I would like to collect feedbacks on the toolkit we are building for zero shot inference, any comments is more than appreciated.