Low-Latency Bayesian Inference: Deploying Models with PyTorch and ONNX