Speeding up the loading of ML models by efficiently splitting the weights across multiple disks then reading them in parallel during model initialization - View it on GitHub