Rotary Position Embedding explained deeply (w/ code)

Поділитися
Вставка
  • Опубліковано 4 лют 2025

КОМЕНТАРІ • 11

  • @datahouse24
    @datahouse24 Місяць тому

    great explanation

    • @jak-zee
      @jak-zee  Місяць тому

      @@datahouse24 thanks! Hope it helps you.

  • @ErenYeager-f7l
    @ErenYeager-f7l 16 днів тому

    Best explanation so far, loved itt!! BTW why only rotating in 2D space?? I mean why not rotating all the dimensions to the complex space and take the real part?

    • @jak-zee
      @jak-zee  16 днів тому

      Appreciate the comment! Good question. The rotation are done in 2D in euclidean space but for each pair of dimensions. The connection with complex space is because the angle you are rotating is dependent on frequency of oscillations found in complex space. Higher frequency means faster rotations for fine gain positioning. The real part is kind how to transfer that information to euclidean. That's how I understand at least.

  • @wilfredomartel7781
    @wilfredomartel7781 2 місяці тому

    Great work!

  • @sachinmohanty4577
    @sachinmohanty4577 Місяць тому

    Good content

    • @jak-zee
      @jak-zee  Місяць тому

      @@sachinmohanty4577 appreciate it!

  • @NikolayUlyanov-q9e
    @NikolayUlyanov-q9e Місяць тому

    keep up good work!

    • @jak-zee
      @jak-zee  Місяць тому +1

      Thanks man! Appreciate it...Hope the content helps.

  • @ErenYeager-f7l
    @ErenYeager-f7l 16 днів тому

    greattt explanation. Btw what’s the app you’re using?

    • @jak-zee
      @jak-zee  16 днів тому +1

      Thanks! App is called Concepts.