Luo , C.; Wang, X. Latency–Throughput Tradeoffs of ONNX Runtime, TensorRT-LLM, VLLM, and Triton: An Empirical Comparison on 1B–3B Parameter LLM Inference. SR 2026, 6.