)
+            If the checkpoint file or directory does not exist.
+        [`ImportError`](https://docs.python.org/3/library/exceptions.html#ImportError)
+            If safetensors or torch is not installed when trying to load a .safetensors file or a PyTorch checkpoint respectively.
+        [`ValueError`](https://docs.python.org/3/library/exceptions.html#ValueError)
+           If the checkpoint path is invalid or if the checkpoint format cannot be determined.
+
+    Example:
+    ```python
+    >>> from huggingface_hub import load_torch_model
+    >>> model = ... # A PyTorch model
+    >>> load_torch_model(model, "path/to/checkpoint")
+    ```
+    """
+    checkpoint_path = Path(checkpoint_path)
+
+    if not checkpoint_path.exists():
+        raise ValueError(f"Checkpoint path {checkpoint_path} does not exist")
+    # 1. Check if checkpoint is a single file
+    if checkpoint_path.is_file():
+        state_dict = load_state_dict_from_file(
+            checkpoint_file=checkpoint_path,
+            map_location=map_location,
+            weights_only=weights_only,
+        )
+        return model.load_state_dict(state_dict, strict=strict)
+
+    # 2. If not, checkpoint_path is a directory
+    if filename_pattern is None:
+        filename_pattern = constants.SAFETENSORS_WEIGHTS_FILE_PATTERN
+        index_path = checkpoint_path / (filename_pattern.format(suffix="") + ".index.json")
+        # Only fallback to pickle format if safetensors index is not found and safe is False.
+        if not index_path.is_file() and not safe:
+            filename_pattern = constants.PYTORCH_WEIGHTS_FILE_PATTERN
+
+    index_path = checkpoint_path / (filename_pattern.format(suffix="") + ".index.json")
+
+    if index_path.is_file():
+        return _load_sharded_checkpoint(
+            model=model,
+            save_directory=checkpoint_path,
+            strict=strict,
+            weights_only=weights_only,
+            filename_pattern=filename_pattern,
+        )
+
+    # Look for single model file
+    model_files = list(checkpoint_path.glob("*.safetensors" if safe else "*.bin"))
+    if len(model_files) == 1:
+        state_dict = load_state_dict_from_file(
+            checkpoint_file=model_files[0],
+            map_location=map_location,
+            weights_only=weights_only,
+            mmap=mmap,
+        )
+        return model.load_state_dict(state_dict, strict=strict)
+
+    raise ValueError(
+        f"Directory '{checkpoint_path}' does not contain a valid checkpoint. "
+        "Expected either a sharded checkpoint with an index file, or a single model file."
+    )
+
+
+def _load_sharded_checkpoint(
+    model: "torch.nn.Module",
+    save_directory: os.PathLike,
+    *,
+    strict: bool = False,
+    weights_only: bool = False,
+    filename_pattern: str = constants.SAFETENSORS_WEIGHTS_FILE_PATTERN,
+) -> NamedTuple:
+    """
+    Loads a sharded checkpoint into a model. This is the same as
+    [`torch.nn.Module.load_state_dict`](https://pytorch.org/docs/stable/generated/torch.nn.Module.html?highlight=load_state_dict#torch.nn.Module.load_state_dict)
+    but for a sharded checkpoint. Each shard is loaded one by one and removed from memory after being loaded into the model.
+
+    Args:
+        model (`torch.nn.Module`):
+            The model in which to load the checkpoint.
+        save_directory (`str` or `os.PathLike`):
+            A path to a folder containing the sharded checkpoint.
+        strict (`bool`, *optional*, defaults to `False`):
+            Whether to strictly enforce that the keys in the model state dict match the keys in the sharded checkpoint.
+        weights_only (`bool`, *optional*, defaults to `False`):
+            If True, only loads the model weights without optimizer states and other metadata.
+            Only supported in PyTorch >= 1.13.
+        filename_pattern (`str`, *optional*, defaults to `"model{suffix}.safetensors"`):
+            The pattern to look for the index file. Pattern must be a string that
+            can be formatted with `filename_pattern.format(suffix=...)` and must contain the keyword `suffix`
+            Defaults to `"model{suffix}.safetensors"`.
+
+    Returns:
+        `NamedTuple`: A named tuple with `missing_keys` and `unexpected_keys` fields,
+            - `missing_keys` is a list of str containing the missing keys
+            - `unexpected_keys` is a list of str containing the unexpected keys
+    """
+
+    # 1. Load and validate index file
+    # The index file contains mapping of parameter names to shard files
+    index_path = filename_pattern.format(suffix="") + ".index.json"
+    index_file = os.path.join(save_directory, index_path)
+    with open(index_file, "r", encoding="utf-8") as f:
+        index = json.load(f)
+
+    # 2. Validate keys if in strict mode
+    # This is done before loading any shards to fail fast
+    if strict:
+        _validate_keys_for_strict_loading(model, index["weight_map"].keys())
+
+    # 3. Load each shard using `load_state_dict`
+    # Get unique shard files (multiple parameters can be in same shard)
+    shard_files = list(set(index["weight_map"].values()))
+    for shard_file in shard_files:
+        # Load shard into memory
+        shard_path = os.path.join(save_directory, shard_file)
+        state_dict = load_state_dict_from_file(
+            shard_path,
+            map_location="cpu",
+            weights_only=weights_only,
+        )
+        # Update model with parameters from this shard
+        model.load_state_dict(state_dict, strict=strict)
+        # Explicitly remove the state dict from memory
+        del state_dict
+
+    # 4. Return compatibility info
+    loaded_keys = set(index["weight_map"].keys())
+    model_keys = set(model.state_dict().keys())
+    return _IncompatibleKeys(
+        missing_keys=list(model_keys - loaded_keys), unexpected_keys=list(loaded_keys - model_keys)
+    )
+
+
+def load_state_dict_from_file(
+    checkpoint_file: Union[str, os.PathLike],
+    map_location: Optional[Union[str, "torch.device"]] = None,
+    weights_only: bool = False,
+    mmap: bool = False,
+) -> Union[Dict[str, "torch.Tensor"], Any]:
+    """
+    Loads a checkpoint file, handling both safetensors and pickle checkpoint formats.
+
+    Args:
+        checkpoint_file (`str` or `os.PathLike`):
+            Path to the checkpoint file to load. Can be either a safetensors or pickle (`.bin`) checkpoint.
+        map_location (`str` or `torch.device`, *optional*):
+            A `torch.device` object, string or a dict specifying how to remap storage locations. It
+            indicates the location where all tensors should be loaded.
+        weights_only (`bool`, *optional*, defaults to `False`):
+            If True, only loads the model weights without optimizer states and other metadata.
+            Only supported for pickle (`.bin`) checkpoints with PyTorch >= 1.13. Has no effect when
+            loading safetensors files.
+        mmap (`bool`, *optional*, defaults to `False`):
+            Whether to use memory-mapped file loading. Memory mapping can improve loading performance
+            for large models in PyTorch >= 2.1.0 with zipfile-based checkpoints. Has no effect when
+            loading safetensors files, as the `safetensors` library uses memory mapping by default.
+
+    Returns:
+        `Union[Dict[str, "torch.Tensor"], Any]`: The loaded checkpoint.
+            - For safetensors files: always returns a dictionary mapping parameter names to tensors.
+            - For pickle files: returns any Python object that was pickled (commonly a state dict, but could be
+              an entire model, optimizer state, or any other Python object).
+
+    Raises:
+        [`FileNotFoundError`](https://docs.python.org/3/library/exceptions.html#FileNotFoundError)
+            If the checkpoint file does not exist.
+        [`ImportError`](https://docs.python.org/3/library/exceptions.html#ImportError)
+            If safetensors or torch is not installed when trying to load a .safetensors file or a PyTorch checkpoint respectively.
+        [`OSError`](https://docs.python.org/3/library/exceptions.html#OSError)
+            If the checkpoint file format is invalid or if git-lfs files are not properly downloaded.
+        [`ValueError`](https://docs.python.org/3/library/exceptions.html#ValueError)
+            If the checkpoint file path is empty or invalid.
+
+    Example:
+    ```python
+    >>> from huggingface_hub import load_state_dict_from_file
+
+    # Load a PyTorch checkpoint
+    >>> state_dict = load_state_dict_from_file("path/to/model.bin", map_location="cpu")
+    >>> model.load_state_dict(state_dict)
+
+    # Load a safetensors checkpoint
+    >>> state_dict = load_state_dict_from_file("path/to/model.safetensors")
+    >>> model.load_state_dict(state_dict)
+    ```
+    """
+    checkpoint_path = Path(checkpoint_file)
+
+    # Check if file exists and is a regular file (not a directory)
+    if not checkpoint_path.is_file():
+        raise FileNotFoundError(
+            f"No checkpoint file found at '{checkpoint_path}'. Please verify the path is correct and "
+            "the file has been properly downloaded."
+        )
+
+    # Load safetensors checkpoint
+    if checkpoint_path.suffix == ".safetensors":
+        try:
+            from safetensors import safe_open
+            from safetensors.torch import load_file
+        except ImportError as e:
+            raise ImportError(
+                "Please install `safetensors` to load safetensors checkpoint. "
+                "You can install it with `pip install safetensors`."
+            ) from e
+
+        # Check format of the archive
+        with safe_open(checkpoint_file, framework="pt") as f:  # type: ignore[attr-defined]
+            metadata = f.metadata()
+        # see comment: https://github.com/huggingface/transformers/blob/3d213b57fe74302e5902d68ed9478c3ad1aaa713/src/transformers/modeling_utils.py#L3966
+        if metadata is not None and metadata.get("format") not in ["pt", "mlx"]:
+            raise OSError(
+                f"The safetensors archive passed at {checkpoint_file} does not contain the valid metadata. Make sure "
+                "you save your model with the `save_torch_model` method."
+            )
+        device = str(map_location.type) if map_location is not None and hasattr(map_location, "type") else map_location
+        # meta device is not supported with safetensors, falling back to CPU
+        if device == "meta":
+            logger.warning("Meta device is not supported with safetensors. Falling back to CPU device.")
+            device = "cpu"
+        return load_file(checkpoint_file, device=device)  # type: ignore[arg-type]
+    # Otherwise, load from pickle
+    try:
+        import torch
+        from torch import load
+    except ImportError as e:
+        raise ImportError(
+            "Please install `torch` to load torch tensors. " "You can install it with `pip install torch`."
+        ) from e
+    # Add additional kwargs, mmap is only supported in torch >= 2.1.0
+    additional_kwargs = {}
+    if version.parse(torch.__version__) >= version.parse("2.1.0"):
+        additional_kwargs["mmap"] = mmap
+
+    # weights_only is only supported in torch >= 1.13.0
+    if version.parse(torch.__version__) >= version.parse("1.13.0"):
+        additional_kwargs["weights_only"] = weights_only
+
+    return load(
+        checkpoint_file,
+        map_location=map_location,
+        **additional_kwargs,
+    )
+
+
+# HELPERS
+
+
+def _validate_keys_for_strict_loading(
+    model: "torch.nn.Module",
+    loaded_keys: Iterable[str],
+) -> None:
+    """
+    Validate that model keys match loaded keys when strict loading is enabled.
+
+    Args:
+        model: The PyTorch model being loaded
+        loaded_keys: The keys present in the checkpoint
+
+    Raises:
+        RuntimeError: If there are missing or unexpected keys in strict mode
+    """
+    loaded_keys_set = set(loaded_keys)
+    model_keys = set(model.state_dict().keys())
+    missing_keys = model_keys - loaded_keys_set  # Keys in model but not in checkpoint
+    unexpected_keys = loaded_keys_set - model_keys  # Keys in checkpoint but not in model
+
+    if missing_keys or unexpected_keys:
+        error_message = f"Error(s) in loading state_dict for {model.__class__.__name__}"
+        if missing_keys:
+            str_missing_keys = ",".join([f'"{k}"' for k in sorted(missing_keys)])
+            error_message += f"\nMissing key(s): {str_missing_keys}."
+        if unexpected_keys:
+            str_unexpected_keys = ",".join([f'"{k}"' for k in sorted(unexpected_keys)])
+            error_message += f"\nUnexpected key(s): {str_unexpected_keys}."
+        raise RuntimeError(error_message)
+
+
 def _get_unique_id(tensor: "torch.Tensor") -> Union[int, Tuple[Any, ...]]:
     """Returns a unique id for plain tensor
     or a (potentially nested) Tuple of unique id for the flattened Tensor
@@ -380,7 +723,7 @@ def _get_unique_id(tensor: "torch.Tensor") -> Union[int, Tuple[Any, ...]]:
         # use some other unique id to distinguish.
         # this is a XLA tensor, it must be created using torch_xla's
         # device. So the following import is safe:
-        import torch_xla
+        import torch_xla  # type: ignore[import]
 
         unique_id = torch_xla._XLAC._xla_get_tensor_id(tensor)
     else:
@@ -444,7 +787,7 @@ def is_torch_tpu_available(check_device=True):
         if check_device:
             # We need to check if `xla_device` can be found, will raise a RuntimeError if not
             try:
-                import torch_xla.core.xla_model as xm
+                import torch_xla.core.xla_model as xm  # type: ignore[import]
 
                 _ = xm.xla_device()
                 return True
@@ -463,7 +806,7 @@ def storage_ptr(tensor: "torch.Tensor") -> Union[int, Tuple[Any, ...]]:
         from torch.utils._python_dispatch import is_traceable_wrapper_subclass
 
         if is_traceable_wrapper_subclass(tensor):
-            return _get_unique_id(tensor)
+            return _get_unique_id(tensor)  # type: ignore
     except ImportError:
         # for torch version less than 2.1, we can fallback to original implementation
         pass
@@ -655,3 +998,18 @@ def _get_dtype_size(dtype: "torch.dtype") -> int:
         _float8_e5m2: 1,
     }
     return _SIZE[dtype]
+
+
+class _IncompatibleKeys(namedtuple("IncompatibleKeys", ["missing_keys", "unexpected_keys"])):
+    """
+    This is used to report missing and unexpected keys in the state dict.
+    Taken from https://github.com/pytorch/pytorch/blob/main/torch/nn/modules/module.py#L52.
+
+    """
+
+    def __repr__(self) -> str:
+        if not self.missing_keys and not self.unexpected_keys:
+            return "<All keys matched successfully>"
+        return super().__repr__()
+
+    __str__ = __repr__

huggingface_hub/utils/_cache_manager.py

@@ -742,7 +742,7 @@ def _scan_cached_repo(repo_path: Path) -> CachedRepoInfo:
 
         for ref_path in refs_path.glob("**/*"):
             # glob("**/*") iterates over all files and directories -> skip directories
-            if ref_path.is_dir():
+            if ref_path.is_dir() or ref_path.name in FILES_TO_IGNORE:
                 continue
 
             ref_name = str(ref_path.relative_to(refs_path))

huggingface_hub/utils/_headers.py

@@ -20,6 +20,7 @@ from huggingface_hub.errors import LocalTokenNotFoundError
 
 from .. import constants
 from ._auth import get_token
+from ._deprecation import _deprecate_arguments
 from ._runtime import (
     get_fastai_version,
     get_fastcore_version,
@@ -35,15 +36,20 @@ from ._runtime import (
 from ._validators import validate_hf_hub_args
 
 
+@_deprecate_arguments(
+    version="1.0",
+    deprecated_args="is_write_action",
+    custom_message="This argument is ignored and we let the server handle the permission error instead (if any).",
+)
 @validate_hf_hub_args
 def build_hf_headers(
     *,
     token: Optional[Union[bool, str]] = None,
-    is_write_action: bool = False,
     library_name: Optional[str] = None,
     library_version: Optional[str] = None,
     user_agent: Union[Dict, str, None] = None,
     headers: Optional[Dict[str, str]] = None,
+    is_write_action: bool = False,
 ) -> Dict[str, str]:
     """
     Build headers dictionary to send in a HF Hub call.
@@ -68,9 +74,6 @@ def build_hf_headers(
                 - if `False`, authorization header is not set
                 - if `None`, the token is read from the machine only except if
                   `HF_HUB_DISABLE_IMPLICIT_TOKEN` env variable is set.
-        is_write_action (`bool`, default to `False`):
-            Set to True if the API call requires a write access. If `True`, the token
-            will be validated (cannot be `None`, cannot start by `"api_org***"`).
         library_name (`str`, *optional*):
             The name of the library that is making the HTTP request. Will be added to
             the user-agent header.
@@ -83,6 +86,8 @@ def build_hf_headers(
         headers (`dict`, *optional*):
             Additional headers to include in the request. Those headers take precedence
             over the ones generated by this function.
+        is_write_action (`bool`):
+            Ignored and deprecated argument.
 
     Returns:
         A `Dict` of headers to pass in your API call.
@@ -105,9 +110,6 @@ def build_hf_headers(
         >>> build_hf_headers() # token is not sent
         {"user-agent": ...}
 
-        >>> build_hf_headers(token="api_org_***", is_write_action=True)
-        ValueError: You must use your personal account token for write-access methods.
-
         >>> build_hf_headers(library_name="transformers", library_version="1.2.3")
         {"authorization": ..., "user-agent": "transformers/1.2.3; hf_hub/0.10.2; python/3.10.4; tensorflow/1.55"}
     ```
@@ -122,7 +124,6 @@ def build_hf_headers(
     """
     # Get auth token to send
     token_to_send = get_token_to_send(token)
-    _validate_token_to_send(token_to_send, is_write_action=is_write_action)
 
     # Combine headers
     hf_headers = {
@@ -171,23 +172,6 @@ def get_token_to_send(token: Optional[Union[bool, str]]) -> Optional[str]:
     return cached_token
 
 
-def _validate_token_to_send(token: Optional[str], is_write_action: bool) -> None:
-    if is_write_action:
-        if token is None:
-            raise ValueError(
-                "Token is required (write-access action) but no token found. You need"
-                " to provide a token or be logged in to Hugging Face with"
-                " `huggingface-cli login` or `huggingface_hub.login`. See"
-                " https://huggingface.co/settings/tokens."
-            )
-        if token.startswith("api_org"):
-            raise ValueError(
-                "You must use your personal account token for write-access methods. To"
-                " generate a write-access token, go to"
-                " https://huggingface.co/settings/tokens"
-            )
-
-
 def _http_user_agent(
     *,
     library_name: Optional[str] = None,

huggingface_hub/utils/tqdm.py

@@ -81,6 +81,8 @@ Group-based control:
 """
 
 import io
+import logging
+import os
 import warnings
 from contextlib import contextmanager
 from pathlib import Path
@@ -196,6 +198,19 @@ def are_progress_bars_disabled(name: Optional[str] = None) -> bool:
     return not progress_bar_states.get("_global", True)
 
 
+def is_tqdm_disabled(log_level: int) -> Optional[bool]:
+    """
+    Determine if tqdm progress bars should be disabled based on logging level and environment settings.
+
+    see https://github.com/huggingface/huggingface_hub/pull/2000 and https://github.com/huggingface/huggingface_hub/pull/2698.
+    """
+    if log_level == logging.NOTSET:
+        return True
+    if os.getenv("TQDM_POSITION") == "-1":
+        return False
+    return None
+
+
 class tqdm(old_tqdm):
     """
     Class to override `disable` argument in case progress bars are globally disabled.

{huggingface_hub-0.26.5.dist-info → huggingface_hub-0.27.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: huggingface-hub
-Version: 0.26.5
+Version: 0.27.0
 Summary: Client library to download and publish models, datasets and other repos on the huggingface.co hub
 Home-page: https://github.com/huggingface/huggingface_hub
 Author: Hugging Face, Inc.
@@ -20,6 +20,7 @@ Classifier: Programming Language :: Python :: 3.9
 Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
 Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
 Requires-Python: >=3.8.0
 Description-Content-Type: text/markdown
@@ -142,10 +143,14 @@ Requires-Dist: types-tqdm; extra == "typing"
 Requires-Dist: types-urllib3; extra == "typing"
 
 <p align="center">
+  <picture>
+    <source media="(prefers-color-scheme: dark)" srcset="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/huggingface_hub-dark.svg">
+    <source media="(prefers-color-scheme: light)" srcset="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/huggingface_hub.svg">
+    <img alt="huggingface_hub library logo" src="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/huggingface_hub.svg" width="352" height="59" style="max-width: 100%;">
+  </picture>
   <br/>
-    <img alt="huggingface_hub library logo" src="https://huggingface.co/datasets/huggingface/documentation-images/raw/main/huggingface_hub.svg" width="376" height="59" style="max-width: 100%;">
   <br/>
-</p>
+</p> 
 
 <p align="center">
     <i>The official Python client for the Huggingface Hub.</i>