Experimenting with different types of channel attention, that would actually reduce the amount of computation needed, by throwing out most of the input.
The idea is to save on buffer reads on inference, by applying some function + gated activation to internal state, and using this result to determine if we should perceive anything
The idea is to save on buffer reads on inference, by applying some function + gated activation to internal state, and using this result to determine if we should perceive anything
https://webgpufundamentals.org/webgpu/lessons/webgpu-wgsl-function-reference.html
handy cheat sheet of WGSL functions
handy cheat sheet of WGSL functions
webgpufundamentals.org
WGSL Function Reference
How to use Buffers
❤1
This media is not supported in your browser
VIEW IN TELEGRAM
translation operator (random input, fixed shift)
texture NCA on Chaos Communication Camp (day 1). WGSL compute -> Vulkan on jetson nano. Mapping is done in the fragment shader stage
NCA particle simulation
https://transdimensional.xyz/projects/neural_ca/index.html
(github: https://github.com/PWhiddy/Growing-Neural-Cellular-Automata-Pytorch)
https://transdimensional.xyz/projects/neural_ca/index.html
(github: https://github.com/PWhiddy/Growing-Neural-Cellular-Automata-Pytorch)
representing radiance field as 3D Gaussians
https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/
https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/
==========================
swiss.gl ⛰👾📽 intro
==========================
This library: http://swiss.gl seems like an interesting model viewing tool for colab (example: last cell here)
and overall should be handy in many situations
A video tutorial with the use of this library has just been released: https://www.youtube.com/watch?v=jKt1Jq4FMRg
swiss.gl ⛰👾📽 intro
==========================
This library: http://swiss.gl seems like an interesting model viewing tool for colab (example: last cell here)
and overall should be handy in many situations
A video tutorial with the use of this library has just been released: https://www.youtube.com/watch?v=jKt1Jq4FMRg
👾1
edge cognition
========================== swiss.gl ⛰👾📽 intro ========================== This library: http://swiss.gl seems like an interesting model viewing tool for colab (example: last cell here) and overall should be handy in many situations A video…
demo from the tutorial: https://google.github.io/swissgl/demo/criticality.html
source code: https://github.com/google/swissgl/blob/main/demo/criticality.html
source code: https://github.com/google/swissgl/blob/main/demo/criticality.html
GitHub
swissgl/demo/criticality.html at main · google/swissgl
SwissGL is a minimalistic wrapper on top of WebGL2 JS API. It's designed to reduce the amount of boilerplate code required to manage GLSL shaders, textures and framebuffers when making proc...
edge cognition
While refactoring NCA to use webgpu arrays, I've encountered a strange performance problem: in 1x1 conv stage, when there is 12*48 cycles to compute state update, my machine initially hanged after ~50 cycles. It turned out, that accessing weights that were…
apparently, it was a bug in tint (google's WGSL compiler)
https://github.com/gpuweb/gpuweb/discussions/4281
thanks @munrocket for reporting it!
https://github.com/gpuweb/gpuweb/discussions/4281
thanks @munrocket for reporting it!
GitHub
Why `const` is slower than `var<private>` for arrays · gpuweb gpuweb · Discussion #4281
This is checked across different devices/OSs and probably not implementation issue. 37 line here explains issue: https://compute.toys/view/359 Try to change B/W top level declaration / usage in cod...
❤1