do not rely on "$ref": "/inference/schemas/text2text-generation/input.json", for Summarization / Translation. Makes things clearer + it's not possible to extend the parameters which was not possible before.
typo in text-to-image
add src_lang and tgt_lang in translation params
use enum for early_stopping parameter (in common defs)
for audio-classification, automatic speech recognition, image classification, image to image, object detection:
mention base64-encoded string as input
mention raw data can be sent if no parameters in the JSON payload
more descriptions in object detection
more descriptions in image segmentation
mention bytes output in text-to-image and image-to-image
This PR should prove useful for the ongoing work of generating documentation pages based on the input/output specs (see https://github.com/huggingface/hub-docs/pull/1379).
In particular, I've made changes when adding new pages to the docs (PRs https://github.com/huggingface/hub-docs/pull/1398 and https://github.com/huggingface/hub-docs/pull/1399).
This PR is now ready for review.
Changes:
oneOf
+ list ofconst
"$ref": "/inference/schemas/text2text-generation/input.json",
for Summarization / Translation. Makes things clearer + it's not possible to extend the parameters which was not possible before.text-to-image
src_lang
andtgt_lang
in translation paramsearly_stopping
parameter (in common defs)audio-classification
,automatic speech recognition
,image classification
,image to image
,object detection
:base64-encoded string
as inputobject detection
image segmentation
text-to-image
andimage-to-image