MariusDrulea

@Unic-X how did you manage to find the solution? It worked for me as well. 

I do use the release: https://github.com/roldanjr/pomatez/releases/download/v1.6.4/Pomatez-v1.6.4-win-x64-setup.exe An update: If I set the app to be "always on top" than it will go "mostly" fullscreen: covers everything except the task-bar.

For debugging I did the following modifications: 1. I only use the first 100 images from the training set: `xtrain, ytrain = MLDatasets.MNIST(split=:train)[1:100]` 2. @time the pullback call: ``` @time...

After some debugging, it seems **the usage of L2 regularization in the model_loss function leads to this issue**. https://github.com/FluxML/model-zoo/blob/master/vision/vae_mnist/vae_mnist.jl#L64 If I comment out the regularization (simply set reg = 0),...

Indeed, the issue is in this line: `reg = λ * sum(x->sum(x.^2), Flux.params(decoder))` See bellow a MWE for the issue. `loss_fun_slow` calls Flux.params(model) each iteration and leads to the mentioned...

Unfortunately, the problem is present again in Flux 0.13.10, so we have to reopen this task. In the following MWE, `loss_slow` compiles at every iteration. Additionally, **the runtime and the...

I wanted to dig a bit into the issue, but I'm facing the following (cumulative) issues: 1. I cannot download any dataset if the proxy (default) is activated. For the...

This issue also appears in julia 1.9.2, https://discourse.julialang.org/t/profileview-jl-makes-my-computer-lag/ `ENV["GTK_AUTO_IDLE"] = false` still works. I think however we should not do this manually. The fix itself is not a big deal,...

This is not advice for generic Julia programming. It's about providing users with a decent debugging experience. As I motivated [here](https://discourse.julialang.org/t/debugging-extremely-slow/53801/56?u=mariusd), all new users use the vscode extension and their...

> I think a short sentence at the end of the docs saying "Here are some alternative debugging options" with links would be fine. I think this is way too...