[inference provider] Add wavespeed.ai as an inference provider

[arabot777]

What’s in this PR
This PR adds wavespeed AI as a new inference provider for conversational tasks.

[arabot777]

Related to huggingface/huggingface.js#1424

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[hanouticelina]

@bot /style

[github-actions]

Style bot fixed some files and pushed the changes.

[hanouticelina]

thanks @arabot777 for the PR! 🤗
could you please update this line

huggingface_hub/src/huggingface_hub/inference/_client.py

Line 2563 in e5532d9

response = provider_helper.get_response(response)

and pass request_parameters into get_response():

response = provider_helper.get_response(response, request_parameters)

also, could you add Wavespeed to the Inference providers table in the documentation here?

I left a few comments to avoid duplicating the payload dictionary construction and use built-in types!

[hanouticelina]

Suggested change

	parameters: Dict,
	parameters: dict,

[hanouticelina]

Suggested change

	response: Union[bytes, Dict],
	response: Union[bytes, dict],

[hanouticelina]

let's use built-in types everywhere

Suggested change

	def _prepare_headers(self, headers: Dict, api_key: str) -> Dict:
	def _prepare_headers(self, headers: dict, api_key: str) -> dict:

[hanouticelina]

Suggested change

	) -> Optional[Dict]:
	) -> Optional[dict]:

[hanouticelina]

WavespeedAITextToVideoTask can inherit from WavespeedAITextToImageTask since it's the same payload preparation when the input modality is text

Suggested change

	class WavespeedAITextToVideoTask(WavespeedAITask):
	def __init__(self):
	super().__init__("text-to-video")
	def _prepare_payload_as_dict(
	self,
	inputs: Any,
	parameters: Dict,
	provider_mapping_info: InferenceProviderMapping,
	) -> Optional[Dict]:
	return {"prompt": inputs, **filter_none(parameters)}
	class WavespeedAITextToVideoTask(WavespeedAITextToImageTask):
	def __init__(self):
	WavespeedAITask.__init__(self, "text-to-video")

[hanouticelina]

since the payload is built the same for both image-to-image and image-to-video

Suggested change

	class WavespeedAIImageToVideoTask(WavespeedAITask):
	def __init__(self):
	super().__init__("image-to-video")
	def _prepare_payload_as_dict(
	self,
	inputs: Any,
	parameters: Dict,
	provider_mapping_info: InferenceProviderMapping,
	) -> Optional[Dict]:
	# Convert inputs to image (URL or base64)
	if isinstance(inputs, str) and inputs.startswith(("http://", "https://")):
	image = inputs
	elif isinstance(inputs, str):
	# If input is a file path, read it first
	with open(inputs, "rb") as f:
	file_content = f.read()
	image_b64 = base64.b64encode(file_content).decode("utf-8")
	image = f"data:image/jpeg;base64,{image_b64}"
	else:
	# If input is binary data
	image_b64 = base64.b64encode(inputs).decode("utf-8")
	image = f"data:image/jpeg;base64,{image_b64}"
	# Extract prompt from parameters if present
	prompt = parameters.pop("prompt", None)
	payload = {"image": image, **filter_none(parameters)}
	if prompt is not None:
	payload["prompt"] = prompt
	return payload
	class WavespeedAIImageToVideoTask(WavespeedAIImageToImageTask):
	def __init__(self):
	WavespeedAITask.__init__(self, "image-to-video")

[arabot777]

@hanouticelina thanks for the review! I've addressed all your feedback:
Ready for another review!

[hanouticelina]

almost ready to be merged! 🤗 thanks for the iteration @arabot777

[hanouticelina]

no need to redefine _prepare_headers, TaskProviderHelper. _prepare_headers already sets the bearer prefix in the authorization header

Suggested change

	def _prepare_headers(self, headers: dict, api_key: str) -> dict:
	headers = super()._prepare_headers(headers, api_key)
	if not api_key.startswith("hf_"):
	headers["Authorization"] = f"Bearer {api_key}"
	return headers

[arabot777]

@hanouticelina Done! Thanks for the thorough review!

[hanouticelina]

@bot /style

[github-actions]

Style bot fixed some files and pushed the changes.

[hanouticelina]

looks good, thank you @arabot777 for the contribution!