Python Runtime v1.1.0 (macOS arm64)

What's New in v1.1.0

MPS GPU Acceleration (5-15x speedup)

• Patched PyTorch 2.10.0 with Metal scalar address-space fix (device → constant in c10/metal/indexing.h)
• Pyannote diarization now runs on Apple Silicon GPU via MPS backend
• Automatic MPS detection with CPU fallback

Architecture Change: Scripts Separated

• Python scripts are no longer in the runtime archive
• Scripts (run_diarization.pyc, download_model.pyc) ship with the app bundle
• Script updates = app update only (no re-downloading 240MB runtime)
• Runtime download only needed when Python/PyTorch version changes

Contents

• Python 3.13.11 (arm64)
• PyTorch 2.10.0 (patched for MPS)
• pyannote.audio + speaker-diarization model

System Requirements

• macOS 14+ (Sonoma or later)
• Apple Silicon (M1/M2/M3/M4)

Size

• Download: ~240 MB (tar.gz)
• Extracted: ~770 MB

Python Runtime v1.0.0 (macOS arm64)

MinuteAI Python Runtime v1.0.0

Pre-built Python runtime with bundled pyannote speaker diarization model.

Contents

• Python 3.13.11 (arm64)
• pyannote.audio 4.0.4
• PyTorch 2.10.0 (CPU-only)
• pyannote/speaker-diarization-community-1 model (CC-BY-4.0)

Details

• Compressed: ~234 MB
• Extracted: ~743 MB
• Platform: macOS arm64 (Apple Silicon)

Usage

Downloaded automatically by MinuteAI when pyannote diarization is enabled.

Model Attribution

Speaker diarization model by Hervé Bredin et al., licensed under CC-BY-4.0.
Original: https://huggingface.co/pyannote/speaker-diarization-community-1