enable local transcription with Whisper
Closed this issue · 1 comments
swiftraccoon commented
add --local
arg
? https://github.com/ggerganov/whisper.cpp
swiftraccoon commented
I tried setting this up this weekend. Sadly I can only get it working with CPU and that's not an acceptable solution for me. Compiling whisper.cpp
with CUDA seems to require me to be on Fedora 37 (I'm on 38, 39 is already released..) due to NVIDIA's cuda toolkit only being available on F37. I did try to bypass that by manually compiling GCC12.2 for F38 but then I get an error:
WHISPER_CUBLAS=1 make -j NVCCFLAGS="-allow-unsupported-compiler"
= nvcc fatal : Unknown option '-fPIC'
so until I bother with setting up a GPU Docker or NVIDIA updates its support this will continue to be stuck for a bit