Hypura: Storage-Tier-Aware LLM Inference Scheduler for Apple Silicon Hits 194 Points

GitHub / Hacker News·medium signal

Hypura is a new open-source LLM inference scheduler designed specifically for Apple Silicon that is aware of the storage hierarchy (unified memory, SSD, swap) and schedules model weight loading accordingly. With 194 points and 75 comments, it signals strong practitioner demand for optimized local inference on Mac hardware. Complements the iPhone 17 Pro demo — both show the SSD-streaming approach gaining traction for running large models on Apple hardware.