diff --git a/docs/source/en/guides/inference.md b/docs/source/en/guides/inference.md index 6fa08f2736..642947f73f 100644 --- a/docs/source/en/guides/inference.md +++ b/docs/source/en/guides/inference.md @@ -162,10 +162,10 @@ You can use [`InferenceClient`] to run chat completion with local inference serv Authentication can be done in two ways: -**Routed through Hugging Face** : Use Hugging Face as a proxy to access third-party providers. The calls will be routed through Hugging Face's infrastructure using our provider keys, and the usage will be billed directly to your Hugging Face account. +**Routed through Hugging Face** : Use Hugging Face as a proxy to access third-party providers. The calls will be routed through Hugging Face's infrastructure using our provider keys, and the usage will be billed directly to your Hugging Face account. You can authenticate using a [User Access Token](https://huggingface.co/docs/hub/security-tokens). You can provide your Hugging Face token directly using the `api_key` parameter: - + ```python >>> client = InferenceClient( provider="replicate", @@ -196,37 +196,37 @@ For more details, refer to the [Inference Providers pricing documentation](https [`InferenceClient`]'s goal is to provide the easiest interface to run inference on Hugging Face models, on any provider. It has a simple API that supports the most common tasks. Here is a table showing which providers support which tasks: -| Task | Black Forest Labs | Cerebras | Cohere | fal-ai | Featherless AI | Fireworks AI | Groq | HF Inference | Hyperbolic | Nebius AI Studio | Novita AI | Replicate | Sambanova | Together | -| --------------------------------------------------- | ----------------- | -------- | ------ | ------ | -------------- | ------------ | ---- | ------------ | ---------- | ---------------- | --------- | --------- | --------- | -------- | -| [`~InferenceClient.audio_classification`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.audio_to_audio`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.automatic_speech_recognition`] | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.chat_completion`] | ❌ | ✅ | ✅ | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | -| [`~InferenceClient.document_question_answering`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.feature_extraction`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ✅ | ❌ | ❌ | ✅ | ❌ | -| [`~InferenceClient.fill_mask`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.image_classification`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.image_segmentation`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.image_to_image`] | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | -| [`~InferenceClient.image_to_video`] | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.image_to_text`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.object_detection`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | | ❌ | -| [`~InferenceClient.question_answering`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.sentence_similarity`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.summarization`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.table_question_answering`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.text_classification`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.text_generation`] | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ | ✅ | -| [`~InferenceClient.text_to_image`] | ✅ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ❌ | ✅ | -| [`~InferenceClient.text_to_speech`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | -| [`~InferenceClient.text_to_video`] | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | -| [`~InferenceClient.tabular_classification`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.tabular_regression`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.token_classification`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.translation`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.visual_question_answering`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.zero_shot_image_classification`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | -| [`~InferenceClient.zero_shot_classification`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| Task | Black Forest Labs | Cerebras | Clarifai | Cohere | fal-ai | Featherless AI | Fireworks AI | Groq | HF Inference | Hyperbolic | Nebius AI Studio | Novita AI | Replicate | Sambanova | Together | +| --------------------------------------------------- | ----------------- | -------- | -------- | ------ | ------ | -------------- | ------------ | ---- | ------------ | ---------- | ---------------- | --------- | --------- | --------- | -------- | +| [`~InferenceClient.audio_classification`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.audio_to_audio`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.automatic_speech_recognition`] | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.chat_completion`] | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | +| [`~InferenceClient.document_question_answering`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.feature_extraction`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ✅ | ❌ | ❌ | ✅ | ❌ | +| [`~InferenceClient.fill_mask`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.image_classification`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.image_segmentation`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.image_to_image`] | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | +| [`~InferenceClient.image_to_video`] | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.image_to_text`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.object_detection`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.question_answering`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.sentence_similarity`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.summarization`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.table_question_answering`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.text_classification`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.text_generation`] | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ | ✅ | +| [`~InferenceClient.text_to_image`] | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ❌ | ✅ | +| [`~InferenceClient.text_to_speech`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | +| [`~InferenceClient.text_to_video`] | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | +| [`~InferenceClient.tabular_classification`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.tabular_regression`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.token_classification`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.translation`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.visual_question_answering`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.zero_shot_image_classification`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | +| [`~InferenceClient.zero_shot_classification`] | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | > [!TIP] > Check out the [Tasks](https://huggingface.co/tasks) page to learn more about each task. @@ -294,7 +294,7 @@ You might wonder why using [`InferenceClient`] instead of OpenAI's client? There ## Function Calling -Function calling allows LLMs to interact with external tools, such as defined functions or APIs. This enables users to easily build applications tailored to specific use cases and real-world tasks. +Function calling allows LLMs to interact with external tools, such as defined functions or APIs. This enables users to easily build applications tailored to specific use cases and real-world tasks. `InferenceClient` implements the same tool calling interface as the OpenAI Chat Completions API. Here is a simple example of tool calling using [Nebius](https://nebius.com/) as the inference provider: ```python @@ -345,7 +345,7 @@ print(response.choices[0].message.tool_calls[0].function.arguments) InferenceClient supports JSON mode for syntactically valid JSON responses and Structured Outputs for schema-enforced responses. JSON mode provides machine-readable data without strict structure, while Structured Outputs guarantee both valid JSON and adherence to a predefined schema for reliable downstream processing. -We follow the OpenAI API specs for both JSON mode and Structured Outputs. You can enable them via the `response_format` argument. Here is an example of Structured Outputs using [Cerebras](https://www.cerebras.ai/) as the inference provider: +We follow the OpenAI API specs for both JSON mode and Structured Outputs. You can enable them via the `response_format` argument. Here is an example of Structured Outputs using [Cerebras](https://www.cerebras.ai/) as the inference provider: ```python from huggingface_hub import InferenceClient diff --git a/src/huggingface_hub/inference/_client.py b/src/huggingface_hub/inference/_client.py index 092512bf33..f50e7d56eb 100644 --- a/src/huggingface_hub/inference/_client.py +++ b/src/huggingface_hub/inference/_client.py @@ -130,7 +130,7 @@ class InferenceClient: Note: for better compatibility with OpenAI's client, `model` has been aliased as `base_url`. Those 2 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"`, `publicai`, `"replicate"`, `"sambanova"`, `"scaleway"`, `"together"` or `"zai-org"`. + Name of the provider to use for inference. Can be `"black-forest-labs"`, `"cerebras"`, `"clarifai"`, `"cohere"`, `"fal-ai"`, `"featherless-ai"`, `"fireworks-ai"`, `"groq"`, `"hf-inference"`, `"hyperbolic"`, `"nebius"`, `"novita"`, `"nscale"`, `"openai"`, `publicai`, `"replicate"`, `"sambanova"`, `"scaleway"`, `"together"` or `"zai-org"`. 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. If model is a URL or `base_url` is passed, then `provider` is not used. token (`str`, *optional*): diff --git a/src/huggingface_hub/inference/_generated/_async_client.py b/src/huggingface_hub/inference/_generated/_async_client.py index e5809617ec..45285d8390 100644 --- a/src/huggingface_hub/inference/_generated/_async_client.py +++ b/src/huggingface_hub/inference/_generated/_async_client.py @@ -118,7 +118,7 @@ class AsyncInferenceClient: Note: for better compatibility with OpenAI's client, `model` has been aliased as `base_url`. Those 2 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"`, `publicai`, `"replicate"`, `"sambanova"`, `"scaleway"`, `"together"` or `"zai-org"`. + Name of the provider to use for inference. Can be `"black-forest-labs"`, `"cerebras"`, `"clarifai"`, `"cohere"`, `"fal-ai"`, `"featherless-ai"`, `"fireworks-ai"`, `"groq"`, `"hf-inference"`, `"hyperbolic"`, `"nebius"`, `"novita"`, `"nscale"`, `"openai"`, `publicai`, `"replicate"`, `"sambanova"`, `"scaleway"`, `"together"` or `"zai-org"`. 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. If model is a URL or `base_url` is passed, then `provider` is not used. token (`str`, *optional*): diff --git a/src/huggingface_hub/inference/_providers/__init__.py b/src/huggingface_hub/inference/_providers/__init__.py index 69051ff914..79d2bd75c8 100644 --- a/src/huggingface_hub/inference/_providers/__init__.py +++ b/src/huggingface_hub/inference/_providers/__init__.py @@ -9,6 +9,7 @@ from ._common import TaskProviderHelper, _fetch_inference_provider_mapping from .black_forest_labs import BlackForestLabsTextToImageTask from .cerebras import CerebrasConversationalTask +from .clarifai import ClarifaiConversationalTask from .cohere import CohereConversationalTask from .fal_ai import ( FalAIAutomaticSpeechRecognitionTask, @@ -50,6 +51,7 @@ PROVIDER_T = Literal[ "black-forest-labs", "cerebras", + "clarifai", "cohere", "fal-ai", "featherless-ai", @@ -78,6 +80,9 @@ "cerebras": { "conversational": CerebrasConversationalTask(), }, + "clarifai": { + "conversational": ClarifaiConversationalTask(), + }, "cohere": { "conversational": CohereConversationalTask(), }, diff --git a/src/huggingface_hub/inference/_providers/_common.py b/src/huggingface_hub/inference/_providers/_common.py index 4194c356b9..366fc3f45d 100644 --- a/src/huggingface_hub/inference/_providers/_common.py +++ b/src/huggingface_hub/inference/_providers/_common.py @@ -24,6 +24,7 @@ # status="live") "cerebras": {}, "cohere": {}, + "clarifai": {}, "fal-ai": {}, "fireworks-ai": {}, "groq": {}, diff --git a/src/huggingface_hub/inference/_providers/clarifai.py b/src/huggingface_hub/inference/_providers/clarifai.py new file mode 100644 index 0000000000..262b8ace32 --- /dev/null +++ b/src/huggingface_hub/inference/_providers/clarifai.py @@ -0,0 +1,10 @@ +from ._common import BaseConversationalTask + + +_PROVIDER = "clarifai" +_BASE_URL = "https://api.clarifai.com/v2/ext/openai" + + +class ClarifaiConversationalTask(BaseConversationalTask): + def __init__(self): + super().__init__(provider=_PROVIDER, base_url=_BASE_URL) diff --git a/tests/test_inference_providers.py b/tests/test_inference_providers.py index 79c668fd47..98df07a325 100644 --- a/tests/test_inference_providers.py +++ b/tests/test_inference_providers.py @@ -17,6 +17,7 @@ recursive_merge, ) from huggingface_hub.inference._providers.black_forest_labs import BlackForestLabsTextToImageTask +from huggingface_hub.inference._providers.clarifai import ClarifaiConversationalTask from huggingface_hub.inference._providers.cohere import CohereConversationalTask from huggingface_hub.inference._providers.fal_ai import ( _POLLING_INTERVAL, @@ -293,6 +294,34 @@ def test_prepare_payload_as_dict(self): } +class TestClarifaiProvider: + def test_prepare_url(self): + helper = ClarifaiConversationalTask() + assert ( + helper._prepare_url("clarifai_api_key", "username/repo_name") + == "https://api.clarifai.com/v2/ext/openai/v1/chat/completions" + ) + + def test_prepare_payload_as_dict(self): + helper = ClarifaiConversationalTask() + payload = helper._prepare_payload_as_dict( + [{"role": "user", "content": "Hello!"}], + {}, + InferenceProviderMapping( + provider="clarifai", + hf_model_id="meta-llama/llama-3.1-8B-Instruct", + providerId="meta-llama/llama-3.1-8B-Instruct", + task="conversational", + status="live", + ), + ) + + assert payload == { + "messages": [{"role": "user", "content": "Hello!"}], + "model": "meta-llama/llama-3.1-8B-Instruct", + } + + class TestFalAIProvider: def test_prepare_headers_fal_ai_key(self): """When using direct call, must use Key authorization."""