Run GPU and CPU work in parallel to speed up AI inference

Original: Unlocking asynchronicity in continuous batching

Source: huggingface.co ↗