bulk_transcribe_youtube_videos_from_playlist icon indicating copy to clipboard operation
bulk_transcribe_youtube_videos_from_playlist copied to clipboard

Cuda detection

Open lockmeister opened this issue 7 months ago • 4 comments

Thank you for your great work. The quality of the transcriptions is great. I just wanted to share with you that the cuda detection did not work for me because it is based on looking for an installation that used Anaconda. I modified your function so that it would find cuda installations in some common situations, such as mine.

def get_cuda_toolkit_path(): # Check for common CUDA toolkit installation directories common_paths = [ '/usr/local/cuda/bin', '/usr/local/cuda-*/bin', ]

# Look for nvcc in common paths
for path in common_paths:
    cuda_paths = glob.glob(path)
    if cuda_paths:
        # Return the first matching path
        return cuda_paths[0]

# As a fallback, use the 'which' command to find the nvcc binary
nvcc_path = os.popen('which nvcc').read().strip()
if nvcc_path:
    # Return the directory containing the nvcc binary
    return os.path.dirname(nvcc_path)

# If the CUDA toolkit path is still not found, return None
return None

lockmeister avatar Nov 15 '23 09:11 lockmeister