huggingface-hub 0.24.0rc0__tar.gz → 0.24.2__tar.gz

{huggingface_hub-0.24.0rc0 → huggingface_hub-0.24.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: huggingface_hub
-Version: 0.24.0rc0
+Version: 0.24.2
 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.

{huggingface_hub-0.24.0rc0 → huggingface_hub-0.24.2}/src/huggingface_hub/__init__.py

@@ -46,7 +46,7 @@ import sys
 from typing import TYPE_CHECKING
 
 
-__version__ = "0.24.0.rc0"
+__version__ = "0.24.2"
 
 # Alphabetical order of definitions is ensured in tests
 # WARNING: any comment added in this dictionary definition will be lost when

{huggingface_hub-0.24.0rc0 → huggingface_hub-0.24.2}/src/huggingface_hub/hf_api.py

@@ -149,7 +149,6 @@ ExpandModelProperty_T = Literal[
     "downloads",
     "downloadsAllTime",
     "gated",
-    "gitalyUid",
     "inference",
     "lastModified",
     "library_name",
@@ -177,7 +176,6 @@ ExpandDatasetProperty_T = Literal[
     "downloads",
     "downloadsAllTime",
     "gated",
-    "gitalyUid",
     "lastModified",
     "likes",
     "paperswithcode_id",
@@ -192,7 +190,6 @@ ExpandSpaceProperty_T = Literal[
     "cardData",
     "datasets",
     "disabled",
-    "gitalyUid",
     "lastModified",
     "createdAt",
     "likes",
@@ -1633,7 +1630,7 @@ class HfApi:
             expand (`List[ExpandModelProperty_T]`, *optional*):
                 List properties to return in the response. When used, only the properties in the list will be returned.
                 This parameter cannot be used if `full`, `cardData` or `fetch_config` are passed.
-                Possible values are `"author"`, `"cardData"`, `"config"`, `"createdAt"`, `"disabled"`, `"downloads"`, `"downloadsAllTime"`, `"gated"`, `"gitalyUid"`, `"inference"`, `"lastModified"`, `"library_name"`, `"likes"`, `"mask_token"`, `"model-index"`, `"pipeline_tag"`, `"private"`, `"safetensors"`, `"sha"`, `"siblings"`, `"spaces"`, `"tags"`, `"transformersInfo"` and `"widgetData"`.
+                Possible values are `"author"`, `"cardData"`, `"config"`, `"createdAt"`, `"disabled"`, `"downloads"`, `"downloadsAllTime"`, `"gated"`, `"inference"`, `"lastModified"`, `"library_name"`, `"likes"`, `"mask_token"`, `"model-index"`, `"pipeline_tag"`, `"private"`, `"safetensors"`, `"sha"`, `"siblings"`, `"spaces"`, `"tags"`, `"transformersInfo"` and `"widgetData"`.
             full (`bool`, *optional*):
                 Whether to fetch all model data, including the `last_modified`,
                 the `sha`, the files and the `tags`. This is set to `True` by
@@ -1836,7 +1833,7 @@ class HfApi:
             expand (`List[ExpandDatasetProperty_T]`, *optional*):
                 List properties to return in the response. When used, only the properties in the list will be returned.
                 This parameter cannot be used if `full` is passed.
-                Possible values are `"author"`, `"cardData"`, `"citation"`, `"createdAt"`, `"disabled"`, `"description"`, `"downloads"`, `"downloadsAllTime"`, `"gated"`, `"gitalyUid"`, `"lastModified"`, `"likes"`, `"paperswithcode_id"`, `"private"`, `"siblings"`, `"sha"` and `"tags"`.
+                Possible values are `"author"`, `"cardData"`, `"citation"`, `"createdAt"`, `"disabled"`, `"description"`, `"downloads"`, `"downloadsAllTime"`, `"gated"`, `"lastModified"`, `"likes"`, `"paperswithcode_id"`, `"private"`, `"siblings"`, `"sha"` and `"tags"`.
             full (`bool`, *optional*):
                 Whether to fetch all dataset data, including the `last_modified`,
                 the `card_data` and  the files. Can contain useful information such as the
@@ -2017,7 +2014,7 @@ class HfApi:
             expand (`List[ExpandSpaceProperty_T]`, *optional*):
                 List properties to return in the response. When used, only the properties in the list will be returned.
                 This parameter cannot be used if `full` is passed.
-                Possible values are `"author"`, `"cardData"`, `"datasets"`, `"disabled"`, `"gitalyUid"`, `"lastModified"`, `"createdAt"`, `"likes"`, `"private"`, `"runtime"`, `"sdk"`, `"siblings"`, `"sha"`, `"subdomain"`, `"tags"` and `"models"`.
+                Possible values are `"author"`, `"cardData"`, `"datasets"`, `"disabled"`, `"lastModified"`, `"createdAt"`, `"likes"`, `"private"`, `"runtime"`, `"sdk"`, `"siblings"`, `"sha"`, `"subdomain"`, `"tags"` and `"models"`.
             full (`bool`, *optional*):
                 Whether to fetch all Spaces data, including the `last_modified`, `siblings`
                 and `card_data` fields.
@@ -2334,7 +2331,7 @@ class HfApi:
             expand (`List[ExpandModelProperty_T]`, *optional*):
                 List properties to return in the response. When used, only the properties in the list will be returned.
                 This parameter cannot be used if `securityStatus` or `files_metadata` are passed.
-                Possible values are `"author"`, `"cardData"`, `"config"`, `"createdAt"`, `"disabled"`, `"downloads"`, `"downloadsAllTime"`, `"gated"`, `"gitalyUid"`, `"inference"`, `"lastModified"`, `"library_name"`, `"likes"`, `"mask_token"`, `"model-index"`, `"pipeline_tag"`, `"private"`, `"safetensors"`, `"sha"`, `"siblings"`, `"spaces"`, `"tags"`, `"transformersInfo"` and `"widgetData"`.
+                Possible values are `"author"`, `"cardData"`, `"config"`, `"createdAt"`, `"disabled"`, `"downloads"`, `"downloadsAllTime"`, `"gated"`, `"inference"`, `"lastModified"`, `"library_name"`, `"likes"`, `"mask_token"`, `"model-index"`, `"pipeline_tag"`, `"private"`, `"safetensors"`, `"sha"`, `"siblings"`, `"spaces"`, `"tags"`, `"transformersInfo"` and `"widgetData"`.
             token (Union[bool, str, None], optional):
                 A valid user access token (string). Defaults to the locally saved
                 token, which is the recommended method for authentication (see
@@ -2408,7 +2405,7 @@ class HfApi:
             expand (`List[ExpandDatasetProperty_T]`, *optional*):
                 List properties to return in the response. When used, only the properties in the list will be returned.
                 This parameter cannot be used if `files_metadata` is passed.
-                Possible values are `"author"`, `"cardData"`, `"citation"`, `"createdAt"`, `"disabled"`, `"description"`, `"downloads"`, `"downloadsAllTime"`, `"gated"`, `"gitalyUid"`, `"lastModified"`, `"likes"`, `"paperswithcode_id"`, `"private"`, `"siblings"`, `"sha"` and `"tags"`.
+                Possible values are `"author"`, `"cardData"`, `"citation"`, `"createdAt"`, `"disabled"`, `"description"`, `"downloads"`, `"downloadsAllTime"`, `"gated"`, `"lastModified"`, `"likes"`, `"paperswithcode_id"`, `"private"`, `"siblings"`, `"sha"` and `"tags"`.
             token (Union[bool, str, None], optional):
                 A valid user access token (string). Defaults to the locally saved
                 token, which is the recommended method for authentication (see
@@ -2481,7 +2478,7 @@ class HfApi:
             expand (`List[ExpandSpaceProperty_T]`, *optional*):
                 List properties to return in the response. When used, only the properties in the list will be returned.
                 This parameter cannot be used if `full` is passed.
-                Possible values are `"author"`, `"cardData"`, `"datasets"`, `"disabled"`, `"gitalyUid"`, `"lastModified"`, `"createdAt"`, `"likes"`, `"private"`, `"runtime"`, `"sdk"`, `"siblings"`, `"sha"`, `"subdomain"`, `"tags"` and `"models"`.
+                Possible values are `"author"`, `"cardData"`, `"datasets"`, `"disabled"`, `"lastModified"`, `"createdAt"`, `"likes"`, `"private"`, `"runtime"`, `"sdk"`, `"siblings"`, `"sha"`, `"subdomain"`, `"tags"` and `"models"`.
             token (Union[bool, str, None], optional):
                 A valid user access token (string). Defaults to the locally saved
                 token, which is the recommended method for authentication (see
@@ -3804,7 +3801,7 @@ class HfApi:
 
             # Get latest commit info
             try:
-                info = self.repo_info(repo_id=repo_id, repo_type=repo_type, revision=revision, token=token)
+                info = self.repo_info(repo_id=repo_id, repo_type=repo_type, revision=unquoted_revision, token=token)
             except RepositoryNotFoundError as e:
                 e.append_to_message(_CREATE_COMMIT_NO_REPO_ERROR_MESSAGE)
                 raise
@@ -3825,7 +3822,7 @@ class HfApi:
             repo_type=repo_type,
             repo_id=repo_id,
             headers=headers,
-            revision=revision,
+            revision=unquoted_revision,
             endpoint=self.endpoint,
         )
         commit_payload = _prepare_commit_payload(

{huggingface_hub-0.24.0rc0 → huggingface_hub-0.24.2}/src/huggingface_hub/inference/_client.py

@@ -66,11 +66,9 @@ from huggingface_hub.inference._common import (
     _fetch_recommended_models,
     _get_unsupported_text_generation_kwargs,
     _import_numpy,
-    _is_chat_completion_server,
     _open_as_binary,
-    _set_as_non_chat_completion_server,
     _set_unsupported_text_generation_kwargs,
-    _stream_chat_completion_response_from_bytes,
+    _stream_chat_completion_response,
     _stream_text_generation_response,
     raise_text_generation_error,
 )
@@ -82,8 +80,6 @@ from huggingface_hub.inference._generated.types import (
     ChatCompletionInputTool,
     ChatCompletionInputToolTypeClass,
     ChatCompletionOutput,
-    ChatCompletionOutputComplete,
-    ChatCompletionOutputMessage,
     ChatCompletionStreamOutput,
     DocumentQuestionAnsweringOutputElement,
     FillMaskOutputElement,
@@ -189,7 +185,7 @@ class InferenceClient:
             )
 
         self.model: Optional[str] = model
-        self.token: Union[str, bool, None] = token or api_key
+        self.token: Union[str, bool, None] = token if token is not None else api_key
         self.headers = CaseInsensitiveDict(build_hf_headers(token=self.token))  # 'authorization' + 'user-agent'
         if headers is not None:
             self.headers.update(headers)
@@ -818,123 +814,52 @@ class InferenceClient:
         # since `chat_completion(..., model=xxx)` is also a payload parameter for the
         # server, we need to handle it differently
         model = self.base_url or self.model or model or self.get_recommended_model("text-generation")
+        is_url = model.startswith(("http://", "https://"))
+
+        # First, resolve the model chat completions URL
+        if model == self.base_url:
+            # base_url passed => add server route
+            model_url = model + "/v1/chat/completions"
+        elif is_url:
+            # model is a URL => use it directly
+            model_url = model
+        else:
+            # model is a model ID => resolve it + add server route
+            model_url = self._resolve_url(model) + "/v1/chat/completions"
+
+        # `model` is sent in the payload. Not used by the server but can be useful for debugging/routing.
+        # If it's a ID on the Hub => use it. Otherwise, we use a random string.
+        model_id = model if not is_url and model.count("/") == 1 else "tgi"
+
+        data = self.post(
+            model=model_url,
+            json=dict(
+                model=model_id,
+                messages=messages,
+                frequency_penalty=frequency_penalty,
+                logit_bias=logit_bias,
+                logprobs=logprobs,
+                max_tokens=max_tokens,
+                n=n,
+                presence_penalty=presence_penalty,
+                response_format=response_format,
+                seed=seed,
+                stop=stop,
+                temperature=temperature,
+                tool_choice=tool_choice,
+                tool_prompt=tool_prompt,
+                tools=tools,
+                top_logprobs=top_logprobs,
+                top_p=top_p,
+                stream=stream,
+            ),
+            stream=stream,
+        )
 
-        if _is_chat_completion_server(model):
-            # First, let's consider the server has a `/v1/chat/completions` endpoint.
-            # If that's the case, we don't have to render the chat template client-side.
-            model_url = self._resolve_url(model)
-            if not model_url.endswith("/chat/completions"):
-                model_url += "/v1/chat/completions"
-
-            # `model` is sent in the payload. Not used by the server but can be useful for debugging/routing.
-            if not model.startswith("http") and model.count("/") == 1:
-                # If it's a ID on the Hub => use it
-                model_id = model
-            else:
-                # Otherwise, we use a random string
-                model_id = "tgi"
-
-            try:
-                data = self.post(
-                    model=model_url,
-                    json=dict(
-                        model=model_id,
-                        messages=messages,
-                        frequency_penalty=frequency_penalty,
-                        logit_bias=logit_bias,
-                        logprobs=logprobs,
-                        max_tokens=max_tokens,
-                        n=n,
-                        presence_penalty=presence_penalty,
-                        response_format=response_format,
-                        seed=seed,
-                        stop=stop,
-                        temperature=temperature,
-                        tool_choice=tool_choice,
-                        tool_prompt=tool_prompt,
-                        tools=tools,
-                        top_logprobs=top_logprobs,
-                        top_p=top_p,
-                        stream=stream,
-                    ),
-                    stream=stream,
-                )
-            except HTTPError as e:
-                if e.response.status_code in (400, 404, 500):
-                    # Let's consider the server is not a chat completion server.
-                    # Then we call again `chat_completion` which will render the chat template client side.
-                    # (can be HTTP 500, HTTP 400, HTTP 404 depending on the server)
-                    _set_as_non_chat_completion_server(model)
-                    logger.warning(
-                        f"Server {model_url} does not seem to support chat completion. Falling back to text generation. Error: {e}"
-                    )
-                    return self.chat_completion(
-                        messages=messages,
-                        model=model,
-                        stream=stream,
-                        max_tokens=max_tokens,
-                        seed=seed,
-                        stop=stop,
-                        temperature=temperature,
-                        top_p=top_p,
-                    )
-                raise
-
-            if stream:
-                return _stream_chat_completion_response_from_bytes(data)  # type: ignore[arg-type]
-
-            return ChatCompletionOutput.parse_obj_as_instance(data)  # type: ignore[arg-type]
-
-        # At this point, we know the server is not a chat completion server.
-        # It means it's a transformers-backed server for which we can send a list of messages directly to the
-        # `text-generation` pipeline. We won't receive a detailed response but only the generated text.
         if stream:
-            raise ValueError(
-                "Streaming token is not supported by the model. This is due to the model not been served by a "
-                "Text-Generation-Inference server. Please pass `stream=False` as input."
-            )
-        if tool_choice is not None or tool_prompt is not None or tools is not None:
-            warnings.warn(
-                "Tools are not supported by the model. This is due to the model not been served by a "
-                "Text-Generation-Inference server. The provided tool parameters will be ignored."
-            )
-        if response_format is not None:
-            warnings.warn(
-                "Response format is not supported by the model. This is due to the model not been served by a "
-                "Text-Generation-Inference server. The provided response format will be ignored."
-            )
+            return _stream_chat_completion_response(data)  # type: ignore[arg-type]
 
-        # generate response
-        text_generation_output = self.text_generation(
-            prompt=messages,  # type: ignore # Not correct type but works implicitly
-            model=model,
-            stream=False,
-            details=False,
-            max_new_tokens=max_tokens,
-            seed=seed,
-            stop_sequences=stop,
-            temperature=temperature,
-            top_p=top_p,
-        )
-
-        # Format as a ChatCompletionOutput with dummy values for fields we can't provide
-        return ChatCompletionOutput(
-            id="dummy",
-            model="dummy",
-            system_fingerprint="dummy",
-            usage=None,  # type: ignore # set to `None` as we don't want to provide false information
-            created=int(time.time()),
-            choices=[
-                ChatCompletionOutputComplete(
-                    finish_reason="unk",  # type: ignore # set to `unk` as we don't want to provide false information
-                    index=0,
-                    message=ChatCompletionOutputMessage(
-                        content=text_generation_output,
-                        role="assistant",
-                    ),
-                )
-            ],
-        )
+        return ChatCompletionOutput.parse_obj_as_instance(data)  # type: ignore[arg-type]
 
     def conversational(
         self,
@@ -2251,7 +2176,12 @@ class InferenceClient:
         if stream:
             return _stream_text_generation_response(bytes_output, details)  # type: ignore
 
-        data = _bytes_to_dict(bytes_output)[0]  # type: ignore[arg-type]
+        data = _bytes_to_dict(bytes_output)  # type: ignore[arg-type]
+
+        # Data can be a single element (dict) or an iterable of dicts where we select the first element of.
+        if isinstance(data, list):
+            data = data[0]
+
         return TextGenerationOutput.parse_obj_as_instance(data) if details else data["generated_text"]
 
     def text_to_image(

{huggingface_hub-0.24.0rc0 → huggingface_hub-0.24.2}/src/huggingface_hub/inference/_common.py

@@ -34,7 +34,6 @@ from typing import (
     Literal,
     NoReturn,
     Optional,
-    Set,
     Union,
     overload,
 )
@@ -61,8 +60,6 @@ from ..utils import (
 )
 from ._generated.types import (
     ChatCompletionStreamOutput,
-    ChatCompletionStreamOutputChoice,
-    ChatCompletionStreamOutputDelta,
     TextGenerationStreamOutput,
 )
 
@@ -271,7 +268,10 @@ def _stream_text_generation_response(
     """Used in `InferenceClient.text_generation`."""
     # Parse ServerSentEvents
     for byte_payload in bytes_output_as_lines:
-        output = _format_text_generation_stream_output(byte_payload, details)
+        try:
+            output = _format_text_generation_stream_output(byte_payload, details)
+        except StopIteration:
+            break
         if output is not None:
             yield output
 
@@ -282,7 +282,10 @@ async def _async_stream_text_generation_response(
     """Used in `AsyncInferenceClient.text_generation`."""
     # Parse ServerSentEvents
     async for byte_payload in bytes_output_as_lines:
-        output = _format_text_generation_stream_output(byte_payload, details)
+        try:
+            output = _format_text_generation_stream_output(byte_payload, details)
+        except StopIteration:
+            break
         if output is not None:
             yield output
 
@@ -293,6 +296,9 @@ def _format_text_generation_stream_output(
     if not byte_payload.startswith(b"data:"):
         return None  # empty line
 
+    if byte_payload == b"data: [DONE]":
+        raise StopIteration("[DONE] signal received.")
+
     # Decode payload
     payload = byte_payload.decode("utf-8")
     json_payload = json.loads(payload.lstrip("data:").rstrip("/n"))
@@ -306,72 +312,41 @@ def _format_text_generation_stream_output(
     return output.token.text if not details else output
 
 
-def _format_chat_completion_stream_output_from_text_generation(
-    item: TextGenerationStreamOutput, created: int
-) -> ChatCompletionStreamOutput:
-    if item.details is None:
-        # new token generated => return delta
-        return ChatCompletionStreamOutput(
-            # explicitly set 'dummy' values to reduce expectations from users
-            id="dummy",
-            model="dummy",
-            system_fingerprint="dummy",
-            choices=[
-                ChatCompletionStreamOutputChoice(
-                    delta=ChatCompletionStreamOutputDelta(
-                        role="assistant",
-                        content=item.token.text,
-                    ),
-                    finish_reason=None,
-                    index=0,
-                )
-            ],
-            created=created,
-        )
-    else:
-        # generation is completed => return finish reason
-        return ChatCompletionStreamOutput(
-            # explicitly set 'dummy' values to reduce expectations from users
-            id="dummy",
-            model="dummy",
-            system_fingerprint="dummy",
-            choices=[
-                ChatCompletionStreamOutputChoice(
-                    delta=ChatCompletionStreamOutputDelta(role="assistant"),
-                    finish_reason=item.details.finish_reason,
-                    index=0,
-                )
-            ],
-            created=created,
-        )
-
-
-def _stream_chat_completion_response_from_bytes(
+def _stream_chat_completion_response(
     bytes_lines: Iterable[bytes],
 ) -> Iterable[ChatCompletionStreamOutput]:
     """Used in `InferenceClient.chat_completion` if model is served with TGI."""
     for item in bytes_lines:
-        output = _format_chat_completion_stream_output_from_text_generation_from_bytes(item)
+        try:
+            output = _format_chat_completion_stream_output(item)
+        except StopIteration:
+            break
         if output is not None:
             yield output
 
 
-async def _async_stream_chat_completion_response_from_bytes(
+async def _async_stream_chat_completion_response(
     bytes_lines: AsyncIterable[bytes],
 ) -> AsyncIterable[ChatCompletionStreamOutput]:
     """Used in `AsyncInferenceClient.chat_completion`."""
     async for item in bytes_lines:
-        output = _format_chat_completion_stream_output_from_text_generation_from_bytes(item)
+        try:
+            output = _format_chat_completion_stream_output(item)
+        except StopIteration:
+            break
         if output is not None:
             yield output
 
 
-def _format_chat_completion_stream_output_from_text_generation_from_bytes(
+def _format_chat_completion_stream_output(
     byte_payload: bytes,
 ) -> Optional[ChatCompletionStreamOutput]:
     if not byte_payload.startswith(b"data:"):
         return None  # empty line
 
+    if byte_payload == b"data: [DONE]":
+        raise StopIteration("[DONE] signal received.")
+
     # Decode payload
     payload = byte_payload.decode("utf-8")
     json_payload = json.loads(payload.lstrip("data:").rstrip("/n"))
@@ -413,17 +388,6 @@ def _get_unsupported_text_generation_kwargs(model: Optional[str]) -> List[str]:
     return _UNSUPPORTED_TEXT_GENERATION_KWARGS.get(model, [])
 
 
-_NON_CHAT_COMPLETION_SERVER: Set[str] = set()
-
-
-def _set_as_non_chat_completion_server(model: str) -> None:
-    _NON_CHAT_COMPLETION_SERVER.add(model)
-
-
-def _is_chat_completion_server(model: str) -> bool:
-    return model not in _NON_CHAT_COMPLETION_SERVER
-
-
 # TEXT GENERATION ERRORS
 # ----------------------
 # Text-generation errors are parsed separately to handle as much as possible the errors returned by the text generation

{huggingface_hub-0.24.0rc0 → huggingface_hub-0.24.2}/src/huggingface_hub/inference/_generated/_async_client.py

@@ -44,7 +44,7 @@ from huggingface_hub.inference._common import (
     TASKS_EXPECTING_IMAGES,
     ContentT,
     ModelStatus,
-    _async_stream_chat_completion_response_from_bytes,
+    _async_stream_chat_completion_response,
     _async_stream_text_generation_response,
     _b64_encode,
     _b64_to_image,
@@ -54,9 +54,7 @@ from huggingface_hub.inference._common import (
     _fetch_recommended_models,
     _get_unsupported_text_generation_kwargs,
     _import_numpy,
-    _is_chat_completion_server,
     _open_as_binary,
-    _set_as_non_chat_completion_server,
     _set_unsupported_text_generation_kwargs,
     raise_text_generation_error,
 )
@@ -68,8 +66,6 @@ from huggingface_hub.inference._generated.types import (
     ChatCompletionInputTool,
     ChatCompletionInputToolTypeClass,
     ChatCompletionOutput,
-    ChatCompletionOutputComplete,
-    ChatCompletionOutputMessage,
     ChatCompletionStreamOutput,
     DocumentQuestionAnsweringOutputElement,
     FillMaskOutputElement,
@@ -174,7 +170,7 @@ class AsyncInferenceClient:
             )
 
         self.model: Optional[str] = model
-        self.token: Union[str, bool, None] = token or api_key
+        self.token: Union[str, bool, None] = token if token is not None else api_key
         self.headers = CaseInsensitiveDict(build_hf_headers(token=self.token))  # 'authorization' + 'user-agent'
         if headers is not None:
             self.headers.update(headers)
@@ -824,123 +820,52 @@ class AsyncInferenceClient:
         # since `chat_completion(..., model=xxx)` is also a payload parameter for the
         # server, we need to handle it differently
         model = self.base_url or self.model or model or self.get_recommended_model("text-generation")
+        is_url = model.startswith(("http://", "https://"))
+
+        # First, resolve the model chat completions URL
+        if model == self.base_url:
+            # base_url passed => add server route
+            model_url = model + "/v1/chat/completions"
+        elif is_url:
+            # model is a URL => use it directly
+            model_url = model
+        else:
+            # model is a model ID => resolve it + add server route
+            model_url = self._resolve_url(model) + "/v1/chat/completions"
+
+        # `model` is sent in the payload. Not used by the server but can be useful for debugging/routing.
+        # If it's a ID on the Hub => use it. Otherwise, we use a random string.
+        model_id = model if not is_url and model.count("/") == 1 else "tgi"
+
+        data = await self.post(
+            model=model_url,
+            json=dict(
+                model=model_id,
+                messages=messages,
+                frequency_penalty=frequency_penalty,
+                logit_bias=logit_bias,
+                logprobs=logprobs,
+                max_tokens=max_tokens,
+                n=n,
+                presence_penalty=presence_penalty,
+                response_format=response_format,
+                seed=seed,
+                stop=stop,
+                temperature=temperature,
+                tool_choice=tool_choice,
+                tool_prompt=tool_prompt,
+                tools=tools,
+                top_logprobs=top_logprobs,
+                top_p=top_p,
+                stream=stream,
+            ),
+            stream=stream,
+        )
 
-        if _is_chat_completion_server(model):
-            # First, let's consider the server has a `/v1/chat/completions` endpoint.
-            # If that's the case, we don't have to render the chat template client-side.
-            model_url = self._resolve_url(model)
-            if not model_url.endswith("/chat/completions"):
-                model_url += "/v1/chat/completions"
-
-            # `model` is sent in the payload. Not used by the server but can be useful for debugging/routing.
-            if not model.startswith("http") and model.count("/") == 1:
-                # If it's a ID on the Hub => use it
-                model_id = model
-            else:
-                # Otherwise, we use a random string
-                model_id = "tgi"
-
-            try:
-                data = await self.post(
-                    model=model_url,
-                    json=dict(
-                        model=model_id,
-                        messages=messages,
-                        frequency_penalty=frequency_penalty,
-                        logit_bias=logit_bias,
-                        logprobs=logprobs,
-                        max_tokens=max_tokens,
-                        n=n,
-                        presence_penalty=presence_penalty,
-                        response_format=response_format,
-                        seed=seed,
-                        stop=stop,
-                        temperature=temperature,
-                        tool_choice=tool_choice,
-                        tool_prompt=tool_prompt,
-                        tools=tools,
-                        top_logprobs=top_logprobs,
-                        top_p=top_p,
-                        stream=stream,
-                    ),
-                    stream=stream,
-                )
-            except _import_aiohttp().ClientResponseError as e:
-                if e.status in (400, 404, 500):
-                    # Let's consider the server is not a chat completion server.
-                    # Then we call again `chat_completion` which will render the chat template client side.
-                    # (can be HTTP 500, HTTP 400, HTTP 404 depending on the server)
-                    _set_as_non_chat_completion_server(model)
-                    logger.warning(
-                        f"Server {model_url} does not seem to support chat completion. Falling back to text generation. Error: {e}"
-                    )
-                    return await self.chat_completion(
-                        messages=messages,
-                        model=model,
-                        stream=stream,
-                        max_tokens=max_tokens,
-                        seed=seed,
-                        stop=stop,
-                        temperature=temperature,
-                        top_p=top_p,
-                    )
-                raise
-
-            if stream:
-                return _async_stream_chat_completion_response_from_bytes(data)  # type: ignore[arg-type]
-
-            return ChatCompletionOutput.parse_obj_as_instance(data)  # type: ignore[arg-type]
-
-        # At this point, we know the server is not a chat completion server.
-        # It means it's a transformers-backed server for which we can send a list of messages directly to the
-        # `text-generation` pipeline. We won't receive a detailed response but only the generated text.
         if stream:
-            raise ValueError(
-                "Streaming token is not supported by the model. This is due to the model not been served by a "
-                "Text-Generation-Inference server. Please pass `stream=False` as input."
-            )
-        if tool_choice is not None or tool_prompt is not None or tools is not None:
-            warnings.warn(
-                "Tools are not supported by the model. This is due to the model not been served by a "
-                "Text-Generation-Inference server. The provided tool parameters will be ignored."
-            )
-        if response_format is not None:
-            warnings.warn(
-                "Response format is not supported by the model. This is due to the model not been served by a "
-                "Text-Generation-Inference server. The provided response format will be ignored."
-            )
-
-        # generate response
-        text_generation_output = await self.text_generation(
-            prompt=messages,  # type: ignore # Not correct type but works implicitly
-            model=model,
-            stream=False,
-            details=False,
-            max_new_tokens=max_tokens,
-            seed=seed,
-            stop_sequences=stop,
-            temperature=temperature,
-            top_p=top_p,
-        )
+            return _async_stream_chat_completion_response(data)  # type: ignore[arg-type]
 
-        # Format as a ChatCompletionOutput with dummy values for fields we can't provide
-        return ChatCompletionOutput(
-            id="dummy",
-            model="dummy",
-            system_fingerprint="dummy",
-            usage=None,  # type: ignore # set to `None` as we don't want to provide false information
-            created=int(time.time()),
-            choices=[
-                ChatCompletionOutputComplete(
-                    finish_reason="unk",  # type: ignore # set to `unk` as we don't want to provide false information
-                    index=0,
-                    message=ChatCompletionOutputMessage(
-                        content=text_generation_output,
-                        role="assistant",
-                    ),
-                )
-            ],
-        )
+        return ChatCompletionOutput.parse_obj_as_instance(data)  # type: ignore[arg-type]
 
     async def conversational(
         self,
@@ -2282,7 +2207,12 @@ class AsyncInferenceClient:
         if stream:
             return _async_stream_text_generation_response(bytes_output, details)  # type: ignore
 
-        data = _bytes_to_dict(bytes_output)[0]  # type: ignore[arg-type]
+        data = _bytes_to_dict(bytes_output)  # type: ignore[arg-type]
+
+        # Data can be a single element (dict) or an iterable of dicts where we select the first element of.
+        if isinstance(data, list):
+            data = data[0]
+
         return TextGenerationOutput.parse_obj_as_instance(data) if details else data["generated_text"]
 
     async def text_to_image(

{huggingface_hub-0.24.0rc0 → huggingface_hub-0.24.2}/src/huggingface_hub.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: huggingface-hub
-Version: 0.24.0rc0
+Version: 0.24.2
 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.