Call for testing: Vulkan-inference support

Hello, since revision 5b61d5e8 we now supports Vulkan-based inference/acceleration. This provides drastically better transcribe speed(~39x) than the poorly optimized CPU inference mode on supported systems:

Screenshot depicting transcribe duration difference between Vulkan-vs-CPU inference modes, where CPU inference takes 99 minutes and Vulkan-based inference only taking about 2.5

As this is a rather big change we would like to hold a call for testing to see whether there's any regressions we need to look into. Please install the build in the beta channel, run your usual audio transcribe workload, and made a reply or file a new issue when you encountered a problem(the following is the typical commands to do so):

sudo snap install --channel=beta whisper-cpp
whisper-cpp.download-ggml-model medium
whisper-cpp.cli -m ggml-medium.bin --threads $(nproc) --language auto --print-progress --flash-attn --output-srt _input_file_

Cheers! 🥂

Edited by 林博仁 Buo-ren Lin