what is "embedding pooling" in Table 8.

[zhaoedf]

In paper, you wrote "We first compute the text-embedding for each prompt by passing it through the textencoder and then create a single classifier head by stacking all the embeddings.".

I don't quite understand, how to use this classifiier to get single similarity score? do mind showing me the corresponding code?
thx in advance.