Nikola Janjušević

Results 12 comments of Nikola Janjušević

@ToucheSir is there any example code you could point to for starting on the path of writing batched_svd / how would you recommend approaching it?

```julia julia> conv(rand(Float16, 16, 16, 16, 1, 1) |> gpu, rand(Float16, 3, 3, 3, 1, 1) |> gpu) ┌ Warning: No valid algorithm found, probably bad params for convolution. └...

There is a similar error for gradients with `conv` and Float16 for 3D/4D/5D tensors as well. ```julia julia> w = rand(Float16, 3, 1, 1) |> gpu; julia> gradient(x->sum(conv(x, w)), rand(Float16,...

Ok, `JULIA_DEBUG=cuDNN` for the 5D conv and 3D gradient cases: ```julia julia> conv(rand(Float16, 16, 16, 16, 1, 1) |> gpu, rand(Float16, 3, 3, 3, 1, 1) |> gpu) ┌ Warning:...

@ToucheSir I'm back to being able to help (busy semester). Do you still want a test with NNlibCUDA v0.2.6?

> Any idea what the test failures are? https://buildkite.com/julialang/nccl-dot-jl/builds/36#0190cca1-710b-4b33-96b5-39d1fc64a262/303-1126 This looks unrelated to the pull request to me... did a configuration for the tests change since the last successful PR?

Can we retrigger the build test now that https://github.com/JuliaGPU/CUDA.jl/pull/2457 is merged?

Trying to get this issue resolved again now that the CUDA patch from earlier is released. Not sure if this build failure is my fault... any ideas @avik-pal @simonbyrne

> I think we need to bump the minimum julia versions in ci @avik-pal looks like julia 1.10 and above should work, do we want to drop support for 1.8...