Skip to content

Conversation

@alicialics
Copy link
Contributor

@alicialics alicialics commented Mar 26, 2023

  • Motivation for features / changes

Splitting an arbitrary vector into 256 parts hurts performance. This change speeds up KNN computation by 3x-5x.
The splitting needs to be done before matrix multiplication step for the cosine similarity computation. Splitting at knn step is already too late. The matrix limit is 2^13 x 2^13 on firefox and 2^14 x 2^14 on chrome (gl.MAX_TEXTURE_SIZE)

  • Technical description of changes
    Remove all code that relates to GPU splitting

  • Screenshots of UI changes

  • Detailed steps to verify changes work correctly (as executed by you)
    Ran vz_projector_test

  • Alternate designs / implementations considered

@alicialics alicialics changed the title Tensor Board: remove knn gpu splitting Embedding Projector: remove knn gpu splitting Apr 4, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

2 participants