RuntimeError: CUDA failed with error CUDA driver version is insufficient for CUDA runtime version

SYSTRAN / faster-whisper

Faster Whisper transcription with CTranslate2

MIT License

12.66k stars 1.06k forks source link

RuntimeError: CUDA failed with error CUDA driver version is insufficient for CUDA runtime version #1115

Closed SwAt1563 closed 2 weeks ago

SwAt1563 commented 3 weeks ago

Issue Summary: I'm attempting to run faster-whisper with GPU support on my device, which has CUDA 12.6 installed. The following package versions are being used:

faster-whisper==1.0.3
ctranslate2==4.4.0
wyoming==1.5.3

The model runs successfully on the CPU; however, it does not utilize the GPU as expected. I am looking for a solution to enable GPU acceleration for faster-whisper using CUDA 12.6.

jhj0517 commented 2 weeks ago

If you're using docker you can try changing https://github.com/SYSTRAN/faster-whisper/blob/814472fdbf7faf5d77d65cdb81b1528c0dead02a/docker/Dockerfile#L5

into

 RUN pip3 install faster-whisper torch --index-url https://download.pytorch.org/whl/cu124

Currently there's a strict version matrix between torch and faster-whisper : https://github.com/SYSTRAN/faster-whisper/issues/1086#issue-2612158224

SwAt1563 commented 2 weeks ago

Thank you for your guidance, but it still doesn't work.