MII makes low-latency and high-throughput inference possible, powered by DeepSpeed. - View it on GitHub