huggingface-hub 0.33.4__py3-none-any.whl → 0.34.0rc0__py3-none-any.whl

huggingface_hub/hf_file_system.py

@@ -367,7 +367,6 @@ class HfFileSystem(fsspec.AbstractFileSystem):
         """
         resolved_path = self.resolve_path(path, revision=revision)
         path = resolved_path.unresolve()
-        kwargs = {"expand_info": detail, **kwargs}
         try:
             out = self._ls_tree(path, refresh=refresh, revision=revision, **kwargs)
         except EntryNotFoundError:
@@ -386,7 +385,7 @@ class HfFileSystem(fsspec.AbstractFileSystem):
         recursive: bool = False,
         refresh: bool = False,
         revision: Optional[str] = None,
-        expand_info: bool = True,
+        expand_info: bool = False,
     ):
         resolved_path = self.resolve_path(path, revision=revision)
         path = resolved_path.unresolve()
@@ -497,8 +496,6 @@ class HfFileSystem(fsspec.AbstractFileSystem):
         Returns:
             `Iterator[Tuple[str, List[str], List[str]]]`: An iterator of (path, list of directory names, list of file names) tuples.
         """
-        # Set expand_info=False by default to get a x10 speed boost
-        kwargs = {"expand_info": kwargs.get("detail", False), **kwargs}
         path = self.resolve_path(path, revision=kwargs.get("revision")).unresolve()
         yield from super().walk(path, *args, **kwargs)
 
@@ -515,8 +512,6 @@ class HfFileSystem(fsspec.AbstractFileSystem):
         Returns:
             `List[str]`: List of paths matching the pattern.
         """
-        # Set expand_info=False by default to get a x10 speed boost
-        kwargs = {"expand_info": kwargs.get("detail", False), **kwargs}
         path = self.resolve_path(path, revision=kwargs.get("revision")).unresolve()
         return super().glob(path, **kwargs)
 
@@ -558,7 +553,6 @@ class HfFileSystem(fsspec.AbstractFileSystem):
             )
         resolved_path = self.resolve_path(path, revision=revision)
         path = resolved_path.unresolve()
-        kwargs = {"expand_info": detail, **kwargs}
         try:
             out = self._ls_tree(path, recursive=True, refresh=refresh, revision=resolved_path.revision, **kwargs)
         except EntryNotFoundError:
@@ -653,7 +647,7 @@ class HfFileSystem(fsspec.AbstractFileSystem):
         Returns:
             `datetime`: Last commit date of the file.
         """
-        info = self.info(path, **kwargs)
+        info = self.info(path, **{**kwargs, "expand_info": True})
         return info["last_commit"]["date"]
 
     def info(self, path: str, refresh: bool = False, revision: Optional[str] = None, **kwargs) -> Dict[str, Any]:
@@ -683,7 +677,7 @@ class HfFileSystem(fsspec.AbstractFileSystem):
         resolved_path = self.resolve_path(path, revision=revision)
         path = resolved_path.unresolve()
         expand_info = kwargs.get(
-            "expand_info", True
+            "expand_info", False
         )  # don't expose it as a parameter in the public API to follow the spec
         if not resolved_path.path_in_repo:
             # Path is the root directory
@@ -691,6 +685,7 @@ class HfFileSystem(fsspec.AbstractFileSystem):
                 "name": path,
                 "size": 0,
                 "type": "directory",
+                "last_commit": None,
             }
             if expand_info:
                 last_commit = self._api.list_repo_commits(
@@ -708,7 +703,7 @@ class HfFileSystem(fsspec.AbstractFileSystem):
             parent_path = self._parent(path)
             if not expand_info and parent_path not in self.dircache:
                 # Fill the cache with cheap call
-                self.ls(parent_path, expand_info=False)
+                self.ls(parent_path)
             if parent_path in self.dircache:
                 # Check if the path is in the cache
                 out1 = [o for o in self.dircache[parent_path] if o["name"] == path]
@@ -779,7 +774,7 @@ class HfFileSystem(fsspec.AbstractFileSystem):
             if kwargs.get("refresh", False):
                 self.invalidate_cache(path)
 
-            self.info(path, **{**kwargs, "expand_info": False})
+            self.info(path, **kwargs)
             return True
         except:  # noqa: E722
             return False
@@ -798,7 +793,7 @@ class HfFileSystem(fsspec.AbstractFileSystem):
             `bool`: True if path is a directory, False otherwise.
         """
         try:
-            return self.info(path, expand_info=False)["type"] == "directory"
+            return self.info(path)["type"] == "directory"
         except OSError:
             return False
 
@@ -816,7 +811,7 @@ class HfFileSystem(fsspec.AbstractFileSystem):
             `bool`: True if path is a file, False otherwise.
         """
         try:
-            return self.info(path, expand_info=False)["type"] == "file"
+            return self.info(path)["type"] == "file"
         except:  # noqa: E722
             return False
 
@@ -942,9 +937,6 @@ class HfFileSystemFile(fsspec.spec.AbstractBufferedFile):
                     f"{e}.\nMake sure the repository and revision exist before writing data."
                 ) from e
             raise
-        # avoid an unnecessary .info() call with expensive expand_info=True to instantiate .details
-        if kwargs.get("mode", "rb") == "rb":
-            self.details = fs.info(self.resolved_path.unresolve(), expand_info=False)
         super().__init__(fs, self.resolved_path.unresolve(), **kwargs)
         self.fs: HfFileSystem
 

huggingface_hub/hub_mixin.py

@@ -491,7 +491,7 @@ class ModelHubMixin:
                 'http://hostname': 'foo.bar:4012'}`. The proxies are used on every request.
             token (`str` or `bool`, *optional*):
                 The token to use as HTTP bearer authorization for remote files. By default, it will use the token
-                cached when running `huggingface-cli login`.
+                cached when running `hf auth login`.
             cache_dir (`str`, `Path`, *optional*):
                 Path to the folder where cached files are stored.
             local_files_only (`bool`, *optional*, defaults to `False`):
@@ -619,7 +619,7 @@ class ModelHubMixin:
                 'http://hostname': 'foo.bar:4012'}`).
             token (`str` or `bool`, *optional*):
                 The token to use as HTTP bearer authorization for remote files. By default, it will use the token
-                cached when running `huggingface-cli login`.
+                cached when running `hf auth login`.
             cache_dir (`str`, `Path`, *optional*):
                 Path to the folder where cached files are stored.
             local_files_only (`bool`, *optional*, defaults to `False`):
@@ -664,7 +664,7 @@ class ModelHubMixin:
                 If `None` (default), the repo will be public unless the organization's default is private.
             token (`str`, *optional*):
                 The token to use as HTTP bearer authorization for remote files. By default, it will use the token
-                cached when running `huggingface-cli login`.
+                cached when running `hf auth login`.
             branch (`str`, *optional*):
                 The git branch on which to push the model. This defaults to `"main"`.
             create_pr (`boolean`, *optional*):

huggingface_hub/inference/_client.py

@@ -130,9 +130,7 @@ class InferenceClient:
             or a URL to a deployed Inference Endpoint. Defaults to None, in which case a recommended model is
             automatically selected for the task.
             Note: for better compatibility with OpenAI's client, `model` has been aliased as `base_url`. Those 2
-            arguments are mutually exclusive. If using `base_url` for chat completion, the `/chat/completions` suffix
-            path will be appended to the base URL (see the [TGI Messages API](https://huggingface.co/docs/text-generation-inference/en/messages_api)
-            documentation for details). When passing a URL as `model`, the client will not append any suffix path to it.
+            arguments are mutually exclusive. If a URL is passed as `model` or `base_url` for chat completion, the `(/v1)/chat/completions` suffix path will be appended to the URL.
         provider (`str`, *optional*):
             Name of the provider to use for inference. Can be `"black-forest-labs"`, `"cerebras"`, `"cohere"`, `"fal-ai"`, `"featherless-ai"`, `"fireworks-ai"`, `"groq"`, `"hf-inference"`, `"hyperbolic"`, `"nebius"`, `"novita"`, `"nscale"`, `"openai"`, `"replicate"`, "sambanova"` or `"together"`.
             Defaults to "auto" i.e. the first of the providers available for the model, sorted by the user's order in https://hf.co/settings/inference-providers.
@@ -947,8 +945,8 @@ class InferenceClient:
         Answer questions on document images.
 
         Args:
-            image (`Union[str, Path, bytes, BinaryIO, PIL.Image.Image]`):
-                The input image for the context. It can be raw bytes, an image file, a URL to an online image, or a PIL Image.
+            image (`Union[str, Path, bytes, BinaryIO]`):
+                The input image for the context. It can be raw bytes, an image file, or a URL to an online image.
             question (`str`):
                 Question to be answered.
             model (`str`, *optional*):
@@ -1338,6 +1336,7 @@ class InferenceClient:
             api_key=self.token,
         )
         response = self._inner_post(request_parameters)
+        response = provider_helper.get_response(response, request_parameters)
         return _bytes_to_image(response)
 
     def image_to_text(self, image: ContentT, *, model: Optional[str] = None) -> ImageToTextOutput:
@@ -1858,23 +1857,23 @@ class InferenceClient:
         return TextClassificationOutputElement.parse_obj_as_list(response)[0]  # type: ignore [return-value]
 
     @overload
-    def text_generation(  # type: ignore
+    def text_generation(
         self,
         prompt: str,
         *,
-        details: Literal[False] = ...,
-        stream: Literal[False] = ...,
+        details: Literal[True],
+        stream: Literal[True],
         model: Optional[str] = None,
         # Parameters from `TextGenerationInputGenerateParameters` (maintained manually)
         adapter_id: Optional[str] = None,
         best_of: Optional[int] = None,
         decoder_input_details: Optional[bool] = None,
-        do_sample: Optional[bool] = False,  # Manual default value
+        do_sample: Optional[bool] = None,
         frequency_penalty: Optional[float] = None,
         grammar: Optional[TextGenerationInputGrammarType] = None,
         max_new_tokens: Optional[int] = None,
         repetition_penalty: Optional[float] = None,
-        return_full_text: Optional[bool] = False,  # Manual default value
+        return_full_text: Optional[bool] = None,
         seed: Optional[int] = None,
         stop: Optional[List[str]] = None,
         stop_sequences: Optional[List[str]] = None,  # Deprecated, use `stop` instead
@@ -1885,26 +1884,26 @@ class InferenceClient:
         truncate: Optional[int] = None,
         typical_p: Optional[float] = None,
         watermark: Optional[bool] = None,
-    ) -> str: ...
+    ) -> Iterable[TextGenerationStreamOutput]: ...
 
     @overload
-    def text_generation(  # type: ignore
+    def text_generation(
         self,
         prompt: str,
         *,
-        details: Literal[True] = ...,
-        stream: Literal[False] = ...,
+        details: Literal[True],
+        stream: Optional[Literal[False]] = None,
         model: Optional[str] = None,
         # Parameters from `TextGenerationInputGenerateParameters` (maintained manually)
         adapter_id: Optional[str] = None,
         best_of: Optional[int] = None,
         decoder_input_details: Optional[bool] = None,
-        do_sample: Optional[bool] = False,  # Manual default value
+        do_sample: Optional[bool] = None,
         frequency_penalty: Optional[float] = None,
         grammar: Optional[TextGenerationInputGrammarType] = None,
         max_new_tokens: Optional[int] = None,
         repetition_penalty: Optional[float] = None,
-        return_full_text: Optional[bool] = False,  # Manual default value
+        return_full_text: Optional[bool] = None,
         seed: Optional[int] = None,
         stop: Optional[List[str]] = None,
         stop_sequences: Optional[List[str]] = None,  # Deprecated, use `stop` instead
@@ -1918,23 +1917,23 @@ class InferenceClient:
     ) -> TextGenerationOutput: ...
 
     @overload
-    def text_generation(  # type: ignore
+    def text_generation(
         self,
         prompt: str,
         *,
-        details: Literal[False] = ...,
-        stream: Literal[True] = ...,
+        details: Optional[Literal[False]] = None,
+        stream: Literal[True],
         model: Optional[str] = None,
         # Parameters from `TextGenerationInputGenerateParameters` (maintained manually)
         adapter_id: Optional[str] = None,
         best_of: Optional[int] = None,
         decoder_input_details: Optional[bool] = None,
-        do_sample: Optional[bool] = False,  # Manual default value
+        do_sample: Optional[bool] = None,
         frequency_penalty: Optional[float] = None,
         grammar: Optional[TextGenerationInputGrammarType] = None,
         max_new_tokens: Optional[int] = None,
         repetition_penalty: Optional[float] = None,
-        return_full_text: Optional[bool] = False,  # Manual default value
+        return_full_text: Optional[bool] = None,  # Manual default value
         seed: Optional[int] = None,
         stop: Optional[List[str]] = None,
         stop_sequences: Optional[List[str]] = None,  # Deprecated, use `stop` instead
@@ -1948,23 +1947,23 @@ class InferenceClient:
     ) -> Iterable[str]: ...
 
     @overload
-    def text_generation(  # type: ignore
+    def text_generation(
         self,
         prompt: str,
         *,
-        details: Literal[True] = ...,
-        stream: Literal[True] = ...,
+        details: Optional[Literal[False]] = None,
+        stream: Optional[Literal[False]] = None,
         model: Optional[str] = None,
         # Parameters from `TextGenerationInputGenerateParameters` (maintained manually)
         adapter_id: Optional[str] = None,
         best_of: Optional[int] = None,
         decoder_input_details: Optional[bool] = None,
-        do_sample: Optional[bool] = False,  # Manual default value
+        do_sample: Optional[bool] = None,
         frequency_penalty: Optional[float] = None,
         grammar: Optional[TextGenerationInputGrammarType] = None,
         max_new_tokens: Optional[int] = None,
         repetition_penalty: Optional[float] = None,
-        return_full_text: Optional[bool] = False,  # Manual default value
+        return_full_text: Optional[bool] = None,
         seed: Optional[int] = None,
         stop: Optional[List[str]] = None,
         stop_sequences: Optional[List[str]] = None,  # Deprecated, use `stop` instead
@@ -1975,26 +1974,26 @@ class InferenceClient:
         truncate: Optional[int] = None,
         typical_p: Optional[float] = None,
         watermark: Optional[bool] = None,
-    ) -> Iterable[TextGenerationStreamOutput]: ...
+    ) -> str: ...
 
     @overload
     def text_generation(
         self,
         prompt: str,
         *,
-        details: Literal[True] = ...,
-        stream: bool = ...,
+        details: Optional[bool] = None,
+        stream: Optional[bool] = None,
         model: Optional[str] = None,
         # Parameters from `TextGenerationInputGenerateParameters` (maintained manually)
         adapter_id: Optional[str] = None,
         best_of: Optional[int] = None,
         decoder_input_details: Optional[bool] = None,
-        do_sample: Optional[bool] = False,  # Manual default value
+        do_sample: Optional[bool] = None,
         frequency_penalty: Optional[float] = None,
         grammar: Optional[TextGenerationInputGrammarType] = None,
         max_new_tokens: Optional[int] = None,
         repetition_penalty: Optional[float] = None,
-        return_full_text: Optional[bool] = False,  # Manual default value
+        return_full_text: Optional[bool] = None,
         seed: Optional[int] = None,
         stop: Optional[List[str]] = None,
         stop_sequences: Optional[List[str]] = None,  # Deprecated, use `stop` instead
@@ -2005,25 +2004,25 @@ class InferenceClient:
         truncate: Optional[int] = None,
         typical_p: Optional[float] = None,
         watermark: Optional[bool] = None,
-    ) -> Union[TextGenerationOutput, Iterable[TextGenerationStreamOutput]]: ...
+    ) -> Union[str, TextGenerationOutput, Iterable[str], Iterable[TextGenerationStreamOutput]]: ...
 
     def text_generation(
         self,
         prompt: str,
         *,
-        details: bool = False,
-        stream: bool = False,
+        details: Optional[bool] = None,
+        stream: Optional[bool] = None,
         model: Optional[str] = None,
         # Parameters from `TextGenerationInputGenerateParameters` (maintained manually)
         adapter_id: Optional[str] = None,
         best_of: Optional[int] = None,
         decoder_input_details: Optional[bool] = None,
-        do_sample: Optional[bool] = False,  # Manual default value
+        do_sample: Optional[bool] = None,
         frequency_penalty: Optional[float] = None,
         grammar: Optional[TextGenerationInputGrammarType] = None,
         max_new_tokens: Optional[int] = None,
         repetition_penalty: Optional[float] = None,
-        return_full_text: Optional[bool] = False,  # Manual default value
+        return_full_text: Optional[bool] = None,
         seed: Optional[int] = None,
         stop: Optional[List[str]] = None,
         stop_sequences: Optional[List[str]] = None,  # Deprecated, use `stop` instead
@@ -2101,7 +2100,7 @@ class InferenceClient:
             typical_p (`float`, *optional`):
                 Typical Decoding mass
                 See [Typical Decoding for Natural Language Generation](https://arxiv.org/abs/2202.00666) for more information
-            watermark (`bool`, *optional`):
+            watermark (`bool`, *optional*):
                 Watermarking with [A Watermark for Large Language Models](https://arxiv.org/abs/2301.10226)
 
         Returns:
@@ -2251,7 +2250,7 @@ class InferenceClient:
             "repetition_penalty": repetition_penalty,
             "return_full_text": return_full_text,
             "seed": seed,
-            "stop": stop if stop is not None else [],
+            "stop": stop,
             "temperature": temperature,
             "top_k": top_k,
             "top_n_tokens": top_n_tokens,
@@ -2305,7 +2304,7 @@ class InferenceClient:
 
         # Handle errors separately for more precise error messages
         try:
-            bytes_output = self._inner_post(request_parameters, stream=stream)
+            bytes_output = self._inner_post(request_parameters, stream=stream or False)
         except HTTPError as e:
             match = MODEL_KWARGS_NOT_USED_REGEX.search(str(e))
             if isinstance(e, BadRequestError) and match:

huggingface_hub/inference/_common.py

@@ -18,6 +18,7 @@ import base64
 import io
 import json
 import logging
+import mimetypes
 from contextlib import contextmanager
 from dataclasses import dataclass
 from pathlib import Path
@@ -160,9 +161,9 @@ def _open_as_binary(
 
 @contextmanager  # type: ignore
 def _open_as_binary(content: Optional[ContentT]) -> Generator[Optional[BinaryT], None, None]:
-    """Open `content` as a binary file, either from a URL, a local path, or raw bytes.
+    """Open `content` as a binary file, either from a URL, a local path, raw bytes, or a PIL Image.
 
-    Do nothing if `content` is None,
+    Do nothing if `content` is None.
 
     TODO: handle base64 as input
     """
@@ -184,15 +185,21 @@ def _open_as_binary(content: Optional[ContentT]) -> Generator[Optional[BinaryT],
         logger.debug(f"Opening content from {content}")
         with content.open("rb") as f:
             yield f
-    elif hasattr(content, "save"):  # PIL Image
-        logger.debug("Converting PIL Image to bytes")
-        buffer = io.BytesIO()
-        content.save(buffer, format="PNG")
-        buffer.seek(0)
-        yield buffer
-    else:
-        # Otherwise: already a file-like object or None
-        yield content
+        return
+
+    # If content is a PIL Image => convert to bytes
+    if is_pillow_available():
+        from PIL import Image
+
+        if isinstance(content, Image.Image):
+            logger.debug("Converting PIL Image to bytes")
+            buffer = io.BytesIO()
+            content.save(buffer, format=content.format or "PNG")
+            yield buffer.getvalue()
+            return
+
+    # Otherwise: already a file-like object or None
+    yield content  # type: ignore
 
 
 def _b64_encode(content: ContentT) -> str:
@@ -202,6 +209,26 @@ def _b64_encode(content: ContentT) -> str:
         return base64.b64encode(data_as_bytes).decode()
 
 
+def _as_url(content: ContentT, default_mime_type: str) -> str:
+    if isinstance(content, str) and (content.startswith("https://") or content.startswith("http://")):
+        return content
+
+    # Handle MIME type detection for different content types
+    mime_type = None
+    if isinstance(content, (str, Path)):
+        mime_type = mimetypes.guess_type(content, strict=False)[0]
+    elif is_pillow_available():
+        from PIL import Image
+
+        if isinstance(content, Image.Image):
+            # Determine MIME type from PIL Image format, in sync with `_open_as_binary`
+            mime_type = f"image/{(content.format or 'PNG').lower()}"
+
+    mime_type = mime_type or default_mime_type
+    encoded_data = _b64_encode(content)
+    return f"data:{mime_type};base64,{encoded_data}"
+
+
 def _b64_to_image(encoded_image: str) -> "Image":
     """Parse a base64-encoded string into a PIL Image."""
     Image = _import_pil_image()
@@ -347,9 +374,12 @@ def _format_chat_completion_stream_output(
 
 
 async def _async_yield_from(client: "ClientSession", response: "ClientResponse") -> AsyncIterable[bytes]:
-    async for byte_payload in response.content:
-        yield byte_payload.strip()
-    await client.close()
+    try:
+        async for byte_payload in response.content:
+            yield byte_payload.strip()
+    finally:
+        # Always close the underlying HTTP session to avoid resource leaks
+        await client.close()
 
 
 # "TGI servers" are servers running with the `text-generation-inference` backend.