“The open-source inference engine market has largely concentrated around three main runtimes: VLLM, SGLang, and TensorRT LLM.”