-
I am working with FFTs on GPU, and in my the complete model I also need the FFT output in half-complex format. Individually, the FFTs and the complex-half-complex transform seem to work, but not toget…
-
The bug appears on file `futhark-ad/rnns/lstm/lstm.fut` compiled with `futhark cuda lstm.fut`
(using the branch `ad-genred-opt`, but very likely also `clean-ad` given that it is likely produced by th…
-
I am writing backward passes for `ImageTransformations.warp!` and other Image agnostic functions. I have been solving most of the problems related to adjoints for Images, but after all I have gotten s…
-
I have a model that needs to run partly on the CPU and partly on the GPU. The part that needs to run on the CPU simply cannot run on the GPU. The part that needs to run on the GPU is far too slow when…
-
It seems to me that there should be a general way of defining
- Presheaf morphisms (i.e. natural transformations)
- Decorating functors from FinOrd to sets of Presheaves (i.e., a function that che…
-
Training Neural ODEs with callbacks currently fails for most of the sensitivity algorithms available to `concrete_solve`: Only `ForwardDiffSensitivity` works for the following MWE. However, this metho…
-
MWE:
```julia
Z = randn(rng, Float32, (n,m))
Z .= Flux.Zygote.OneElement(0.1, (2,), axes(Z))
vec(Z) .= vec(Flux.Zygote.OneElement(0.1, (2,), axes(Z)))
```
-
Here's a thing that isn't currently possible and is, I believe, something that we might actually want to care about. Consider the pullback for `AbstractMatrix` multiplication:
```julia
function rrul…
-
I am so sorry to post this kind of mess. But I am just not sure what is going on here. Why is it so cryptic?
```
> remotes::install_github("tidymodels/tidyposterior")
Downloading GitHub repo ti…
-
< -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ -->
## The problem
ERROR: lazy loading failed for package ‘tidyposterior’
I'm having trouble with ...
## Reproducible exam…