Support text-generation in InferenceClient

docs/source/guides/inference.md

@@ -139,7 +139,7 @@ has a simple API that supports the most common tasks. Here is a list of the curr
 | | [Summarization](https://huggingface.co/tasks/summarization)                  | ✅ | [`~InferenceClient.summarization`] |
 | | [Table Question Answering](https://huggingface.co/tasks/table-question-answering)       | |  |
 | | [Text Classification](https://huggingface.co/tasks/text-classification)            | |  |
-| | [Text Generation](https://huggingface.co/tasks/text-generation)   | |  |
+| | [Text Generation](https://huggingface.co/tasks/text-generation)   | ✅ | [`~InferenceClient.text_generation`] |
 | | [Token Classification](https://huggingface.co/tasks/token-classification)           | |  |
 | | [Translation](https://huggingface.co/tasks/translation)       | |  |
 | | [Zero Shot Classification](https://huggingface.co/tasks/zero-shot-image-classification)       | |  |

docs/source/package_reference/inference_client.md

@@ -36,6 +36,35 @@ For most tasks, the return value has a built-in type (string, list, image...). H
 
 [[autodoc]] huggingface_hub.inference._types.ImageSegmentationOutput
 
+### Text generation types
+
+[`~InferenceClient.text_generation`] task has a greater support than other tasks in `InferenceClient`. In
+particular, user inputs and server outputs are validated using [Pydantic](https://docs.pydantic.dev/latest/)
+if this package is installed. Therefore, we recommend installing it (`pip install pydantic`)
+for a better user experience.
+
+You can find below the dataclasses used to validate data and in particular [`~huggingface_hub.inference._text_generation.TextGenerationParameters`] (input),
+[`~huggingface_hub.inference._text_generation.TextGenerationResponse`] (output) and 
+[`~huggingface_hub.inference._text_generation.TextGenerationStreamResponse`] (streaming output).
+
+[[autodoc]] huggingface_hub.inference._text_generation.TextGenerationParameters
+
+[[autodoc]] huggingface_hub.inference._text_generation.TextGenerationResponse
+
+[[autodoc]] huggingface_hub.inference._text_generation.TextGenerationStreamResponse
+
+[[autodoc]] huggingface_hub.inference._text_generation.InputToken
+
+[[autodoc]] huggingface_hub.inference._text_generation.Token
+
+[[autodoc]] huggingface_hub.inference._text_generation.FinishReason
+
+[[autodoc]] huggingface_hub.inference._text_generation.BestOfSequence
+
+[[autodoc]] huggingface_hub.inference._text_generation.Details
+
+[[autodoc]] huggingface_hub.inference._text_generation.StreamDetails
+
 ## InferenceAPI
 
 [`InferenceAPI`] is the legacy way to call the Inference API. The interface is more simplistic and requires knowing

pyproject.toml

@@ -7,6 +7,9 @@ preview = true
 ignore_missing_imports = true
 no_implicit_optional = true
 scripts_are_modules = true
+plugins = [
+  "pydantic.mypy"
+]
 
 [tool.ruff]
 # Ignored rules:

setup.py

@@ -55,6 +55,7 @@ def get_version() -> str:
     "Pillow",
     "gradio",  # to test webhooks
     "numpy",  # for embeddings
+    "pydantic",  # for text-generation-inference
 ]
 
 # Typing extra dependencies list is duplicated in `.pre-commit-config.yaml`
@@ -66,6 +67,7 @@ def get_version() -> str:
     "types-toml",
     "types-tqdm",
     "types-urllib3",
+    "pydantic",  # for text-generation dataclasses
 ]
 
 extras["quality"] = [