Parameter setting of the zero-shot inference

[yifliu3]

Hi,

I'm curious about the parameter settings used in the inference period, and I suppose that you use the same setting across the downstream classification datasets as the provided demo. Could you please share more details?