Fast and optimized open-model inference on multi-region infrastructure with zero request retention.