Jan

with the hardcoding or your changes, this will give you an "Surface does not support the adapter's queue family" in `wgpuSurfaceConfigure` and the following advice https://github.com/gfx-rs/wgpu/issues/1344#issuecomment-823448056 note that the D3D12...

The warnings also show in wgpu-native when I force the backend to D3D12 there. But otherwise it does work. Apart from compute shaders, which I don't think are meant to...

> Oh, interesting ... I will try wgpu-native when I'm back at the office (where I have a Windows machine) tomorrow. What happens on your machine is you modify `cube.py`...

I will try this on both machines when I am home in ~90 minutes

checked on both machines, seems to be good now. examples work with gui, pytest passes. Also tried with forced Vulkan and D3D12. I will collect the fatal errors and raise...

> can we use only repetition penalty Likely yes in most practical settings. But in the case of runnings generation for eval benchmarks, some require greedy decoding. Using any of...

It's A but y flipped. See the version at the bottom of the readme here https://github.com/gfx-rs/wgpu?tab=readme-ov-file#coordinate-systems

ah, now I get the question. I recently saw [this video](https://youtu.be/d6tp43wZqps) which makes use of this sampling method. So it seems answer C is correct, it's normalized to [-1..1]

I understand this. But can there be a solution that doesn't require the end user to write this shema by hand(in my case there is some fields that contain a...

also have this where it lists a dependency twice... might be some issue with other packages? I don't see what changed.