A model card with a customer handler to deploy the MSFT Kosmos2 model as an inference endpoint. Enjoy!
Hit me up on X/Twitter @yanMachX and let me know what you guys are building!
Expected Request payload
{
"inputs": "you can just leave this empty, for some reason, inference endpoint expects this",
# base64 encoded string representation of the image blob, here's an example
"image" : "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAAABGdBTUEAALGPC"
}
There is a python script that provides example API call the inference endpoint.