gpu-programming

import taichi as ti

loss = ti.field(dtype = ti.f32, shape = (), needs_grad = True)
x = ti.field(dtype = ti.f32, shape = (), needs_grad = True)

@ti.kernel
def work():
    loss[None] = x[None] * ti.random()
    # loss[None] = x[None] * 0.5 # this works

x[None] = 10
with ti.Tape(loss):
    work()
print(x.grad[None])

One submission of the ICLR 2021 challenge reports the following bug:

"We were unable to project points from the Grassmannian or the Stiefel manifold to the tangent space using Geomstats combined with the default canonical metric, or our custom metrics. We tried both geomstats.learning.preprocessing.ToTangentSpace and the newer geomstats.geometry.grassmannian.Grassmannian.to_tangent, but bot

May	JUN	Aug
	20
2020	2021	2022

gpu-programming

Here are 166 public repositories matching this topic...

taichi-dev / taichi

Kernel with ti.random() runtime error while using autodiff

[Doc] Update doc on ti.Matrix (like zeros, ones, identity)

[Lang] Deprecate ti.field(ti.f32), use ti.field(ti.f32, None) instead for advanced layouts

taskflow / taskflow

plasma-umass / scalene

potential enhancement: incorporate flame graphs as in py-spy

EmbarkStudios / rust-gpu

QianMo / Game-Programmer-Study-Notes

uber / aresdb

calebwin / emu

QianMo / GPU-Gems-Book-Source-Code

geomstats / geomstats

Projection to tangent space in Stiefel and Grassmannian fails on einsum

Formulas not compiled in documentation

Clean Pytorch Backend

brucefan1983 / CUDA-Programming

QianMo / GPU-Pro-Books-Source-Code

Zielon / PBRVulkan

JuliaGPU / CuArrays.jl

Glavnokoman / vuh

gpufit / Gpufit

fastflow / fastflow

stetre / moonlibs

johannesugb / VolumetricLinesUnity

rAzoR8 / SpvGenTwo

adamnemecek / awesome-metal

arctern-io / arctern

hollance / metal-gpgpu

Heteroflow / Heteroflow

eyalroz / cuda-kat

ysh329 / OpenCL-101

LanLou123 / Webgl-Erosion

Glavnokoman / vulkan-compute-example

weissenberger / gpuhd

xmartlabs / cuda-calculator

bcosenza / sycl-bench

Improve this page

Add this topic to your repo