Is it possible to track the attention (embeddings) through fine tunning process

[Coolchong]

Hello,

Very cool model! I tried it with our data set, and it gave great accuracy in terms of cell type prediction. I am wondering, since I am not trained as a data scientist, how could I extract the attentions of the sentences or embeddings (I hope these are the correct terms) during the fine tuning process to understand how C2S make predictions.

Thanks!