Where is the vocabulary list used as a source for the prototypes?

[Med-Rokaimi]

Thank you so much for this work.
In the paper, you mentioned that you used statistic list of words to build prototypes that describe patches.

For example, one patch could be translated to “short up then steadily down”.

I wanted to modify this list, but I cannot figure out where this list is combined to the model inputs.

Can you please provide more explaination about this ?
Thank you

[IansSUn]

Excuse me, have you solved it? I also have this question.
Thank you

[IansSUn]

self.mapping_layer(self.word_embeddings.permute(1, 0)).permute(1, 0)
I think the code here refers to the content of " A simple solution is to maintain a small collection of text prototypes by linearly probing E, ...", it makes the vocab_size from the LLM's original vocabulary size to 1000, 1000 refers to the number of words related to the changing trend, like "short up then steadily down". It's just that these 1000 embeddings need training.
This is what I think, I don't know if it's right. Looking forward to your reply.