[feature]: GPU acceleration for Apple Silicon

[ramen]

Feature Description

Enable GPU-optimized transcription on Mac. Currently, we only support CUDA, which requires NVIDIA. On Mac, we may be able to use MPS. We need to evaluate the current state of support for MPS with Whisper/PyTorch to see if we're able to support this yet, and if so, it may just be a matter of upgrading a few packages and tweaking some Python code.

Use Case

GPU acceleration provides for much faster transcription. This would help bring Mac performance in line with other operating systems.

Additional Context

• https://pytorch.org/blog/introducing-accelerated-pytorch-training-on-mac/
• M1 GPU support openai/whisper#91
• Uses MPS (Mac acceleration) by default when available openai/whisper#382

[zachvalenti]

Thrilled to see this here! In case it might be of any use, I figured I'd drop a link to this propject that could bridge the gap: https://github.com/ggerganov/whisper.cpp