[Inference Providers] fold OpenAI support into provider parameter (#2949)

* add openai as a provider

* fix

Author: hanouticelina

src/huggingface_hub/inference/_client.py

@@ -133,7 +133,7 @@ class InferenceClient:
             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.
         provider (`str`, *optional*):
-            Name of the provider to use for inference. Can be `"black-forest-labs"`, `"cerebras"`, `"cohere"`, `"fal-ai"`, `"fireworks-ai"`, `"hf-inference"`, `"hyperbolic"`, `"nebius"`, `"novita"`, `"replicate"`, "sambanova"` or `"together"`.
+            Name of the provider to use for inference. Can be `"black-forest-labs"`, `"cerebras"`, `"cohere"`, `"fal-ai"`, `"fireworks-ai"`, `"hf-inference"`, `"hyperbolic"`, `"nebius"`, `"novita"`, `"openai"`, `"replicate"`, "sambanova"` or `"together"`.
             defaults to hf-inference (Hugging Face Serverless Inference API).
             If model is a URL or `base_url` is passed, then `provider` is not used.
         token (`str`, *optional*):

src/huggingface_hub/inference/_generated/_async_client.py

@@ -121,7 +121,7 @@ class AsyncInferenceClient:
             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.
         provider (`str`, *optional*):
-            Name of the provider to use for inference. Can be `"black-forest-labs"`, `"cerebras"`, `"cohere"`, `"fal-ai"`, `"fireworks-ai"`, `"hf-inference"`, `"hyperbolic"`, `"nebius"`, `"novita"`, `"replicate"`, "sambanova"` or `"together"`.
+            Name of the provider to use for inference. Can be `"black-forest-labs"`, `"cerebras"`, `"cohere"`, `"fal-ai"`, `"fireworks-ai"`, `"hf-inference"`, `"hyperbolic"`, `"nebius"`, `"novita"`, `"openai"`, `"replicate"`, "sambanova"` or `"together"`.
             defaults to hf-inference (Hugging Face Serverless Inference API).
             If model is a URL or `base_url` is passed, then `provider` is not used.
         token (`str`, *optional*):

src/huggingface_hub/inference/_providers/__init__.py

@@ -15,6 +15,7 @@
 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 .together import TogetherConversationalTask, TogetherTextGenerationTask, TogetherTextToImageTask
@@ -30,6 +31,7 @@
     "hyperbolic",
     "nebius",
     "novita",
+    "openai",
     "replicate",
     "sambanova",
     "together",
@@ -97,6 +99,9 @@
         "conversational": NovitaConversationalTask(),
         "text-to-video": NovitaTextToVideoTask(),
     },
+    "openai": {
+        "conversational": OpenAIConversationalTask(),
+    },
     "replicate": {
         "text-to-image": ReplicateTask("text-to-image"),
         "text-to-speech": ReplicateTextToSpeechTask(),

src/huggingface_hub/inference/_providers/new_provider.md

@@ -23,7 +23,11 @@ class MyNewProviderTaskProviderHelper(TaskProviderHelper):
         """Define high-level parameters."""
         super().__init__(provider=..., base_url=..., task=...)
 
-    def get_response(self, response: Union[bytes, Dict]) -> Any:
+    def get_response(
+        self,
+        response: Union[bytes, Dict],
+        request_params: Optional[RequestParameters] = None,
+    ) -> Any:
         """
         Return the response in the expected format.
 
@@ -37,7 +41,7 @@ class MyNewProviderTaskProviderHelper(TaskProviderHelper):
         """
         return super()._prepare_headers(headers, api_key)
 
-    def _prepare_route(self, mapped_model: str) -> str:
+    def _prepare_route(self, mapped_model: str, api_key: str) -> str:
         """Return the route to use for the request.
 
         Override this method in subclasses for customized routes.

src/huggingface_hub/inference/_providers/openai.py

@@ -0,0 +1,22 @@
+from typing import Optional
+
+from huggingface_hub.inference._providers._common import BaseConversationalTask
+
+
+class OpenAIConversationalTask(BaseConversationalTask):
+    def __init__(self):
+        super().__init__(provider="openai", base_url="https://api.openai.com")
+
+    def _prepare_api_key(self, api_key: Optional[str]) -> str:
+        if api_key is None:
+            raise ValueError("You must provide an api_key to work with OpenAI API.")
+        if api_key.startswith("hf_"):
+            raise ValueError(
+                "OpenAI provider is not available through Hugging Face routing, please use your own OpenAI API key."
+            )
+        return api_key
+
+    def _prepare_mapped_model(self, model: Optional[str]) -> str:
+        if model is None:
+            raise ValueError("Please provide an OpenAI model ID, e.g. `gpt-4o` or `o1`.")
+        return model

tests/test_inference_providers.py

@@ -37,6 +37,7 @@
     NovitaConversationalTask,
     NovitaTextGenerationTask,
 )
+from huggingface_hub.inference._providers.openai import OpenAIConversationalTask
 from huggingface_hub.inference._providers.replicate import ReplicateTask, ReplicateTextToSpeechTask
 from huggingface_hub.inference._providers.sambanova import SambanovaConversationalTask
 from huggingface_hub.inference._providers.together import (
@@ -707,6 +708,12 @@ def test_prepare_url_conversational(self):
         assert url == "https://api.novita.ai/v3/openai/chat/completions"
 
 
+class TestOpenAIProvider:
+    def test_prepare_url(self):
+        helper = OpenAIConversationalTask()
+        assert helper._prepare_url("sk-XXXXXX", "gpt-4o-mini") == "https://api.openai.com/v1/chat/completions"
+
+
 class TestReplicateProvider:
     def test_prepare_headers(self):
         helper = ReplicateTask("text-to-image")