Closed atiorh closed 5 months ago
@jkrukowski has expressed interest in taking this on, here is a suggested approach:
Starting with CLI:
--audio-path
accepts an array of audio files now--batch-size
param which stands for maximum concurrent inference that can happen in the queue (default is unlimited). Open to suggestions on terminology, because batch
isn't exactly the same concept as concurrent async
Adopting Core ML's async batch prediction APIs will be crucial for high-throughput batch transcription with WhisperKit.
Reference: https://developer.apple.com/videos/play/wwdc2023/10049/