Maybe consider CLBlast for acceleration on older GPUs (also AMD ones)

[codingbutstillalive]

Feature request

The whisper.cpp project makes use of it, if I am not mistaken. See here e.g. ggml-org/whisper.cpp#173

Motivation

I think the inference part of the models needs to be improved on non state of the art hardware.

Your contribution

NAN

[niansa]

Related: #746

[brlin-tw]

@codingbutstillalive

The whisper.cpp project makes use of it, if I am not mistaken.

I believe the patch isn't merged to whisper.cpp (also the issue is in an open state), so no.

[cebtenzzre]

The CLBlast backend has been removed upstream, as the Vulkan backend is superior.