And now self-attention is like a little bit more complicated. I don't want to go again to go too deep of how how that works, but essentially at- attention is there to to understand the relationships between the tokens. Uh Uh essentially what what is the important like if I say the sky is blue, blue and sky have a a very big correlation. Uh so that's what attention does. Based on the how you've trained your weights, you will understand what tokens should be attending to each other and put higher emphasis on those specific relationships. Now again going back to what I was say- saying before about the what that's why the tokenizer matters a lot cuz sky and blue are very easy to very easy for the model to make this relationship. While in our case you'll have to like combine different groups of tokens together the different characters together, that's quite a bit harder. But this that's that's what attention does. It's It's the what what each token should be attending to in the past and what has most importance for it.