How to explain Q, K and V of Self Attention in Transformers (BERT)?

Поделиться
HTML-код
  • Опубликовано: 13 янв 2025

Комментарии • 26

  • @alexfeng75
    @alexfeng75 Год назад +9

    this is probably the clearest explanation I have seen on QKV. Thanks!

  • @Tuscani2005GT
    @Tuscani2005GT 8 месяцев назад

    This is the best channel on this subject. You actually made the QKV concept click for me in 15 minutes when hours of videos could not. Thank you for not just regurgitating Attention is all you need content and actually reforming it into a simpler explanation.

  • @exxzxxe
    @exxzxxe Год назад +2

    One of the best explanations on RUclips.

  • @hkchee9858
    @hkchee9858 Год назад +4

    the video provides an extremely clear picture of QKV in self-attention

  • @SyntharaPrime
    @SyntharaPrime 2 года назад +2

    Incredibly excellent explanation. Really good. Thank you very much.

    • @code4AI
      @code4AI  2 года назад

      Glad it was helpful!

  • @rikki146
    @rikki146 Год назад

    best explanation i can find on the internet. thank you!

  • @vassilisworld
    @vassilisworld Год назад +2

    you are born to teach!! thank you

  • @RuairiODonnellFOTO
    @RuairiODonnellFOTO Год назад +2

    I thought the multi-headed aspect of the algorithm was to allow for small variants in the word positions. I think Attention is all you need used a sin and cosine to have a wave function to model the word proximity in a sentence.

  • @Sunbee2000
    @Sunbee2000 Год назад

    Excellent explanation. Thanks a lot.

  • @fatemehfathiniazi4932
    @fatemehfathiniazi4932 Год назад

    Thanks for the great explanation!

  • @hongt1930
    @hongt1930 Год назад

    thanks you for the explaination of self dot product attention and connect it with the paper! It would be even better if you can tap into the backpropogation since you mentioned it in the selecting the vectors K, Q, V.

  • @lorrancmlopes
    @lorrancmlopes 7 месяцев назад

    Excelent video! Thanks a lot!

  • @anandgs
    @anandgs Год назад

    You are amazing Sir!!

  • @haroldsu1696
    @haroldsu1696 5 месяцев назад

    you did a good job! Sir , and Thank you.

    • @code4AI
      @code4AI  5 месяцев назад

      Glad to help

  • @BasmaSH1
    @BasmaSH1 Год назад

    Thanks a lot for your explanation.

  • @terryliu3635
    @terryliu3635 7 месяцев назад

    Great explanation! Could you pls help explain how tensor q, k, v values are initiated? I've checked more than 5 videos and none of them have explained where q, k, v values come from... Thank you!

  • @unclecode
    @unclecode Год назад +1

    The video is fantastic! By the way, based on the beginning, I assumed the you intended to explain how context-aware embedding can be generated and then proceed to explain Q, K, and V. However, it seems that the embedding part was missed, and you moved straight to the Q, K, and V concept. Do you have any video for that?

  • @AdamCole1
    @AdamCole1 9 месяцев назад

    Brilliant!

  • @punk3900
    @punk3900 8 месяцев назад

    ❤❤❤

  • @boutroschalouhy2931
    @boutroschalouhy2931 2 года назад +1

    Thanks

    • @code4AI
      @code4AI  Год назад

      You are welcome.

  • @masonhzx
    @masonhzx Год назад +1

    can you give data, number sample of how Q, K, V get calculated for text like "Welcome how are you"?