huggingface-hub 0.30.1__py3-none-any.whl → 0.31.0rc0__py3-none-any.whl

huggingface_hub/inference/_generated/types/automatic_speech_recognition.py

@@ -75,11 +75,10 @@ class AutomaticSpeechRecognitionGenerationParameters(BaseInferenceType):
 class AutomaticSpeechRecognitionParameters(BaseInferenceType):
     """Additional inference parameters for Automatic Speech Recognition"""
 
+    generation_parameters: Optional[AutomaticSpeechRecognitionGenerationParameters] = None
+    """Parametrization of the text generation process"""
     return_timestamps: Optional[bool] = None
     """Whether to output corresponding timestamps with the generated text"""
-    # Will be deprecated in the future when the renaming to `generation_parameters` is implemented in transformers
-    generate_kwargs: Optional[AutomaticSpeechRecognitionGenerationParameters] = None
-    """Parametrization of the text generation process"""
 
 
 @dataclass_with_extra

huggingface_hub/inference/_generated/types/chat_completion.py

@@ -25,8 +25,8 @@ class ChatCompletionInputMessageChunk(BaseInferenceType):
 
 @dataclass_with_extra
 class ChatCompletionInputFunctionDefinition(BaseInferenceType):
-    arguments: Any
     name: str
+    parameters: Any
     description: Optional[str] = None
 
 
@@ -45,7 +45,7 @@ class ChatCompletionInputMessage(BaseInferenceType):
     tool_calls: Optional[List[ChatCompletionInputToolCall]] = None
 
 
-ChatCompletionInputGrammarTypeType = Literal["json", "regex"]
+ChatCompletionInputGrammarTypeType = Literal["json", "regex", "json_schema"]
 
 
 @dataclass_with_extra
@@ -189,7 +189,7 @@ class ChatCompletionOutputLogprobs(BaseInferenceType):
 
 @dataclass_with_extra
 class ChatCompletionOutputFunctionDefinition(BaseInferenceType):
-    arguments: Any
+    arguments: str
     name: str
     description: Optional[str] = None
 

huggingface_hub/inference/_generated/types/image_to_text.py

@@ -75,11 +75,10 @@ class ImageToTextGenerationParameters(BaseInferenceType):
 class ImageToTextParameters(BaseInferenceType):
     """Additional inference parameters for Image To Text"""
 
+    generation_parameters: Optional[ImageToTextGenerationParameters] = None
+    """Parametrization of the text generation process"""
     max_new_tokens: Optional[int] = None
     """The amount of maximum tokens to generate."""
-    # Will be deprecated in the future when the renaming to `generation_parameters` is implemented in transformers
-    generate_kwargs: Optional[ImageToTextGenerationParameters] = None
-    """Parametrization of the text generation process"""
 
 
 @dataclass_with_extra

huggingface_hub/inference/_generated/types/text_generation.py

@@ -8,7 +8,7 @@ from typing import Any, List, Literal, Optional
 from .base import BaseInferenceType, dataclass_with_extra
 
 
-TypeEnum = Literal["json", "regex"]
+TypeEnum = Literal["json", "regex", "json_schema"]
 
 
 @dataclass_with_extra

huggingface_hub/inference/_generated/types/text_to_audio.py

@@ -75,8 +75,7 @@ class TextToAudioGenerationParameters(BaseInferenceType):
 class TextToAudioParameters(BaseInferenceType):
     """Additional inference parameters for Text To Audio"""
 
-    # Will be deprecated in the future when the renaming to `generation_parameters` is implemented in transformers
-    generate_kwargs: Optional[TextToAudioGenerationParameters] = None
+    generation_parameters: Optional[TextToAudioGenerationParameters] = None
     """Parametrization of the text generation process"""
 
 

huggingface_hub/inference/_generated/types/text_to_speech.py

@@ -75,8 +75,7 @@ class TextToSpeechGenerationParameters(BaseInferenceType):
 class TextToSpeechParameters(BaseInferenceType):
     """Additional inference parameters for Text To Speech"""
 
-    # Will be deprecated in the future when the renaming to `generation_parameters` is implemented in transformers
-    generate_kwargs: Optional[TextToSpeechGenerationParameters] = None
+    generation_parameters: Optional[TextToSpeechGenerationParameters] = None
     """Parametrization of the text generation process"""
 
 

huggingface_hub/inference/_providers/__init__.py

@@ -1,6 +1,8 @@
-from typing import Dict, Literal
+from typing import Dict, Literal, Optional, Union
 
-from ._common import TaskProviderHelper
+from huggingface_hub.utils import logging
+
+from ._common import TaskProviderHelper, _fetch_inference_provider_mapping
 from .black_forest_labs import BlackForestLabsTextToImageTask
 from .cerebras import CerebrasConversationalTask
 from .cohere import CohereConversationalTask
@@ -11,16 +13,24 @@ from .fal_ai import (
     FalAITextToVideoTask,
 )
 from .fireworks_ai import FireworksAIConversationalTask
-from .hf_inference import HFInferenceBinaryInputTask, HFInferenceConversational, HFInferenceTask
+from .hf_inference import (
+    HFInferenceBinaryInputTask,
+    HFInferenceConversational,
+    HFInferenceFeatureExtractionTask,
+    HFInferenceTask,
+)
 from .hyperbolic import HyperbolicTextGenerationTask, HyperbolicTextToImageTask
 from .nebius import NebiusConversationalTask, NebiusTextGenerationTask, NebiusTextToImageTask
 from .novita import NovitaConversationalTask, NovitaTextGenerationTask, NovitaTextToVideoTask
 from .openai import OpenAIConversationalTask
-from .replicate import ReplicateTask, ReplicateTextToSpeechTask
-from .sambanova import SambanovaConversationalTask
+from .replicate import ReplicateTask, ReplicateTextToImageTask, ReplicateTextToSpeechTask
+from .sambanova import SambanovaConversationalTask, SambanovaFeatureExtractionTask
 from .together import TogetherConversationalTask, TogetherTextGenerationTask, TogetherTextToImageTask
 
 
+logger = logging.get_logger(__name__)
+
+
 PROVIDER_T = Literal[
     "black-forest-labs",
     "cerebras",
@@ -37,6 +47,8 @@ PROVIDER_T = Literal[
     "together",
 ]
 
+PROVIDER_OR_POLICY_T = Union[PROVIDER_T, Literal["auto"]]
+
 PROVIDERS: Dict[PROVIDER_T, Dict[str, TaskProviderHelper]] = {
     "black-forest-labs": {
         "text-to-image": BlackForestLabsTextToImageTask(),
@@ -65,7 +77,7 @@ PROVIDERS: Dict[PROVIDER_T, Dict[str, TaskProviderHelper]] = {
         "audio-classification": HFInferenceBinaryInputTask("audio-classification"),
         "automatic-speech-recognition": HFInferenceBinaryInputTask("automatic-speech-recognition"),
         "fill-mask": HFInferenceTask("fill-mask"),
-        "feature-extraction": HFInferenceTask("feature-extraction"),
+        "feature-extraction": HFInferenceFeatureExtractionTask(),
         "image-classification": HFInferenceBinaryInputTask("image-classification"),
         "image-segmentation": HFInferenceBinaryInputTask("image-segmentation"),
         "document-question-answering": HFInferenceTask("document-question-answering"),
@@ -103,12 +115,13 @@ PROVIDERS: Dict[PROVIDER_T, Dict[str, TaskProviderHelper]] = {
         "conversational": OpenAIConversationalTask(),
     },
     "replicate": {
-        "text-to-image": ReplicateTask("text-to-image"),
+        "text-to-image": ReplicateTextToImageTask(),
         "text-to-speech": ReplicateTextToSpeechTask(),
         "text-to-video": ReplicateTask("text-to-video"),
     },
     "sambanova": {
         "conversational": SambanovaConversationalTask(),
+        "feature-extraction": SambanovaFeatureExtractionTask(),
     },
     "together": {
         "text-to-image": TogetherTextToImageTask(),
@@ -118,24 +131,49 @@ PROVIDERS: Dict[PROVIDER_T, Dict[str, TaskProviderHelper]] = {
 }
 
 
-def get_provider_helper(provider: PROVIDER_T, task: str) -> TaskProviderHelper:
+def get_provider_helper(
+    provider: Optional[PROVIDER_OR_POLICY_T], task: str, model: Optional[str]
+) -> TaskProviderHelper:
     """Get provider helper instance by name and task.
 
     Args:
-        provider (str): Name of the provider
-        task (str): Name of the task
-
+        provider (`str`, *optional*): name of the provider, or "auto" to automatically select the provider for the model.
+        task (`str`): Name of the task
+        model (`str`, *optional*): Name of the model
     Returns:
         TaskProviderHelper: Helper instance for the specified provider and task
 
     Raises:
         ValueError: If provider or task is not supported
     """
-    if provider not in PROVIDERS:
-        raise ValueError(f"Provider '{provider}' not supported. Available providers: {list(PROVIDERS.keys())}")
-    if task not in PROVIDERS[provider]:
+
+    if (model is None and provider in (None, "auto")) or (
+        model is not None and model.startswith(("http://", "https://"))
+    ):
+        provider = "hf-inference"
+
+    if provider is None:
+        logger.info(
+            "Defaulting to 'auto' which will select the first provider available for the model, sorted by the user's order in https://hf.co/settings/inference-providers."
+        )
+        provider = "auto"
+
+    if provider == "auto":
+        if model is None:
+            raise ValueError("Specifying a model is required when provider is 'auto'")
+        provider_mapping = _fetch_inference_provider_mapping(model)
+        provider = next(iter(provider_mapping))
+
+    provider_tasks = PROVIDERS.get(provider)  # type: ignore
+    if provider_tasks is None:
+        raise ValueError(
+            f"Provider '{provider}' not supported. Available values: 'auto' or any provider from {list(PROVIDERS.keys())}."
+            "Passing 'auto' (default value) will automatically select the first provider available for the model, sorted "
+            "by the user's order in https://hf.co/settings/inference-providers."
+        )
+
+    if task not in provider_tasks:
         raise ValueError(
-            f"Task '{task}' not supported for provider '{provider}'. "
-            f"Available tasks: {list(PROVIDERS[provider].keys())}"
+            f"Task '{task}' not supported for provider '{provider}'. Available tasks: {list(provider_tasks.keys())}"
         )
-    return PROVIDERS[provider][task]
+    return provider_tasks[task]

huggingface_hub/inference/_providers/_common.py

@@ -2,21 +2,24 @@ from functools import lru_cache
 from typing import Any, Dict, Optional, Union
 
 from huggingface_hub import constants
+from huggingface_hub.hf_api import InferenceProviderMapping
 from huggingface_hub.inference._common import RequestParameters
 from huggingface_hub.utils import build_hf_headers, get_token, logging
 
 
 logger = logging.get_logger(__name__)
 
-
 # Dev purposes only.
 # If you want to try to run inference for a new model locally before it's registered on huggingface.co
 # for a given Inference Provider, you can add it to the following dictionary.
-HARDCODED_MODEL_ID_MAPPING: Dict[str, Dict[str, str]] = {
-    # "HF model ID" => "Model ID on Inference Provider's side"
+HARDCODED_MODEL_INFERENCE_MAPPING: Dict[str, Dict[str, InferenceProviderMapping]] = {
+    # "HF model ID" => InferenceProviderMapping object initialized with "Model ID on Inference Provider's side"
     #
     # Example:
-    # "Qwen/Qwen2.5-Coder-32B-Instruct": "Qwen2.5-Coder-32B-Instruct",
+    # "Qwen/Qwen2.5-Coder-32B-Instruct": InferenceProviderMapping(hf_model_id="Qwen/Qwen2.5-Coder-32B-Instruct",
+    #                                    provider_id="Qwen2.5-Coder-32B-Instruct",
+    #                                    task="conversational",
+    #                                    status="live")
     "cerebras": {},
     "cohere": {},
     "fal-ai": {},
@@ -61,28 +64,30 @@ class TaskProviderHelper:
         api_key = self._prepare_api_key(api_key)
 
         # mapped model from HF model ID
-        mapped_model = self._prepare_mapped_model(model)
+        provider_mapping_info = self._prepare_mapping_info(model)
 
         # default HF headers + user headers (to customize in subclasses)
         headers = self._prepare_headers(headers, api_key)
 
         # routed URL if HF token, or direct URL (to customize in '_prepare_route' in subclasses)
-        url = self._prepare_url(api_key, mapped_model)
+        url = self._prepare_url(api_key, provider_mapping_info.provider_id)
 
         # prepare payload (to customize in subclasses)
-        payload = self._prepare_payload_as_dict(inputs, parameters, mapped_model=mapped_model)
+        payload = self._prepare_payload_as_dict(inputs, parameters, provider_mapping_info=provider_mapping_info)
         if payload is not None:
             payload = recursive_merge(payload, extra_payload or {})
 
         # body data (to customize in subclasses)
-        data = self._prepare_payload_as_bytes(inputs, parameters, mapped_model, extra_payload)
+        data = self._prepare_payload_as_bytes(inputs, parameters, provider_mapping_info, extra_payload)
 
         # check if both payload and data are set and return
         if payload is not None and data is not None:
             raise ValueError("Both payload and data cannot be set in the same request.")
         if payload is None and data is None:
             raise ValueError("Either payload or data must be set in the request.")
-        return RequestParameters(url=url, task=self.task, model=mapped_model, json=payload, data=data, headers=headers)
+        return RequestParameters(
+            url=url, task=self.task, model=provider_mapping_info.provider_id, json=payload, data=data, headers=headers
+        )
 
     def get_response(
         self,
@@ -107,7 +112,7 @@ class TaskProviderHelper:
             )
         return api_key
 
-    def _prepare_mapped_model(self, model: Optional[str]) -> str:
+    def _prepare_mapping_info(self, model: Optional[str]) -> InferenceProviderMapping:
         """Return the mapped model ID to use for the request.
 
         Usually not overwritten in subclasses."""
@@ -115,8 +120,8 @@ class TaskProviderHelper:
             raise ValueError(f"Please provide an HF model ID supported by {self.provider}.")
 
         # hardcoded mapping for local testing
-        if HARDCODED_MODEL_ID_MAPPING.get(self.provider, {}).get(model):
-            return HARDCODED_MODEL_ID_MAPPING[self.provider][model]
+        if HARDCODED_MODEL_INFERENCE_MAPPING.get(self.provider, {}).get(model):
+            return HARDCODED_MODEL_INFERENCE_MAPPING[self.provider][model]
 
         provider_mapping = _fetch_inference_provider_mapping(model).get(self.provider)
         if provider_mapping is None:
@@ -132,7 +137,7 @@ class TaskProviderHelper:
             logger.warning(
                 f"Model {model} is in staging mode for provider {self.provider}. Meant for test purposes only."
             )
-        return provider_mapping.provider_id
+        return provider_mapping
 
     def _prepare_headers(self, headers: Dict, api_key: str) -> Dict:
         """Return the headers to use for the request.
@@ -168,7 +173,9 @@ class TaskProviderHelper:
         """
         return ""
 
-    def _prepare_payload_as_dict(self, inputs: Any, parameters: Dict, mapped_model: str) -> Optional[Dict]:
+    def _prepare_payload_as_dict(
+        self, inputs: Any, parameters: Dict, provider_mapping_info: InferenceProviderMapping
+    ) -> Optional[Dict]:
         """Return the payload to use for the request, as a dict.
 
         Override this method in subclasses for customized payloads.
@@ -177,7 +184,11 @@ class TaskProviderHelper:
         return None
 
     def _prepare_payload_as_bytes(
-        self, inputs: Any, parameters: Dict, mapped_model: str, extra_payload: Optional[Dict]
+        self,
+        inputs: Any,
+        parameters: Dict,
+        provider_mapping_info: InferenceProviderMapping,
+        extra_payload: Optional[Dict],
     ) -> Optional[bytes]:
         """Return the body to use for the request, as bytes.
 
@@ -199,8 +210,10 @@ class BaseConversationalTask(TaskProviderHelper):
     def _prepare_route(self, mapped_model: str, api_key: str) -> str:
         return "/v1/chat/completions"
 
-    def _prepare_payload_as_dict(self, inputs: Any, parameters: Dict, mapped_model: str) -> Optional[Dict]:
-        return {"messages": inputs, **filter_none(parameters), "model": mapped_model}
+    def _prepare_payload_as_dict(
+        self, inputs: Any, parameters: Dict, provider_mapping_info: InferenceProviderMapping
+    ) -> Optional[Dict]:
+        return {"messages": inputs, **filter_none(parameters), "model": provider_mapping_info.provider_id}
 
 
 class BaseTextGenerationTask(TaskProviderHelper):
@@ -215,8 +228,10 @@ class BaseTextGenerationTask(TaskProviderHelper):
     def _prepare_route(self, mapped_model: str, api_key: str) -> str:
         return "/v1/completions"
 
-    def _prepare_payload_as_dict(self, inputs: Any, parameters: Dict, mapped_model: str) -> Optional[Dict]:
-        return {"prompt": inputs, **filter_none(parameters), "model": mapped_model}
+    def _prepare_payload_as_dict(
+        self, inputs: Any, parameters: Dict, provider_mapping_info: InferenceProviderMapping
+    ) -> Optional[Dict]:
+        return {"prompt": inputs, **filter_none(parameters), "model": provider_mapping_info.provider_id}
 
 
 @lru_cache(maxsize=None)

huggingface_hub/inference/_providers/black_forest_labs.py

@@ -1,6 +1,7 @@
 import time
 from typing import Any, Dict, Optional, Union
 
+from huggingface_hub.hf_api import InferenceProviderMapping
 from huggingface_hub.inference._common import RequestParameters, _as_dict
 from huggingface_hub.inference._providers._common import TaskProviderHelper, filter_none
 from huggingface_hub.utils import logging
@@ -27,7 +28,9 @@ class BlackForestLabsTextToImageTask(TaskProviderHelper):
     def _prepare_route(self, mapped_model: str, api_key: str) -> str:
         return f"/v1/{mapped_model}"
 
-    def _prepare_payload_as_dict(self, inputs: Any, parameters: Dict, mapped_model: str) -> Optional[Dict]:
+    def _prepare_payload_as_dict(
+        self, inputs: Any, parameters: Dict, provider_mapping_info: InferenceProviderMapping
+    ) -> Optional[Dict]:
         parameters = filter_none(parameters)
         if "num_inference_steps" in parameters:
             parameters["steps"] = parameters.pop("num_inference_steps")

huggingface_hub/inference/_providers/fal_ai.py

@@ -4,6 +4,8 @@ from abc import ABC
 from typing import Any, Dict, Optional, Union
 from urllib.parse import urlparse
 
+from huggingface_hub import constants
+from huggingface_hub.hf_api import InferenceProviderMapping
 from huggingface_hub.inference._common import RequestParameters, _as_dict
 from huggingface_hub.inference._providers._common import TaskProviderHelper, filter_none
 from huggingface_hub.utils import get_session, hf_raise_for_status
@@ -34,7 +36,9 @@ class FalAIAutomaticSpeechRecognitionTask(FalAITask):
     def __init__(self):
         super().__init__("automatic-speech-recognition")
 
-    def _prepare_payload_as_dict(self, inputs: Any, parameters: Dict, mapped_model: str) -> Optional[Dict]:
+    def _prepare_payload_as_dict(
+        self, inputs: Any, parameters: Dict, provider_mapping_info: InferenceProviderMapping
+    ) -> Optional[Dict]:
         if isinstance(inputs, str) and inputs.startswith(("http://", "https://")):
             # If input is a URL, pass it directly
             audio_url = inputs
@@ -61,14 +65,31 @@ class FalAITextToImageTask(FalAITask):
     def __init__(self):
         super().__init__("text-to-image")
 
-    def _prepare_payload_as_dict(self, inputs: Any, parameters: Dict, mapped_model: str) -> Optional[Dict]:
-        parameters = filter_none(parameters)
-        if "width" in parameters and "height" in parameters:
-            parameters["image_size"] = {
-                "width": parameters.pop("width"),
-                "height": parameters.pop("height"),
+    def _prepare_payload_as_dict(
+        self, inputs: Any, parameters: Dict, provider_mapping_info: InferenceProviderMapping
+    ) -> Optional[Dict]:
+        payload: Dict[str, Any] = {
+            "prompt": inputs,
+            **filter_none(parameters),
+        }
+        if "width" in payload and "height" in payload:
+            payload["image_size"] = {
+                "width": payload.pop("width"),
+                "height": payload.pop("height"),
             }
-        return {"prompt": inputs, **parameters}
+        if provider_mapping_info.adapter_weights_path is not None:
+            lora_path = constants.HUGGINGFACE_CO_URL_TEMPLATE.format(
+                repo_id=provider_mapping_info.hf_model_id,
+                revision="main",
+                filename=provider_mapping_info.adapter_weights_path,
+            )
+            payload["loras"] = [{"path": lora_path, "scale": 1}]
+            if provider_mapping_info.provider_id == "fal-ai/lora":
+                # little hack: fal requires the base model for stable-diffusion-based loras but not for flux-based
+                # See payloads in https://fal.ai/models/fal-ai/lora/api vs https://fal.ai/models/fal-ai/flux-lora/api
+                payload["model_name"] = "stabilityai/stable-diffusion-xl-base-1.0"
+
+        return payload
 
     def get_response(self, response: Union[bytes, Dict], request_params: Optional[RequestParameters] = None) -> Any:
         url = _as_dict(response)["images"][0]["url"]
@@ -79,8 +100,10 @@ class FalAITextToSpeechTask(FalAITask):
     def __init__(self):
         super().__init__("text-to-speech")
 
-    def _prepare_payload_as_dict(self, inputs: Any, parameters: Dict, mapped_model: str) -> Optional[Dict]:
-        return {"lyrics": inputs, **filter_none(parameters)}
+    def _prepare_payload_as_dict(
+        self, inputs: Any, parameters: Dict, provider_mapping_info: InferenceProviderMapping
+    ) -> Optional[Dict]:
+        return {"text": inputs, **filter_none(parameters)}
 
     def get_response(self, response: Union[bytes, Dict], request_params: Optional[RequestParameters] = None) -> Any:
         url = _as_dict(response)["audio"]["url"]
@@ -104,7 +127,9 @@ class FalAITextToVideoTask(FalAITask):
             return f"/{mapped_model}?_subdomain=queue"
         return f"/{mapped_model}"
 
-    def _prepare_payload_as_dict(self, inputs: Any, parameters: Dict, mapped_model: str) -> Optional[Dict]:
+    def _prepare_payload_as_dict(
+        self, inputs: Any, parameters: Dict, provider_mapping_info: InferenceProviderMapping
+    ) -> Optional[Dict]:
         return {"prompt": inputs, **filter_none(parameters)}
 
     def get_response(

huggingface_hub/inference/_providers/hf_inference.py

@@ -1,10 +1,11 @@
 import json
 from functools import lru_cache
 from pathlib import Path
-from typing import Any, Dict, Optional
+from typing import Any, Dict, Optional, Union
 
 from huggingface_hub import constants
-from huggingface_hub.inference._common import _b64_encode, _open_as_binary
+from huggingface_hub.hf_api import InferenceProviderMapping
+from huggingface_hub.inference._common import RequestParameters, _b64_encode, _bytes_to_dict, _open_as_binary
 from huggingface_hub.inference._providers._common import TaskProviderHelper, filter_none
 from huggingface_hub.utils import build_hf_headers, get_session, get_token, hf_raise_for_status
 
@@ -23,9 +24,9 @@ class HFInferenceTask(TaskProviderHelper):
         # special case: for HF Inference we allow not providing an API key
         return api_key or get_token()  # type: ignore[return-value]
 
-    def _prepare_mapped_model(self, model: Optional[str]) -> str:
+    def _prepare_mapping_info(self, model: Optional[str]) -> InferenceProviderMapping:
         if model is not None and model.startswith(("http://", "https://")):
-            return model
+            return InferenceProviderMapping(providerId=model, hf_model_id=model, task=self.task, status="live")
         model_id = model if model is not None else _fetch_recommended_models().get(self.task)
         if model_id is None:
             raise ValueError(
@@ -33,7 +34,7 @@ class HFInferenceTask(TaskProviderHelper):
                 " explicitly. Visit https://huggingface.co/tasks for more info."
             )
         _check_supported_task(model_id, self.task)
-        return model_id
+        return InferenceProviderMapping(providerId=model_id, hf_model_id=model_id, task=self.task, status="live")
 
     def _prepare_url(self, api_key: str, mapped_model: str) -> str:
         # hf-inference provider can handle URLs (e.g. Inference Endpoints or TGI deployment)
@@ -41,13 +42,15 @@ class HFInferenceTask(TaskProviderHelper):
             return mapped_model
         return (
             # Feature-extraction and sentence-similarity are the only cases where we handle models with several tasks.
-            f"{self.base_url}/pipeline/{self.task}/{mapped_model}"
+            f"{self.base_url}/models/{mapped_model}/pipeline/{self.task}"
             if self.task in ("feature-extraction", "sentence-similarity")
             # Otherwise, we use the default endpoint
             else f"{self.base_url}/models/{mapped_model}"
         )
 
-    def _prepare_payload_as_dict(self, inputs: Any, parameters: Dict, mapped_model: str) -> Optional[Dict]:
+    def _prepare_payload_as_dict(
+        self, inputs: Any, parameters: Dict, provider_mapping_info: InferenceProviderMapping
+    ) -> Optional[Dict]:
         if isinstance(inputs, bytes):
             raise ValueError(f"Unexpected binary input for task {self.task}.")
         if isinstance(inputs, Path):
@@ -56,11 +59,17 @@ class HFInferenceTask(TaskProviderHelper):
 
 
 class HFInferenceBinaryInputTask(HFInferenceTask):
-    def _prepare_payload_as_dict(self, inputs: Any, parameters: Dict, mapped_model: str) -> Optional[Dict]:
+    def _prepare_payload_as_dict(
+        self, inputs: Any, parameters: Dict, provider_mapping_info: InferenceProviderMapping
+    ) -> Optional[Dict]:
         return None
 
     def _prepare_payload_as_bytes(
-        self, inputs: Any, parameters: Dict, mapped_model: str, extra_payload: Optional[Dict]
+        self,
+        inputs: Any,
+        parameters: Dict,
+        provider_mapping_info: InferenceProviderMapping,
+        extra_payload: Optional[Dict],
     ) -> Optional[bytes]:
         parameters = filter_none({k: v for k, v in parameters.items() if v is not None})
         extra_payload = extra_payload or {}
@@ -82,9 +91,12 @@ class HFInferenceBinaryInputTask(HFInferenceTask):
 
 class HFInferenceConversational(HFInferenceTask):
     def __init__(self):
-        super().__init__("text-generation")
+        super().__init__("conversational")
 
-    def _prepare_payload_as_dict(self, inputs: Any, parameters: Dict, mapped_model: str) -> Optional[Dict]:
+    def _prepare_payload_as_dict(
+        self, inputs: Any, parameters: Dict, provider_mapping_info: InferenceProviderMapping
+    ) -> Optional[Dict]:
+        mapped_model = provider_mapping_info.provider_id
         payload_model = parameters.get("model") or mapped_model
 
         if payload_model is None or payload_model.startswith(("http://", "https://")):
@@ -165,3 +177,13 @@ def _check_supported_task(model: str, task: str) -> None:
             f"Model '{model}' doesn't support task '{task}'. Supported tasks: '{pipeline_tag}', got: '{task}'"
         )
     return
+
+
+class HFInferenceFeatureExtractionTask(HFInferenceTask):
+    def __init__(self):
+        super().__init__("feature-extraction")
+
+    def get_response(self, response: Union[bytes, Dict], request_params: Optional[RequestParameters] = None) -> Any:
+        if isinstance(response, bytes):
+            return _bytes_to_dict(response)
+        return response

huggingface_hub/inference/_providers/hyperbolic.py

@@ -1,6 +1,7 @@
 import base64
 from typing import Any, Dict, Optional, Union
 
+from huggingface_hub.hf_api import InferenceProviderMapping
 from huggingface_hub.inference._common import RequestParameters, _as_dict
 from huggingface_hub.inference._providers._common import BaseConversationalTask, TaskProviderHelper, filter_none
 
@@ -12,7 +13,10 @@ class HyperbolicTextToImageTask(TaskProviderHelper):
     def _prepare_route(self, mapped_model: str, api_key: str) -> str:
         return "/v1/images/generations"
 
-    def _prepare_payload_as_dict(self, inputs: Any, parameters: Dict, mapped_model: str) -> Optional[Dict]:
+    def _prepare_payload_as_dict(
+        self, inputs: Any, parameters: Dict, provider_mapping_info: InferenceProviderMapping
+    ) -> Optional[Dict]:
+        mapped_model = provider_mapping_info.provider_id
         parameters = filter_none(parameters)
         if "num_inference_steps" in parameters:
             parameters["steps"] = parameters.pop("num_inference_steps")

huggingface_hub/inference/_providers/nebius.py

@@ -1,6 +1,7 @@
 import base64
 from typing import Any, Dict, Optional, Union
 
+from huggingface_hub.hf_api import InferenceProviderMapping
 from huggingface_hub.inference._common import RequestParameters, _as_dict
 from huggingface_hub.inference._providers._common import (
     BaseConversationalTask,
@@ -14,6 +15,16 @@ class NebiusTextGenerationTask(BaseTextGenerationTask):
     def __init__(self):
         super().__init__(provider="nebius", base_url="https://api.studio.nebius.ai")
 
+    def get_response(self, response: Union[bytes, Dict], request_params: Optional[RequestParameters] = None) -> Any:
+        output = _as_dict(response)["choices"][0]
+        return {
+            "generated_text": output["text"],
+            "details": {
+                "finish_reason": output.get("finish_reason"),
+                "seed": output.get("seed"),
+            },
+        }
+
 
 class NebiusConversationalTask(BaseConversationalTask):
     def __init__(self):
@@ -27,7 +38,10 @@ class NebiusTextToImageTask(TaskProviderHelper):
     def _prepare_route(self, mapped_model: str, api_key: str) -> str:
         return "/v1/images/generations"
 
-    def _prepare_payload_as_dict(self, inputs: Any, parameters: Dict, mapped_model: str) -> Optional[Dict]:
+    def _prepare_payload_as_dict(
+        self, inputs: Any, parameters: Dict, provider_mapping_info: InferenceProviderMapping
+    ) -> Optional[Dict]:
+        mapped_model = provider_mapping_info.provider_id
         parameters = filter_none(parameters)
         if "guidance_scale" in parameters:
             parameters.pop("guidance_scale")

huggingface_hub/inference/_providers/novita.py

@@ -1,5 +1,6 @@
 from typing import Any, Dict, Optional, Union
 
+from huggingface_hub.hf_api import InferenceProviderMapping
 from huggingface_hub.inference._common import RequestParameters, _as_dict
 from huggingface_hub.inference._providers._common import (
     BaseConversationalTask,
@@ -22,6 +23,16 @@ class NovitaTextGenerationTask(BaseTextGenerationTask):
         # there is no v1/ route for novita
         return "/v3/openai/completions"
 
+    def get_response(self, response: Union[bytes, Dict], request_params: Optional[RequestParameters] = None) -> Any:
+        output = _as_dict(response)["choices"][0]
+        return {
+            "generated_text": output["text"],
+            "details": {
+                "finish_reason": output.get("finish_reason"),
+                "seed": output.get("seed"),
+            },
+        }
+
 
 class NovitaConversationalTask(BaseConversationalTask):
     def __init__(self):
@@ -39,7 +50,9 @@ class NovitaTextToVideoTask(TaskProviderHelper):
     def _prepare_route(self, mapped_model: str, api_key: str) -> str:
         return f"/v3/hf/{mapped_model}"
 
-    def _prepare_payload_as_dict(self, inputs: Any, parameters: Dict, mapped_model: str) -> Optional[Dict]:
+    def _prepare_payload_as_dict(
+        self, inputs: Any, parameters: Dict, provider_mapping_info: InferenceProviderMapping
+    ) -> Optional[Dict]:
         return {"prompt": inputs, **filter_none(parameters)}
 
     def get_response(self, response: Union[bytes, Dict], request_params: Optional[RequestParameters] = None) -> Any: