Diego Devesa

It still happens in Anki 2.1. In AnkiDroid this can be fixed with css: ``` .night_mode .latex { filter: invert(1); } ``` However the .night_mode selector also doesn't work with...

Just updated to 2.1 beta 23 (Windows 10/Qt 5.9.2 PyQt 5.9) and it seems to work now without the need of any extra css. Only tested in the editor/card preview...

I don't think that this is a bug. The search behavior of cmake for the CUDA toolkit is documented [here](https://cmake.org/cmake/help/latest/module/FindCUDAToolkit.html#search-behavior). For non-standard paths, your workaround seems good enough. Setting the...

The prompt is already ingested in parallel, and you can specify the batch size with -b, however the benefits of doing that on CPU are not as big as on...

I don't know why that may be, these targets are listed [in the documentation](https://cmake.org/cmake/help/latest/module/FindCUDAToolkit.html#cuda-toolkit-cublas).

What does software that depend on it (or other libraries in the CUDA toolkit) do, distribute the dll files along with the software?

@WilliamTambellini this no longer happens with the CUDA backend after the sync with ggml from yesterday.

> cuBLAS dequantizes `x` to F16 and casts `y` to F16 and performs F16 mat mul, while OpenBLAS dequantizes `x` to F32 and performs F32 mat mul (if I'm not...

@NawafAlansari That looks correct. When the baby-llama example was created, graphs had a fixed size and could be allocated in the stack, but that was changed a while ago and...

We have a mechanism in place for using multiple backends simultaneously with `ggml_backend_sched`. Have you looked into this?