Enable inference to work on HF endpoints

[coyotte508]

See https://huggingface.co/docs/inference-endpoints/guides/test_endpoint

Instead of model we could have endpoint in the calls. Or we could keep model but specify the url instead of the name.